# Sesi Context Pack ## Project & Language ### project **architecture**: **execution_engines**: - tree_walking_interpreter - bytecode_compiler_and_vm **primary_components**: - AI runtime and multi-provider model integration - Sesi language lexer, parser, AST, compiler, and VM - built-in runtime library - static semantic and type checking - standard modules for games, audio, drawing, browser automation, APIs, and databases - package manager and CLI execution entry points **description**: A scripting language and runtime with a tree-walking interpreter, bytecode VM, AI model integration, built-in tooling, package management, and standard modules. **entry_points**: - **exports**: - Lexer - Parser - Interpreter - Environment - Compiler - VM - disassemble - runInstall **module**: index - **functions**: - runSesi - runSesiFile **module**: runtime **entrypoints**: **cli**: **functions**: - runSesi - runSesiFile **library**: **exports**: - Lexer - Parser - Interpreter - Compiler - VM - Environment - disassemble - runInstall **language**: TypeScript **name**: Sesi **runtime**: Node.js ### title Sesi Built-in Functions Reference ### description Reference for Sesi built-in functions, standard-library modules, normal invocation, and pipe operator usage. ### language **compiler**: **bytecode**: **capabilities**: - stack_operations - globals - locals - closures - upvalues - control_flow - collections - builtins - AI_calls - imports - memory_initialization - exception_handling **class**: OpCode **class**: Compiler **output**: Chunk **diagnostics**: - lexical_errors - parser_errors - compiler_errors - dry_run_semantic_checks - type_checks - undefined_symbol_detection - unused_symbol_warnings **execution_modes**: - bytecode_vm - tree_walking_interpreter **keywords**: - let - fn - async - await - if - else - while - for - in - to - return - break - continue - try - catch - finally - prompt - model - image - convert - memory - import - export - allow - make **name**: Sesi **syntax_features**: - let_declarations - functions - async_functions - closures_and_upvalues - make_object_templates - arrays - objects - loops - conditions - try_catch_finally - imports - exports - memory_bindings - prompt_expressions - AI_model_expressions - image_generation_expressions - video_generation_expressions - conversion_expressions - structured_output - tool_calls - await_expressions - type_annotations - variables - default_parameters - member_access - index_access - assignments - if_else - while_loops - for_in_loops - range_loops - allow_namespace_imports - prompts - model_calls - image_calls - video_calls - await - pipeline_operator **type_annotations**: **compound**: - array - object - T? - union_types **primitives**: - number - string - bool - any ### syntax **normal_call**: function_name(argument1, argument2) **note**: Piping passes the value on the left as the first function argument. Normal calls are generally recommended for readability. **pipe_call**: value | function_name(other_arguments) ### sesi_language **compiler**: **features**: - global_variables - local_variables - closures - upvalues - control_flow_jumps - exception_handlers - native_ai_opcodes - built_in_fast_path **target**: stack_based_bytecode **lexical_features**: **expressions**: - binary_operations - unary_operations - logical_operations - assignments - calls - member_access - index_access - prompt - model - image - video - structured_output - tool_call - convert - await **literals**: - number - string - boolean - array - object **statements**: - let - fn - async_fn - make - if_else - while - for_in - for_range - return - break - continue - try_catch_finally - import - allow - export - memory **parser**: **class_syntax**: **implementation**: parser_desugaring_to_factory_function **keyword**: make **strategy**: recursive_descent **supported_comments**: - line_comments - block_comments **type_annotations**: **composites**: - array - object - optional_types - union_types **primitives**: - number - string - bool - any **vm**: **execution_model**: stack_based **supports**: - closures - open_and_closed_upvalues - function_frames - native_builtin_calls - async_builtin_calls - try_catch_finally - model_calls - image_calls - video_calls - memory_bindings - timeout_enforcement ### language_engine **bytecode**: **opcodes**: - CONSTANT - DEFINE_GLOBAL - GET_GLOBAL - SET_GLOBAL - GET_LOCAL - SET_LOCAL - ADD - SUBTRACT - MULTIPLY - DIVIDE - MODULO - JUMP - JUMP_IF_FALSE - LOOP - BUILD_ARRAY - BUILD_OBJECT - CLOSURE - CALL - CALL_BUILTIN - CALL_MODEL - CALL_IMAGE - CALL_VIDEO - PRINT - TRY_START - IMPORT - ALLOW - INITIALIZE_MEMORY - CONVERT **source**: chunk.ts **compiler**: **optimizations**: - local_slot_variables - closure_upvalues - builtin_call_fast_path - dedicated_ai_opcodes **source**: compiler.ts **target**: bytecode **execution_modes**: **default**: bytecode_vm **diagnostics**: - token_listing - AST_output - bytecode_disassembly - dry_run_semantic_checks - type_checks - profiling **fallback**: tree_walking_interpreter **lexer**: **source**: lexer.ts **supports**: - identifiers - strings - numbers - comments - arrays - objects - arithmetic - logical_operators - type_annotations - async_await - imports - prompts - model_calls **parser**: **source**: parser.ts **style**: recursive_descent **supports**: **expressions**: - literals - arrays - objects - function_calls - member_access - indexing - assignments - pipe_operator - prompts - model - image - video - convert - structured_output - tool_call **statements**: - let - fn - async_fn - make - if_else - while - for - return - break - continue - try_catch_finally - import - allow - export - memory ### language_features **syntax**: **ai_primitives**: - model - image - video - structured_output - tool_call **control_flow**: - if_else - while - for_in - range_for - break - continue - try_catch_finally - return **declarations**: - let - const - fn - async - make - memory - prompt **expressions**: - literals - arrays - objects - indexing - member_access - assignments - arithmetic - comparisons - logical_short_circuiting - conditional_expressions - await **modules**: - import - allow - export - std_modules ### language_runtime **bytecode**: **opcodes**: **ai**: - CALL_MODEL - CALL_IMAGE - CALL_VIDEO **arithmetic**: - ADD - SUBTRACT - MULTIPLY - DIVIDE - MODULO - NEGATE **collections**: - BUILD_ARRAY - BUILD_OBJECT - GET_INDEX - SET_INDEX - GET_PROPERTY - SET_PROPERTY **comparison**: - EQUAL - NOT_EQUAL - LESS - LESS_EQUAL - GREATER - GREATER_EQUAL **constants**: - CONSTANT - NIL - true - false **control_flow**: - JUMP - JUMP_IF_FALSE - LOOP - RETURN - RETURN_VOID **conversion**: - CONVERT **exceptions**: - TRY_START - TRY_END - FINALLY_START - FINALLY_END **functions**: - CLOSURE - CALL - CALL_BUILTIN - CLOSE_UPVALUE **memory**: - INITIALIZE_MEMORY **modules**: - IMPORT - ALLOW **variables**: - DEFINE_GLOBAL - GET_GLOBAL - SET_GLOBAL - GET_LOCAL - SET_LOCAL - GET_UPVALUE - SET_UPVALUE **compiler**: **class**: Compiler **optimization**: **builtin_fast_path**: CALL_BUILTIN **supports**: - globals - local_slots - closures - upvalues - default_parameters - loops - break_continue - try_catch_finally - imports - memory - AI_call_opcodes **execution_engines**: **bytecode_vm**: **default**: true **modules**: - chunk - compiler - vm **tree_walking_interpreter**: **available**: true **enabled_with**: treeWalker **language**: Sesi **lexer**: **class**: Lexer **supports**: - identifiers - numbers - strings - escaped_strings - comments - arrays - objects - functions - async_await - control_flow - imports - AI expressions - type_annotations **parser**: **class**: Parser **language_features**: **ai_syntax**: - prompt - model - image - video - structured_output - tool_call - convert **control_flow**: - if_else - while - for_in - for_range - try_catch_finally - return - break - continue **declarations**: - let - fn - async_fn - make - memory - import - allow - export **expressions**: - arithmetic - comparisons - logical_operations - assignments - conditional_expressions - arrays - objects - indexing - member_access - function_calls - pipes - await **style**: recursive_descent ### language_toolchain **bytecode**: **core_opcodes**: - CONSTANT - NIL - true - false - POP - DEFINE_GLOBAL - GET_GLOBAL - SET_GLOBAL - GET_LOCAL - SET_LOCAL - ADD - SUBTRACT - MULTIPLY - DIVIDE - MODULO - NEGATE - EQUAL - NOT_EQUAL - LESS - LESS_EQUAL - GREATER - GREATER_EQUAL - NOT - JUMP - JUMP_IF_FALSE - LOOP - BUILD_ARRAY - BUILD_OBJECT - GET_INDEX - SET_INDEX - GET_PROPERTY - SET_PROPERTY - CLOSURE - CALL - RETURN - CALL_BUILTIN - CALL_MODEL - CALL_IMAGE - CALL_VIDEO - PRINT - TRY_START - TRY_END - FINALLY_START - FINALLY_END - IMPORT - ALLOW - INITIALIZE_MEMORY - GET_UPVALUE - SET_UPVALUE - CLOSE_UPVALUE - CONVERT **compiler**: **supports**: - global_variables - local_variables - closures - upvalues - default_function_parameters - control_flow - loops - break_continue - try_catch_finally - builtin_fast_path - model_image_video_opcodes - imports - memory_initialization - type_conversion **target**: bytecode_vm **execution_modes**: **default**: bytecode_vm **diagnostics**: - token_dump - ast_dump - bytecode_dump - dry_run_semantic_checks - static_type_checks - profiling - execution_timeout **fallback**: tree_walking_interpreter **lexer**: **supports**: - identifiers - numbers - strings - arrays - objects - comments - operators - type_annotations - async_await - imports_exports - model_image_video_calls - prompt_blocks - memory_blocks **parser**: **architecture**: recursive_descent **supports**: **expressions**: - literals - arrays - objects - member_access - index_access - assignments - unary_operations - binary_operations - logical_operations - function_calls - await - model - image - video - convert - structured_output - tool_call - prompts **statements**: - let - fn - async_fn - make - if_else - while - for - return - break - continue - try_catch_finally - import - allow - export - memory **syntactic_sugar**: **make**: **description**: Desugars class-like make declarations into factory functions with self-bound fields, methods, and optional start constructors. --- ## AI Integration ### AIRuntime **cache**: **cache_key_includes**: - model - resolved local model - local dtype - local device - local system prompt - prompt - temperature - maxTokens - topK - topP - ratio - size - images - systemPrompt - thinkingLevel - search - tools **cached_response_behavior**: **cached**: true **usage**: **inputTokens**: 0 **outputTokens**: 0 **thinkingTokens**: 0 **disabled_when**: request.cache is false **enabled_by_default**: true **file**: .sesi_cache.json **key_algorithm**: SHA-256 **location**: Current working directory **embeddings**: **cache_key**: SHA-256 of input text **fallback_models**: - gemini-embedding-001 - gemini-embedding-2 **method**: embedText **provider**: Gemini **image_generation**: **gemini**: **condition**: request.model contains image **output**: Base64 encoded image **response_modality**: IMAGE **openai**: **handler**: callGPTImageModel **ratio_mapping**: **16:9**: 1536x1024 **1:1**: 1024x1024 **3:4**: 1024x1536 **4:3**: 1536x1024 **9:16**: 1024x1536 **supported_sizes**: - 1024x1024 - 1536x1024 - 1024x1536 **memory**: **operations**: **appendToMemory**: **behavior**: Adds a content entry to a memory history **autoTrimMemory**: **behavior**: Trims only when automatic summarization is enabled **getMemory**: **behavior**: Returns newline-joined history **initializeMemory**: **behavior**: Initializes a memory ID with one entry **searchMemory**: **behavior**: Embeds the query and each memory entry, calculates cosine similarity, and returns the highest scoring entries. **trimMemory**: **behavior**: Forces memory compression under a supplied token limit **updateMemory**: **behavior**: Replaces history with one entry **summarization**: **defaults**: **enabled**: true **maxTokens**: 900000 **summaryModel**: gemini-3.5-flash-lite **targetTokens**: 60% of maxTokens **environment_variables**: **SESI_MEMORY_AUTO_SUMMARIZE**: **disable_value**: false **SESI_MEMORY_MAX_TOKENS**: Maximum memory token budget **SESI_MEMORY_SUMMARY_MODEL**: Model used for summarization **SESI_MEMORY_TARGET_TOKENS**: Target compressed memory token budget **strategy**: - Preserves recent memory using approximately 55% of target token budget - Summarizes older memory with an AI model - Stores summary under "[Memory Summary]" - Stores retained recent content under "[Recent Memory]" - Applies per-memory locks to prevent concurrent trim races **summary_requirements**: - facts - decisions - constraints - user preferences - unresolved tasks - names - dates - identifiers **model_routing**: **gemini**: **condition**: Default provider **handler**: Gemini generateContent APIs **local**: **condition**: model equals local or starts with local: **handler**: callLocalModel **openai_image**: **condition**: model starts with gpt- and contains image **handler**: callGPTImageModel **openai_text**: **condition**: model starts with gpt- **handler**: callGPTModel **providers**: **gemini**: **authentication**: **environment_variable**: GEMINI_API_KEY **continuation**: **behavior**: When generation ends with MAX_TOKENS, appends partial output to the conversation and asks the model to continue. **maximum_polls**: 10 **model_normalization**: **behavior**: Adds the models/ prefix unless the model is already a models/, projects/, publishers/, or slash-qualified identifier. **model_selector**: Gemini model names and Google model paths **sdk**: @google/genai **supports**: - text generation - streaming - image input - audio input - image generation - video generation - text-to-speech - speech transcription - function tools - Google Search - native token counting - embeddings **local**: **caching**: **fallback**: If a local cached model fails to load, retries while allowing remote download. **validates**: - config.json - tokenizer.json - ONNX model file - ONNX external data file for small ONNX headers **configuration**: **SESI_LOCAL_CACHE_DIR**: **default**: ~/.cache/sesi/models **purpose**: Model cache directory **SESI_LOCAL_DEVICE**: **default**: cpu **purpose**: Inference device **SESI_LOCAL_DTYPE**: **default**: q4 **purpose**: ONNX model datatype **SESI_LOCAL_MODEL**: **purpose**: Overrides the default local model **SESI_LOCAL_SYSTEM_PROMPT**: **purpose**: Default system prompt for local models **SESI_LOCAL_WARN_TOKENS**: **default**: 2048 **purpose**: Warning threshold for large local prompts **default_model**: onnx-community/Qwen2.5-0.5B-Instruct **model_selector**: **accepted**: - local - local: **runtime**: @huggingface/transformers **supported**: - text generation - streaming - function tools - native tokenizer-based token counting **unsupported**: - images - audio - web search **openai**: **APIs**: **image**: **condition**: GPT model name contains 'image' **endpoint**: images.generate **supports**: - prompt-based image generation - ratio-to-size mapping **unsupported**: - audio input - reference images - tools - web search - streaming **text**: **endpoint**: responses.create **supports**: - text generation - streaming - images as input - function tools - web search - reasoning effort - native input token counting **authentication**: **environment_variable**: OPENAI_API_KEY **model_selector**: Models whose names begin with gpt- **reasoning_effort_mapping**: **high**: high **low**: low **medium**: medium **minimal**: minimal **no**: minimal **sdk**: openai **request_features**: **audio**: **supported_by**: - Gemini text - Gemini transcription **unsupported_by**: - local - OpenAI GPT text - OpenAI GPT image **context_injection**: **enabled_for**: - OpenAI text - Gemini text **format**: [System context: Current date and time is ] **images**: **encoding**: Base64 **source**: Local filesystem paths **supported_by**: - OpenAI text - Gemini text - Gemini image generation **search**: **gemini**: **tool**: googleSearch **openai**: **tool**: web_search **streaming**: **behavior**: **callback**: Calls request.stream when it is a function **stdout**: Writes deltas to stdout when request.stream is true **supported_by**: - local - OpenAI text - Gemini text **tool_calls**: **gemini**: **response_format**: JSON string representing functionCall or call **local**: **accepted_formats**: - structured tool_calls - JSON - [TOOL_CALLS] JSON - fenced JSON **response_format**: JSON string with TOOL_CALL finish reason **openai**: **response_format**: JSON string containing name, args, and optional call_id **speech**: **synthesis**: **default_model**: gemini-2.5-flash-preview-tts **default_voice**: Vindemiatrix **method**: synthesizeSpeech **output**: Base64 encoded WAV **processing**: - Receives PCM audio from Gemini - Extracts sample rate from MIME type - Constructs a 44-byte WAV header **provider**: Gemini **transcription**: **default_model**: gemini-3.5-flash-lite **method**: transcribeSpeech **optional_language_instruction**: true **output**: Plain transcript text **provider**: Gemini **state**: **_client**: **purpose**: Lazy-initialized Google Gemini client **_openAIClient**: **purpose**: Lazy-initialized OpenAI client **conversationHistory**: **purpose**: Stores memory entries by memory ID **type**: Map **embeddingCache**: **purpose**: Caches embeddings by SHA-256 text hash **type**: Map **localPipelines**: **purpose**: Caches pending and initialized local text-generation pipelines **type**: Map> **localTokenizers**: **purpose**: Caches pending and initialized local tokenizers **type**: Map> **memorySummaryConfigs**: **type**: Map **memoryTrimLocks**: **purpose**: Prevents concurrent trimming of the same memory **type**: Map> **structured_output**: **behavior**: - Attempts to extract a JSON object from model output - Filters parsed values to keys declared in the supplied schema - If no JSON is found, asks Gemini to convert the response to JSON - Returns an empty object if parsing fails **method**: parseStructuredOutput **token_counting**: **gemini**: **endpoint**: models.countTokens **heuristic**: **formula**: ceil(text.length / 4) **local**: **fallback**: Character-based estimate **method**: Local tokenizer encode method **openai**: **endpoint**: responses.inputTokens.count **video_generation**: **handler**: callVideo **modes**: **omni**: **condition**: model name contains omni **endpoint**: interactions.create **supports**: - prompt - multiple reference images - aspect ratio - task configuration **standard**: **endpoint**: models.generateVideos **polling**: **default_interval_ms**: 10000 **minimum_interval_ms**: 100 **supports**: - prompt - first reference image - aspect ratio - duration - resolution - negative prompt - audio generation **output**: Base64 encoded MP4 video **provider**: Gemini ### ai_features **caching**: **file**: .sesi_cache.json **hash_algorithm**: sha256 **includes**: - model - prompt - generation_parameters - image_inputs - system_prompt - tools - local_model_configuration **memory**: **defaults**: **max_tokens**: 900000 **summary_model**: gemini-3.5-flash-lite **target_tokens_ratio**: 0.6 **features**: - conversation_history - embedding_search - automatic_summarization - token_budget_trimming - per_memory_configuration **tool_orchestration**: **max_tool_calls_default**: 8 **safety_restrictions**: - exec - run - spawn - python - js - ffmpeg - gif - video ### ai_runtime **cache**: **algorithm**: sha256 **cache_key_includes**: - model - local_model_configuration - prompt - sampling_configuration - image_paths - system_prompt - thinking_level - search - tools **file**: .sesi_cache.json **caching**: **algorithm**: sha256 **behavior**: **cache_hit**: **streams_cached_text**: true **usage**: **input_tokens**: 0 **output_tokens**: 0 **thinking_tokens**: 0 **cached_usage**: **inputTokens**: 0 **outputTokens**: 0 **thinkingTokens**: 0 **cache_key_includes**: - model - local model configuration - prompt - generation parameters - images - system prompt - thinking level - search setting - tools - generation_parameters - system_prompt - thinking_level - local_model_configuration **cache_key_inputs**: - model - prompt - local_model_configuration - temperature - max_tokens - sampling_parameters - image_inputs - system_prompt - thinking_level - search - tools **file**: .sesi_cache.json **hash_algorithm**: sha256 **includes**: - model - local_model_settings - prompt - sampling_settings - images - system_prompt - thinking_level - search - tools - resolved_local_model - local_dtype - local_device - generation_parameters - image_paths - tool_definitions - local_model_configuration **class**: AIRuntime **common_features**: - request caching in .sesi_cache.json - SHA-256 cache keys - current UTC time context injection - token usage reporting - streaming callbacks or stdout streaming - structured-output parsing - prototype stripping for untrusted objects - tool-call extraction - image path resolution - model-specific request normalization **default_local_model**: onnx-community/Qwen2.5-0.5B-Instruct **default_local_warning_tokens**: 2048 **embeddings**: **fallback_models**: - gemini-embedding-001 - gemini-embedding-2 **similarity**: cosine **instance**: aiRuntime **local_cache**: **file**: .sesi_cache.json **includes**: - model - prompt - sampling_configuration - image_paths - system_prompt - tool_configuration - local_model_configuration **strategy**: SHA-256 **local_model_handling**: **allowed_model_id_format**: organization/model **generation_defaults**: **max_new_tokens**: 512 **sampling_enabled_when_temperature_positive**: true **model_id_validation**: **rejects**: - . - .. - IDs without exactly one slash **segment_pattern**: ^[A-Za-z0-9._-]+$ **tool_call_formats_supported**: - native_tool_calls - {...} - [TOOL_CALLS] {...} - fenced_json - plain_json **media**: **image_generation**: **gemini**: **response_modality**: IMAGE **supports_reference_images**: true **openai**: **ratio_mapping**: **16:9**: 1536x1024 **1:1**: 1024x1024 **3:4**: 1024x1536 **4:3**: 1536x1024 **9:16**: 1024x1536 **supported_sizes**: - 1024x1024 - 1536x1024 - 1024x1536 **speech_synthesis**: **default_model**: gemini-2.5-flash-preview-tts **default_voice**: Vindemiatrix **output**: base64_encoded_wav **provider**: gemini **transcription**: **default_model**: gemini-3.5-flash-lite **output**: plain_transcript **provider**: gemini **video_generation**: **modes**: - omni_interactions - generateVideos_operation_polling **outputs**: base64_encoded_video **provider**: gemini **memory**: **automatic_summarization**: true **capabilities**: - initialize - append - retrieve - update - semantic_search - automatic_summarization - token_budget_trimming - initialize memory - append memory - retrieve memory - update memory - semantic memory search - automatic summarization - token-budget trimming - concurrent trim locking **conversation_history**: in_memory_map **default_max_tokens**: 900000 **default_summary_configuration**: **enabled**: true **max_tokens**: 900000 **model**: gemini-3.5-flash-lite **summary_model**: gemini-3.5-flash-lite **target_tokens_ratio**: 0.6 **default_summary_model**: gemini-3.5-flash-lite **default_target_ratio**: 0.6 **defaults**: **auto_summarize**: true **max_tokens**: 900000 **summary_model**: gemini-3.5-flash-lite **target_tokens_ratio**: 0.6 **embedding_cache**: in_memory_map **embedding_models**: - gemini-embedding-001 - gemini-embedding-2 **embeddings**: **models**: - gemini-embedding-001 - gemini-embedding-2 **similarity**: cosine_similarity **environment_variables**: **SESI_MEMORY_AUTO_SUMMARIZE**: Set to false to disable automatic summarization. **SESI_MEMORY_MAX_TOKENS**: Maximum estimated token count. **SESI_MEMORY_SUMMARY_MODEL**: Model used for memory summaries. **SESI_MEMORY_TARGET_TOKENS**: Target token count after compaction. **features**: - initialize - append - update - retrieve - semantic_search - automatic_summarization - token_budget_trimming - per_memory_configuration - concurrent_trim_locking **functions**: - initializeMemory - appendToMemory - getMemory - updateMemory - autoTrimMemory - trimMemory - searchMemory - configureMemorySummary **operations**: - initialize - append - get - update - semantic_search - automatic_trim - manual_trim - configuration **semantic_search**: **embedding_cache**: sha256_keyed **embedding_models**: - gemini-embedding-001 - gemini-embedding-2 **similarity**: cosine_similarity **storage**: in_memory_conversation_history **summarization**: **default_max_tokens**: 900000 **default_model**: gemini-3.5-flash-lite **default_target_fraction**: 0.6 **environment_variables**: - SESI_MEMORY_AUTO_SUMMARIZE - SESI_MEMORY_MAX_TOKENS - SESI_MEMORY_TARGET_TOKENS - SESI_MEMORY_SUMMARY_MODEL **summary_format**: - [Memory Summary] - [Recent Memory] **multimedia**: **image_generation**: **providers**: - OpenAI GPT image models - Gemini image models **speech**: **synthesis**: **default_voice**: Vindemiatrix **output**: Base64-encoded WAV **provider**: Gemini **transcription**: **default_model**: gemini-3.5-flash-lite **provider**: Gemini **video_generation**: **features**: - reference images - aspect ratio - duration - resolution - negative prompt - audio generation - polling interval **provider**: Gemini **supported_modes**: - Omni interaction video generation - asynchronous generateVideos polling **persistent_response_cache**: **algorithm**: SHA-256 **cache_key_includes**: - model - local_model_configuration - prompt - generation_settings - images - system_prompt - reasoning_settings - search - tools **file**: .sesi_cache.json **providers**: **gemini**: **APIs**: - models.generateContent - models.generateContentStream - models.countTokens - models.embedContent - models.generateVideos - interactions.create **authentication**: GEMINI_API_KEY **capabilities**: - text_generation - image_input - audio_input - image_generation - video_generation - speech_synthesis - speech_transcription - embeddings - token_counting - search - tools - streaming - text generation - streaming output - image inputs - audio inputs - image generation - video generation - speech synthesis - speech transcription - native token counting - Google Search tools - function declarations - thinking configuration - google_search_grounding - function_tools - text_to_speech - speech_to_text - function_calling - google_search - web_search - tool_calls - native_token_counting - audio_transcription - memory_summarization **credential**: GEMINI_API_KEY **default_provider**: true **default_transcription_model**: gemini-3.5-flash-lite **default_tts_model**: gemini-2.5-flash-preview-tts **embedding_models**: - gemini-embedding-001 - gemini-embedding-2 **features**: - text_generation - image_input - audio_transcription - image_generation - video_generation - text_to_speech - streaming - function_tools - google_search - embeddings - token_counting - reasoning_configuration **model_detection**: gemini-* **model_name_normalization**: **default_prefix**: models/ **prefix**: models/ **prefixes_preserved**: - models/ - projects/ - publishers/ **preserves_fully_qualified_prefixes**: - models/ - projects/ - publishers/ **model_pattern**: ^(models/)?gemini- **model_prefixes**: - gemini- - models/gemini- **package**: @google/genai **required_environment**: - GEMINI_API_KEY **sdk**: @google/genai **sdk_package**: @google/genai **supported_features**: - text_generation - streaming - image_input - audio_input - image_generation - video_generation - embeddings - text_to_speech - transcription - native_token_counting - function_tools - google_search **trigger_models**: - gemini-* - models/gemini-* **local**: **cache_directory**: ~/.cache/sesi/models **cache_validation**: **onnx_directory**: onnx **required_files**: - config.json - tokenizer.json **supports_split_onnx_data_files**: true **caching**: **cache_directory_default**: ~/.cache/sesi/models **checks**: - config.json - tokenizer.json - ONNX model file - ONNX data file for small model headers **fallback_behavior**: If an apparently cached model cannot load, retry with remote downloads enabled. **local_model_validation**: **checks_external_onnx_data**: true **required_files**: - config.json - tokenizer.json - onnx/model_.onnx **pipeline_cache_key**: - model - dtype - device - cache_directory **tokenizer_cache_key**: - model - cache_directory **capabilities**: - text_generation - token_counting - streaming - tool_calls - tokenizer_based_token_counting - tool_call_parsing - local_model_caching - local_token_counting - tool_call_extraction - local_model_download_and_cache **configuration**: **cache_directory_default**: ~/.cache/sesi/models **default_warning_tokens**: 2048 **environment_variables**: **SESI_LOCAL_CACHE_DIR**: Local model cache directory **SESI_LOCAL_DEVICE**: Execution device; default cpu **SESI_LOCAL_DTYPE**: Model quantization/type; default q4 **SESI_LOCAL_MODEL**: Override default local Hugging Face model ID **SESI_LOCAL_SYSTEM_PROMPT**: Default local model system prompt **SESI_LOCAL_WARN_TOKENS**: CPU prompt-size warning threshold **default_device**: cpu **default_dtype**: q4 **default_model**: onnx-community/Qwen2.5-0.5B-Instruct **defaults**: **cache_directory**: ~/.cache/sesi/models **device**: cpu **device_env**: SESI_LOCAL_DEVICE **dtype**: q4 **dtype_env**: SESI_LOCAL_DTYPE **engine**: @huggingface/transformers **environment_variables**: **SESI_LOCAL_CACHE_DIR**: Model-cache directory. **SESI_LOCAL_DEVICE**: Inference device; defaults to cpu. **SESI_LOCAL_DTYPE**: Local ONNX model data type; defaults to q4. **SESI_LOCAL_MODEL**: Overrides the default local Hugging Face model. **SESI_LOCAL_SYSTEM_PROMPT**: Default local-model system prompt. **SESI_LOCAL_WARN_TOKENS**: Input warning threshold; 0 disables warnings. **features**: - text_generation - streaming - tool_call_normalization - token_counting - local_model_cache - Hugging Face Transformers dynamic import - Node.js remote-model support - local model cache detection - tokenizer caching - text generation - tool-call normalization and parsing - input token warnings **implementation**: @huggingface/transformers **library**: @huggingface/transformers **limitations**: - text_only_input - no_audio_input - no_search - no_image_input - no_web_search **model_prefixes**: - local - local: - local: - local: **model_selector**: **aliases**: - local - local: **default_model**: onnx-community/Qwen2.5-0.5B-Instruct **package**: @huggingface/transformers **restrictions**: - images_not_supported - audio_not_supported - search_not_supported - text_only - no_audio_input - no_search **selector**: **accepted_models**: - local - local: **supported_features**: - text_generation - streaming - tool_calls - local_token_counting **task**: text-generation **transformer_package**: @huggingface/transformers **transformers_package**: @huggingface/transformers **trigger_models**: - local - local: - local:/ **unsupported_features**: - image_input - audio_input - web_search **unsupported_inputs**: - images - audio - search - web search **openai**: **APIs**: - responses.create - responses.inputTokens.count - images.generate - Responses API - Images API - Input token counting API **api**: - responses.create - responses.inputTokens.count - images.generate **authentication**: OPENAI_API_KEY **capabilities**: - text_generation - image_input - image_generation - tool_calls - web_search - streaming - native_token_counting - text generation - streaming output - image inputs through base64 data URLs - image generation - function tools - web search - reasoning-effort mapping - function_tools - reasoning_effort - responses_api - token_counting - reasoning_effort_mapping **credential**: OPENAI_API_KEY **features**: - text_generation - image_input - image_generation - streaming - function_tools - web_search - reasoning_effort - token_counting **image_model_detection**: GPT model name contains 'image'. **image_models**: **detected_when**: GPT model name contains 'image' **supported_sizes**: - 1024x1024 - 1536x1024 - 1024x1536 **image_sizes**: - 1024x1024 - 1536x1024 - 1024x1536 **model_detection**: ^gpt- **model_pattern**: ^gpt- **model_prefix**: gpt- **package**: openai **reasoning_effort_levels**: - minimal - low - medium - high **reasoning_efforts**: - minimal - low - medium - high **required_environment**: - OPENAI_API_KEY **sdk**: openai **sdk_package**: openai **selector**: ^gpt- **supported_features**: - responses_api - streaming - image_input - tools - web_search - token_counting - image_generation **supported_image_sizes**: - 1024x1024 - 1536x1024 - 1024x1536 **trigger_models**: - gpt-* **request_features**: - model - prompt - system_prompt - temperature - max_tokens - top_k - top_p - ratio - image_size - images - audio - tools - search - thinking_level - streaming - caching - cache - size - image_ratio - stream - system_prompts - reasoning_or_thinking_level - tool_schemas - web_search **response_features**: - text - finish_reason - token_usage - thinking_token_usage - cached **response_fields**: - text - finishReason - usage.inputTokens - usage.outputTokens - usage.thinkingTokens - cached **safety**: **local_model_id_validation**: true **local_tool_call_normalization**: true **prototype_stripping**: true **tool_argument_json_parsing**: true **safety_features**: - prototype_stripping_for_external_data - local_model_id_validation - cached_model_fallback_to_remote_download - tool_call_normalization - automated_tool_limit **singleton**: aiRuntime **singleton_export**: aiRuntime **speech**: **synthesis**: **default_model**: gemini-2.5-flash-preview-tts **default_voice**: Vindemiatrix **output**: base64_encoded_wav **provider**: Gemini **transcription**: **default_model**: gemini-3.5-flash-lite **output**: transcript_text **provider**: Gemini **supported_providers**: **gemini**: **authentication**: GEMINI_API_KEY **capabilities**: - text_generation - image_input - audio_transcription - image_generation - video_generation - speech_synthesis - embeddings - native_token_counting - Google_Search_grounding - function_tools - streaming **model_prefixes**: - gemini- - models/gemini- **sdk**: @google/genai **local**: **cache_directory**: ~/.cache/sesi/models **capabilities**: - text_generation - local_token_counting - streaming - tool_call_extraction **default_device**: cpu **default_dtype**: q4 **default_model**: onnx-community/Qwen2.5-0.5B-Instruct **environment_variables**: - SESI_LOCAL_MODEL - SESI_LOCAL_DTYPE - SESI_LOCAL_DEVICE - SESI_LOCAL_CACHE_DIR - SESI_LOCAL_SYSTEM_PROMPT - SESI_LOCAL_WARN_TOKENS **invocation**: dynamic_import **limitations**: - text_only_input - no_audio_input - no_search **package**: @huggingface/transformers **openai**: **APIs**: - responses.create - responses.inputTokens.count - images.generate **authentication**: OPENAI_API_KEY **capabilities**: - text_generation - image_input - streaming - function_tools - web_search - image_generation - token_counting **model_detection**: ^gpt- **sdk**: openai **utilities**: - strip_prototypes - cosine_similarity - token_estimation - native_token_counting - embedding_cache - image_to_base64_conversion - audio_pcm_to_wav_conversion **video_generation**: **output**: base64_encoded_video **supports**: - Gemini video generation operations - Omni interaction-based video generation - reference images - aspect ratios - durations - resolutions - negative prompts - generated audio - operation polling ### ai_usage - **description**: Estimates paid-tier text-token costs in USD. **name**: estimate_cost **pipe_supported**: true **result_fields**: - input_tokens - output_tokens - input_cost_usd - output_cost_usd - total_cost_usd - input_per_million - output_per_million - pricing_source - pricing_snapshot_date **signature**: estimate_cost(model, input, output = 0, rates = null) -> object|null - **description**: Returns usage and estimated cost for the latest model() call. **name**: model_usage **notable_fields**: - total_tokens - thinking_tokens - billable_output_tokens - total_cost_usd - cached **signature**: model_usage() -> object|null --- ## Builtins & Modules ### builtins **AI**: - speech - from_speech - translate - count_tokens - estimate_tokens - estimate_cost - model_usage - memory_search - memory_trim - memory_config - workflow - set_alias - define_tool - list_tools **ai**: - speech - from_speech - translate - count_tokens - estimate_tokens - estimate_cost - model_usage - memory_search - memory_trim - memory_config - set_alias - workflow - define_tool - list_tools **archive_support**: **external_archives**: - 7z - rar - tar - tar.gz - tgz - tar.bz2 - tbz2 - tar.xz - txz **protections**: - validates all archive entry paths - rejects absolute archive paths - rejects traversal entries - disables non-ZIP archives in safe mode **zip_containers**: - zip - jar - apk - docx - xlsx - pptx - epub **categories**: **AI**: - count_tokens - estimate_tokens - estimate_cost - model_usage - memory_search - memory_trim - memory_config - define_tool - list_tools - set_alias - workflow **AI_and_memory**: - workflow - memory_search - memory_trim - memory_config - set_alias - define_tool - list_tools - retry - lazy - force - timeout **ai**: - speech - from_speech - translate - count_tokens - estimate_tokens - estimate_cost - model_usage - memory_search - memory_trim - memory_config - workflow - define_tool - list_tools - tokenize - set_alias **ai_and_memory**: - workflow - memory_search - memory_trim - memory_config - set_alias - define_tool - list_tools **basic**: - show - debug - input - len - type - str - num - float - bool - time - random - env **collections**: - range - push - append - pop - keys - values - join - split - map - filter - reduce - find - reverse - sort - unique - flatten **control_flow**: - retry - lazy - force - timeout - multi_req **core**: - show - debug - input - len - type - str - num - float - bool - range - env - time - random **cryptography**: - encrypt - decrypt **data**: - len - type - str - num - float - bool - to_json - from_json - keys - values - range - push - append - pop - join - split - slice - reverse - sort - unique - flatten **errors**: - error_type - raise_error **external_execution**: - exec - run - spawn - python - js - sesi **files**: - read_file - write_file - append_file - write_image - list_dir - get_ext - exists - make_dir - rename - archive - trash - zip **filesystem**: - read_file - write_file - append_file - write_image - open_file - open - list_dir - get_ext - exists - make_dir - rename - archive - trash - zip **functional**: - map - filter - reduce - find - multi_req - lazy - force - timeout - retry - workflow **general**: - show - debug - input - len - type - str - num - float - bool - time - random - env - range **json_and_crypto**: - to_json - from_json - encrypt - decrypt **matrices**: - matrix_dot - matrix_transpose - matrix_add - matrix_sub - matrix_mul_elements - matrix_scale - matrix_sigmoid - matrix_dsigmoid - matrix_sum_rows - matrix_mse **matrix**: - matrix_dot - matrix_transpose - matrix_add - matrix_sub - matrix_mul_elements - matrix_scale - matrix_sigmoid - matrix_dsigmoid - matrix_sum_rows - matrix_mse **media**: - speech - from_speech - translate - ffmpeg - gif - video **network**: - web_get - web_send - listen - api - live **networking**: - web_get - web_send - listen - api - live **process**: - exec - run - spawn - python - js - sesi **processes**: - spawn - exec - run - python - js - sesi **profiling**: - profile - profile_start - profile_end - profile_report **runtime_control**: - retry - lazy - force - timeout - profile - profile_start - profile_end - profile_report - multi_req - live **serialization_and_crypto**: - to_json - from_json - encrypt - decrypt **strings**: - to_upper - to_lower - trim - slice - swap - contains - locate - starts_with - ends_with - index_of - repeat - includes - regex - tokenize - split - join **strings_and_collections**: - range - push - append - pop - keys - values - join - split - regex - to_upper - to_lower - trim - slice - swap - contains - locate - starts_with - ends_with - index_of - repeat - includes - reverse - sort - unique - flatten **system**: - open - open_file - spawn - exec - run - python - js - ffmpeg - gif - video **tokenization**: **GPT_5_compatibility_encoding**: o200k_base **default_model**: gpt-5.6-sol **library**: js-tiktoken **tokenization_and_costs**: - tokenize - count_tokens - estimate_tokens - estimate_cost - model_usage **tokens_and_cost**: - tokenize - count_tokens - estimate_tokens - estimate_cost - model_usage **type_checks**: - is_function - is_array - is_object - is_string - is_number - is_bool - is_null - name - arity **web**: - web_get - web_send - listen - api - open - open_file - live **collections**: - range - push - append - pop - keys - values - join - split - map - filter - reduce - find - reverse - sort - unique - flatten **conversion**: - num - float - bool - trunc - to_upper - to_lower - trim **core**: - show - debug - input - len - type - str - num - float - bool - range - push - append - pop - keys - values - join - split - map - filter - reduce - find - sort - unique - flatten **crypto**: - encrypt - decrypt **cryptography**: - encrypt - decrypt - **algorithm**: aes-256-cbc - **key_derivation**: sha256_password_hash **data_and_collections**: - len - type - str - num - float - bool - range - push - append - pop - keys - values - join - split - slice - reverse - sort - unique - flatten - map - filter - reduce - find **errors**: - error_type - raise_error **external_processes**: - exec - run - spawn - python - js **file_system_security**: **behavior**: - Resolves paths against a base directory. - Restricts access to configured allowed paths in safe mode. - Includes script directory in permitted paths. - Rejects path traversal outside permitted directories. **function**: ensureSafePath **files**: - read_file - write_file - append_file - write_image - list_dir - exists - get_ext - make_dir - rename - archive - trash - zip **filesystem**: - read_file - write_file - append_file - write_image - list_dir - make_dir - rename - archive - trash - zip - exists - get_ext **functional**: - lazy - force - timeout - retry - multi_req - map - filter - reduce - find - workflow - profile - profile_start - profile_end - profile_report **general**: - show - debug - input - len - type - str - to_json - from_json - env - time - random - num - float - bool **introspection**: - name - arity - is_function - is_array - is_object - is_string - is_number - is_bool - is_null **json_and_crypto**: - to_json - from_json - encrypt - decrypt **math_matrices**: - matrix_dot - matrix_transpose - matrix_add - matrix_sub - matrix_mul_elements - matrix_scale - matrix_sigmoid - matrix_dsigmoid - matrix_sum_rows - matrix_mse **matrices**: - matrix_dot - matrix_transpose - matrix_add - matrix_sub - matrix_mul_elements - matrix_scale - matrix_sigmoid - matrix_dsigmoid - matrix_sum_rows - matrix_mse **matrix**: - matrix_dot - matrix_transpose - matrix_add - matrix_sub - matrix_mul_elements - matrix_scale - matrix_sigmoid - matrix_dsigmoid - matrix_sum_rows - matrix_mse **media**: - ffmpeg - gif - video - speech - from_speech - translate **networking**: - web_get - web_send - listen - api **process**: - exec - run - spawn - python - js - sesi **process_and_scripting**: - exec - run - spawn - python - js - sesi **profiling**: - profile - profile_start - profile_end - profile_report **security**: **local_filesystem_override**: SESI_LOCAL_FS=true **protected_operations**: - exec - run - spawn - python - js - open - open_file - ffmpeg - gif - video - native_http_server - websocket_server - external_archive_access **safe_mode_default**: true **safe_mode_override**: SESI_SAFE_MODE=false **source_module**: builtins.ts **string_and_collection**: - split - join - slice - swap - contains - locate - starts_with - ends_with - index_of - repeat - includes - reverse - sort - unique - flatten - keys - values - push - pop - append - range **string_and_regex**: - to_upper - to_lower - trim - swap - contains - locate - starts_with - ends_with - index_of - repeat - includes - regex **strings**: - to_upper - to_lower - trim - slice - swap - contains - locate - starts_with - ends_with - index_of - repeat - includes - regex **text**: - to_upper - to_lower - trim - slice - swap - contains - locate - regex - tokenize **web**: - web_get - web_send - listen - api ### runtime_builtins **ai_and_tokens**: - speech - from_speech - translate - count_tokens - estimate_tokens - estimate_cost - model_usage - workflow - memory_search - memory_trim - memory_config - set_alias - define_tool - list_tools **collections**: - range - push - append - pop - keys - values - join - split - map - filter - reduce - find - reverse - sort - unique - flatten **conversion_and_types**: - num - float - bool - is_array - is_object - is_string - is_number - is_bool - is_null - is_function - name - arity **files_and_directories**: - read_file - write_file - append_file - write_image - open - open_file - list_dir - get_ext - exists - make_dir - rename - archive - trash - zip **general**: - show - debug - input - len - type - str - to_json - from_json - env - time - random **math**: - exp - matrix_dot - matrix_transpose - matrix_add - matrix_sub - matrix_mul_elements - matrix_scale - matrix_sigmoid - matrix_dsigmoid - matrix_sum_rows - matrix_mse **processes_and_scripting**: - exec - run - spawn - python - js - sesi - ffmpeg - gif - video **runtime_control**: - retry - lazy - force - timeout - profile - profile_start - profile_end - profile_report - error_type - raise_error **strings**: - to_upper - to_lower - trim - trunc - slice - swap - contains - locate - starts_with - ends_with - index_of - repeat - regex - tokenize **web_and_servers**: - web_get - web_send - listen - api - live ### modules **ai_runtime**: **cache**: **algorithm**: sha256 **cache_key_includes**: - model - prompt - temperature - max_tokens - top_k - top_p - images - system_prompt - thinking_level - search - tools - local_model_configuration **file**: .sesi_cache.json **exports**: - DEFAULT_LOCAL_MODEL - DEFAULT_LOCAL_MODEL_WARNING_TOKENS - AIRuntime - aiRuntime **key_methods**: - **name**: callModel **purpose**: Routes an AI request to local Transformers, OpenAI GPT, or Gemini; supports caching, streaming, tools, media, and generation controls. - **name**: countTokens **purpose**: Counts tokens through local tokenizer, OpenAI API, or Gemini API. - **name**: callVideo **purpose**: Generates videos using Gemini video or Omni-style interaction APIs. - **name**: synthesizeSpeech **purpose**: Produces WAV-compatible base64 audio through Gemini TTS. - **name**: transcribeSpeech **purpose**: Transcribes provided base64 audio via Gemini. - **name**: parseStructuredOutput **purpose**: Extracts and validates JSON-like structured output from model text. - **name**: embedText **purpose**: Generates cached Gemini embeddings. - **name**: searchMemory **purpose**: Uses cosine similarity over embeddings to retrieve memory chunks. - **name**: autoTrimMemory **purpose**: Automatically summarizes memory exceeding configured token limits. - **name**: trimMemory **purpose**: Compresses a memory entry using an AI summary model. **memory_summary**: **defaults**: **enabled**: true **max_tokens**: 900000 **summary_model**: gemini-3.5-flash-lite **target_tokens_ratio**: 0.6 **environment_variables**: **SESI_MEMORY_AUTO_SUMMARIZE**: Set to false to disable automatic summarization. **SESI_MEMORY_MAX_TOKENS**: Maximum memory token budget. **SESI_MEMORY_SUMMARY_MODEL**: Model used to summarize old history. **SESI_MEMORY_TARGET_TOKENS**: Desired compressed-memory token budget. **providers**: **gemini**: **api_key_environment_variable**: GEMINI_API_KEY **model_detection**: Models beginning with gemini- or models/gemini- **sdk_package**: @google/genai **supported_features**: - text_generation - streaming - function_tools - google_search - image_input - audio_input - image_generation - video_generation - text_to_speech - speech_to_text - embeddings - token_counting **local**: **default_model**: onnx-community/Qwen2.5-0.5B-Instruct **environment_variables**: **SESI_LOCAL_CACHE_DIR**: Custom local model-cache directory. **SESI_LOCAL_DEVICE**: Inference device; defaults to cpu. **SESI_LOCAL_DTYPE**: Model data type; defaults to q4. **SESI_LOCAL_MODEL**: Optional default Hugging Face model ID. **SESI_LOCAL_SYSTEM_PROMPT**: Default local-model system prompt. **SESI_LOCAL_WARN_TOKENS**: Input-token warning threshold. **package**: @huggingface/transformers **supported_features**: - text_generation - streaming - tool_calls - token_counting **unsupported_features**: - image_input - audio_input - web_search **openai**: **APIs**: - responses.create - responses.inputTokens.count - images.generate **api_key_environment_variable**: OPENAI_API_KEY **model_detection**: Models beginning with gpt- **sdk_package**: openai **supported_features**: - text_generation - streaming - tools - web_search - image_input - reasoning_effort - image_generation **builtins**: **categories**: **filesystem**: - read_file - write_file - append_file - write_image - list_dir - exists - get_ext - make_dir - rename - archive - trash - zip **general**: - show - debug - input - len - type - str - num - float - bool - time - random - env **matrix**: - matrix_dot - matrix_transpose - matrix_add - matrix_sub - matrix_mul_elements - matrix_scale - matrix_sigmoid - matrix_dsigmoid - matrix_sum_rows - matrix_mse **media**: - speech - from_speech - translate - ffmpeg - gif - video **memory_and_tools**: - memory_search - memory_trim - memory_config - set_alias - define_tool - list_tools - error_type - raise_error **networking**: - web_get - web_send - listen - api **process_and_scripting**: - exec - run - spawn - sesi - python - js **runtime_control**: - retry - lazy - force - timeout - profile - profile_start - profile_end - profile_report - map - filter - reduce - find - multi_req **serialization_and_crypto**: - to_json - from_json - encrypt - decrypt **string_and_collection**: - range - push - append - pop - keys - values - join - split - regex - slice - swap - contains - locate - to_upper - to_lower - trim - starts_with - ends_with - index_of - repeat - includes - reverse - sort - unique - flatten **tokenization_and_costs**: - tokenize - count_tokens - estimate_tokens - estimate_cost - model_usage **workflow**: - workflow - live **exports**: - getBuiltins - ensureSafePath - isTruthy - stripPrototypes - isEqual - stringify - compareValues **security**: **disabled_in_safe_mode**: - open - open_file - spawn - exec - run - python - js - ffmpeg - gif - video - non_zip_archive_access - native_http_server - websocket_server - std_game_run **local_filesystem_override**: SESI_LOCAL_FS **path_restriction**: File access is restricted to allowed directories unless local filesystem access is explicitly enabled outside safe mode. **safe_mode_default**: true **safe_mode_override**: SESI_SAFE_MODE **bytecode**: **exports**: - OpCode - makeChunk - addConstant - emitByte - emitBytes - emit16 - emitJump - patchJump - emitLoop - read16 - disassemble **module**: chunk **opcode_groups**: **ai**: - CALL_MODEL - CALL_IMAGE - CALL_VIDEO **collections**: - BUILD_ARRAY - BUILD_OBJECT - GET_INDEX - SET_INDEX - GET_PROPERTY - SET_PROPERTY **control_flow**: - JUMP - JUMP_IF_FALSE - LOOP - TRY_START - TRY_END - FINALLY_START - FINALLY_END **conversion**: - CONVERT **functions**: - CLOSURE - CALL - RETURN - RETURN_VOID - CALL_BUILTIN **modules_and_memory**: - IMPORT - ALLOW - INITIALIZE_MEMORY **operations**: - ADD - SUBTRACT - MULTIPLY - DIVIDE - MODULO - NEGATE - EQUAL - NOT_EQUAL - LESS - LESS_EQUAL - GREATER - GREATER_EQUAL - NOT **stack_and_constants**: - CONSTANT - NIL - true - false - POP **variables**: - DEFINE_GLOBAL - GET_GLOBAL - SET_GLOBAL - GET_LOCAL - SET_LOCAL - GET_UPVALUE - SET_UPVALUE - CLOSE_UPVALUE **compiler**: **class**: Compiler **compilation_support**: **expressions**: - literals - identifiers - binary_operations - unary_operations - logical_operations - assignments - function_calls - properties - indexes - arrays - objects - prompts - model_calls - image_calls - video_calls - structured_output - tool_calls - conversion - conditional_expressions - await_expressions **optimizations**: - Known builtin calls compile to CALL_BUILTIN. - show(...) compiles to PRINT. - Local variables use stack-slot access. - Nested functions capture upvalues. - Loop break and continue jumps are back-patched. **statements**: - let - function - expression - block - if_else - while - for - return - break - continue - try_catch_finally - import - allow - memory - export **input**: Sesi AST **output**: Chunk bytecode **dry_run_semantic_checks**: **diagnostics**: **undefined_symbol**: **code**: undefined-symbol **severity**: error **unused_symbol**: **code**: unused-symbol **severity**: warning **export**: runDryRunSemanticChecks **features**: - comment_stripping - lexical_tokenization - nested_scope_tracking - declaration_tracking - reference_tracking - builtin_symbol_exemptions - unused_binding_detection - make_scope_self_parameter_suppression **game_module**: **asset_handling**: **local_assets**: Embedded as base64 data URLs. **remote_assets**: Allows http, https, and data URLs. **export**: createGameModule **exports**: **create**: **game_methods**: **add**: **purpose**: Adds a validated game entity. **build**: **purpose**: Writes a standalone HTML game file. **rule**: **purpose**: Adds collision, key, timer, or bounds behavior. **run**: **purpose**: Starts a local preview server and optionally opens a browser. **returns**: game_object **module_name**: std/game **preview**: **host**: 127.0.0.1 **safe_mode_behavior**: disabled **purpose**: Provides a data-driven Canvas game system that creates self-contained HTML previews or game files. **supported_entity_shapes**: - rect - circle - sprite - text **supported_rule_actions**: - destroy - spawn - set - addScore - reverseVelocity - sound **supported_rule_events**: - collision - key - timer - bounds **std_api**: **defaults**: **cors**: true **cors_origin**: * **docs_path**: /docs **openapi_path**: /openapi.json **description**: FastAPI-style HTTP API framework with Swagger UI and OpenAPI 3.1. **functions**: - API.create_app(config = null) -> app - app.get(path, schema, handler) - app.post(path, schema, handler) - app.put(path, schema, handler) - app.patch(path, schema, handler) - app.delete(path, schema, handler) - app.use(middleware) - app.openapi() -> object - app.routes() -> array - app.listen(port, options = null) -> server **import**: allow "std/api" in as API **std_audio**: **functions**: - Audio.beep(frequency, duration) - Audio.play(note, duration, options = null) - Audio.synth(frequency_or_note, duration, type, options = null) -> base64_wav - Audio.save(path, frequency_or_note, duration, type, options = null) - Audio.load(path) -> audio_sample - Audio.kick(duration = 300, volume = 1.0) -> base64_wav - Audio.snare(duration = 200, volume = 1.0) -> base64_wav - Audio.hat(duration = 50, volume = 1.0) -> base64_wav - Audio.sequence(path, notes_array, type = null, options = null) - Audio.mix(path, tracks_array, type = null, options = null) - Audio.midi(path, tracks) - Audio.sf2(path, options = null) -> fn **import**: allow "std/audio" in as Audio **waveform_types**: - sine - square - saw - triangle - noise - kick - snare - hat - clap **std_base64**: **functions**: - Base64.encode(value, mode = text) - Base64.decode(base64_text, mode = text) **import**: allow "std/base64" in as Base64 **modes**: - text - bytes **std_browser**: **functions**: - launch(options = null) -> browser - browser.newPage() -> page - browser.close() - page.goto(url) - page.content() -> string - page.screenshot(options = null) -> string - page.click(selector) - page.fill(selector, value) - page.type(selector, value) - page.press(selector, key) - page.inner_text(selector) -> string - page.attribute(selector, name) -> string - page.evaluate(script) -> any - page.title() -> string - page.close() - page.pdf(options = null) -> string - page.wait_for_selector(selector, options = null) - page.wait_for_timeout(ms) **import**: allow "std/browser" in with {launch} **std_db**: **encryption**: Optional AES-256-CBC encryption is enabled by passing a database password. **functions**: - db_open(filename, password = null) - db.collection(name) - collection.insert(object) - collection.find(query_object = null) - collection.update(query_object, update_object) - collection.delete(query_object) **import**: allow "std/db" in with {db_open} **std_draw**: **functions**: - Draw.clear() - Draw.circle(x, y, radius, fill, options = {}) - Draw.rect(x, y, width, height, fill, options = {}) - Draw.line(x1, y1, x2, y2, stroke, options = {}) - Draw.text(x, y, content, size, fill, options = {}) - Draw.ellipse(cx, cy, rx, ry, fill, options = {}) - Draw.polygon(points, fill, options = {}) - Draw.path(d, fill, options = {}) - Draw.gradient(type, id, stops, options = {}) - Draw.style(cssText) - Draw.raw(svgCode) - Draw.render(width, height) -> string - Draw.save_svg(path, width, height) -> bool - Draw.pixel(x, y, color) - Draw.pixel_grid(grid, palette, scale = 1, x = 0, y = 0) - Draw.save_png(path, width, height, background = "transparent") -> bool **import**: allow "std/draw" in as Draw **std_game**: **bounds_modes**: - clamp - wrap - bounce **entity_shapes**: - rect - circle - sprite - text **functions**: - Game.create(config) -> game - game.add(entity) - game.rule(rule) - game.build(path) - game.run(options = null) -> handle **import**: allow "std/game" in as Game **rule_actions**: - destroy - spawn - set - addScore - reverseVelocity - sound **rule_events**: - collision - key - timer - bounds **std_terminal**: **formatting_functions**: - Terminal.color(text, color) - Terminal.style(text, styles) - Terminal.background(text, color) - Terminal.rgbBackground(text, red, green, blue) - Terminal.rgb(text, red, green, blue) **import**: allow "std/terminal" in as Terminal **terminal_functions**: - Terminal.clear(mode = screen) - Terminal.eraseLine(mode = all) - Terminal.eraseScreen(mode = all) - Terminal.cursor(x, y) - Terminal.move(x, y) - Terminal.up(amount = 1) - Terminal.down(amount = 1) - Terminal.left(amount = 1) - Terminal.right(amount = 1) - Terminal.saveCursor() - Terminal.restoreCursor() - Terminal.hideCursor() - Terminal.showCursor() - Terminal.write(text) - Terminal.line(text = null) - Terminal.title(text) - Terminal.bell() - Terminal.size() -> object **std_theory**: **chord_types**: - M - m - dim - aug - 7 - M7 - m7 - sus2 - sus4 **functions**: - Music.chord(root, type) -> array - Music.scale(root, type) -> array - Music.transpose(notes, semitones) -> array - Music.duration(minutes, seconds) -> number - Music.bar(bars, bpm, beatsPerBar = 4) -> number **import**: allow "std/theory" in as Music **scale_types**: - major - minor - dorian - phrygian - lydian - mixolydian - locrian **std_time**: **functions**: - Time.now() - Time.sleep(ms) - Time.format(timestamp, options = null) **import**: allow "std/time" in as Time ### standard_modules **std/api**: **capabilities**: - route_registration - middleware - query_validation - JSON_body_validation - CORS - OpenAPI_3_1 - Swagger_UI - routes - request_validation - OpenAPI_3_1_document_generation - validation - OpenAPI_generation **features**: - HTTP_routing - middleware - request_validation - CORS - OpenAPI_3_1_generation - Swagger_UI - query_validation - JSON_body_validation - Swagger_UI_docs - route_registration - cors - openapi_3_1 - swagger_ui **purpose**: HTTP API framework with OpenAPI and Swagger UI support **requires_safe_mode_disabled**: true **restricted_in_safe_mode**: true **route_methods**: - get - post - put - delete - patch - options - head **std/audio**: **capabilities**: - wav_synthesis - note_playback - waveform_generation - MIDI_export - SoundFont_rendering - track_mixing **exports**: - beep - play - synth - save - load - kick - snare - hat - sf2 - sequence - mix - midi - render - comp **features**: - wav_synthesis - note_playback - audio_sample_loading - sequencing - mixing - MIDI_generation - SoundFont_rendering - WAV_synthesis - audio_mixing - waveform_generation - music_theory_integration - synthesis - wav_generation - midi_generation - soundfont_rendering **purpose**: WAV_synthesis_and_MIDI_generation **std/base64**: **exports**: - encode - decode **modes**: - text - bytes **std/browser**: **capabilities**: - launch - newPage - goto - content - screenshot - click - fill - type - press - inner_text - attribute - evaluate - pdf - navigation - screenshots - PDF_generation - DOM_interaction - page_evaluation - Chromium_launch - PDF_export **dependency**: playwright **features**: - launch_browser - navigation - page_content - screenshots - PDF_generation - DOM_interaction - page_evaluation - launch - new_page - JavaScript_evaluation - chromium_launch - page_navigation - pdf_generation - dom_interaction **implementation**: Playwright **purpose**: Playwright_browser_automation **requires_safe_mode_disabled**: true **restricted_in_safe_mode**: true **std/db**: **capabilities**: - collection - insert - find - update - delete - optional_AES_256_CBC_encryption **collection_methods**: - insert - find - update - delete **collection_operations**: - insert - find - update - delete **description**: JSON-file document database with optional AES-256-CBC encryption. **exports**: - db_open **features**: - JSON_document_database - collections - insert - find - update - delete - optional_AES_256_CBC_encryption **operations**: - db_open - collection - insert - find - update - delete **optional_encryption**: aes-256-cbc **purpose**: JSON-file document database **storage**: JSON_file_database **std/draw**: **capabilities**: - vector_shapes - gradients - raw_SVG - pixel_art **exports**: - clear - circle - rect - pixel - pixel_grid - line - text - ellipse - polygon - path - gradient - style - raw - render - save_svg - save_png **features**: - SVG_generation - PNG_pixel_rendering - gradients - raw_svg - primitive_shapes - styles - svg_drawing - pixel_art - png_output **output_formats**: - SVG - PNG **purpose**: SVG_and_pixel_PNG_rendering **std/game**: **description**: Data-driven Canvas game construction and browser preview. **entity_shapes**: - rect - circle - sprite - text **exports**: - create **features**: - entities - sprite_assets - rules - collisions - keyboard_input - pointer_input - scores - timers - browser_preview - static_html_build - sprites - keyboard_controls - collision_rules - scoring - HTML_export - canvas_game_generation - entity_management - audio_effects - html_build - timer_rules - score_tracking - asset_embedding - local_preview_server - HTML_build_output **purpose**: data_driven_canvas_game_creation **rule_actions**: - destroy - spawn - set - addScore - reverseVelocity - sound **rule_events**: - collision - key - timer - bounds **type**: browser_canvas_game_builder **std/math**: **exports**: - PI - E - sin - cos - tan - sqrt - floor - ceil - abs - pow - log - exp **std/terminal**: **capabilities**: - colors - styles - cursor_control - screen_erasure - terminal_title - terminal_size - ANSI_styling - terminal_dimensions - screen_erasing **features**: - ANSI_colors - text_styles - cursor_control - screen_erasure - terminal_dimensions - terminal_title - ANSI_styling - screen_control - terminal_size - ANSI_styles - terminal_screen_control - ansi_styles - terminal_output **purpose**: ANSI_terminal_control **std/theory**: **exports**: - chord - scale - transpose - duration - bar **purpose**: music_theory_helpers **std/time**: **exports**: - now - sleep - format ### game_module **API**: **game.add**: **purpose**: Add a game entity **supported_shapes**: - rect - circle - sprite - text **game.build**: **output**: HTML file **game.rule**: **supported_actions**: - destroy - spawn - set - addScore - reverseVelocity - sound **supported_events**: - collision - key - timer - bounds **game.run**: **disabled_in_safe_mode**: true **output**: Local preview server object **preview_methods**: - url - stop **browser_runtime_features**: - canvas_rendering - keyboard_input - pointer_input - entity_physics - gravity - collision_detection - entity_bounds_modes - sprite_assets - score_hud - sound_effects - timer_rules **capabilities**: - create canvas games from data definitions - rectangle entities - circle entities - sprite entities - text entities - entity motion and gravity - keyboard controls - pointer input - collision detection - bounds handling - score tracking - timer rules - sound effects - HTML export - local preview server **exported_function**: create **game_configuration**: **optional**: - title - background - input - scores **required**: - width - height **implementation**: Native Canvas HTML runtime **methods**: - Game.create - game.add - game.rule - game.build - game.run **module**: std/game **rule_actions**: - destroy - spawn - set - addScore - reverseVelocity - sound **rule_events**: - collision - key - timer - bounds **type**: native_data_driven_canvas_module ### collections **arrays**: - **description**: Reverses an array in place. **mutates_input**: true **name**: reverse **pipe_supported**: true **signature**: reverse(array) -> array - **description**: Sorts an array in place. **mutates_input**: true **name**: sort **pipe_supported**: true **signature**: sort(array, compareFn = null) -> array - **description**: Returns a new array with duplicates removed. **name**: unique **pipe_supported**: true **signature**: unique(array) -> array - **description**: Flattens nested arrays by one level. **name**: flatten **pipe_supported**: true **signature**: flatten(array) -> array - **callback_arguments**: - item - index - array **name**: map **pipe_supported**: true **signature**: map(array, callback) -> array - **callback_arguments**: - item - index - array **name**: filter **pipe_supported**: true **signature**: filter(array, callback) -> array - **callback_arguments**: - accumulator - currentValue - index - array **name**: reduce **pipe_supported**: true **signature**: reduce(array, callback, initialValue = null) -> any - **callback_arguments**: - item - index - array **name**: find **pipe_supported**: true **signature**: find(array, callback) -> any|null **core**: - **description**: Returns the length of a string, array, or object. **name**: len **pipe_supported**: true **signature**: len(collection) -> number|null - **description**: Alias for len(). **name**: length **pipe_supported**: true **signature**: length(collection) -> number|null - **description**: Creates integers from 0 up to, but excluding, n. **name**: range **pipe_supported**: true **signature**: range(n) -> array - **description**: Adds an element to the end of an array in place. **mutates_input**: true **name**: push **pipe_supported**: true **signature**: push(array, value) -> array - **description**: Appends to an array in place or concatenates strings. **mutates_arrays**: true **name**: append **pipe_supported**: true **signature**: append(collection, value) -> array|string|null - **description**: Removes and returns the final array element. **mutates_input**: true **name**: pop **pipe_supported**: true **signature**: pop(array) -> any|null - **name**: join **pipe_supported**: true **signature**: join(array, separator) -> string - **name**: split **pipe_supported**: true **signature**: split(string, separator) -> array - **name**: keys **pipe_supported**: true **signature**: keys(object) -> array|null - **name**: values **pipe_supported**: true **signature**: values(object) -> array|null **strings**: - **description**: Runs a JavaScript-compatible regular expression. **modes**: - match - test - replace - split **name**: regex **options**: **flags**: JavaScript regex flags, such as i, m, s, and g. **limit**: Maximum matches or split parts; defaults to 10000. **replacement**: Required when mode is replace. **pipe_supported**: true **signature**: regex(pattern, text, options = null) -> array|bool|string - **description**: Tokenizes text with a tiktoken-style encoding or simple whitespace mode. **name**: tokenize **options**: **encoding**: Optional explicit tokenizer encoding. **mode**: Use "simple" for whitespace tokenization. **model**: Tokenizer model; defaults to "gpt-5.6-sol". **pipe_supported**: true **signature**: tokenize(string, options = null) -> array|array|null - **description**: Counts request tokens using a provider-native endpoint where applicable. **name**: count_tokens **pipe_supported**: true **providers**: **gemini**: Uses Gemini models.countTokens. **gpt**: Uses OpenAI responses/input_tokens. **local**: Uses cached local tokenizer. **signature**: count_tokens(string, options = null) -> number|null - **description**: Estimates token count locally without a provider request. **name**: estimate_tokens **pipe_supported**: true **signature**: estimate_tokens(string, options = null) -> number|null - **name**: to_upper **pipe_supported**: true **signature**: to_upper(string) -> string|null - **name**: to_lower **pipe_supported**: true **signature**: to_lower(string) -> string|null - **name**: trim **pipe_supported**: true **signature**: trim(string) -> string|null - **description**: Returns a sliced string or array without changing the original. **name**: slice **pipe_supported**: true **signature**: slice(collection, start, end = null) -> string|array|null - **description**: Replaces every occurrence of target in a string. **name**: swap **pipe_supported**: true **signature**: swap(string, target, replacement) -> string|null - **name**: contains **pipe_supported**: true **signature**: contains(string, sub) -> bool|null - **description**: Returns the first matching index or -1 when not found. **name**: locate **pipe_supported**: true **signature**: locate(string, sub) -> number|null - **name**: starts_with **pipe_supported**: true **signature**: starts_with(string, prefix) -> bool - **name**: ends_with **pipe_supported**: true **signature**: ends_with(string, suffix) -> bool - **description**: Returns first matching index or -1. **name**: index_of **pipe_supported**: true **signature**: index_of(collection, value) -> number - **name**: includes **pipe_supported**: true **signature**: includes(collection, value) -> bool - **name**: repeat **pipe_supported**: true **signature**: repeat(string, count) -> string ### math **builtins**: - **description**: Returns Euler's number raised to x. **name**: exp **pipe_supported**: true **signature**: exp(x) -> number - **description**: Truncates numeric fractional parts or string content. **name**: trunc **pipe_supported**: true **signature**: trunc(value, length = 0) -> number|string|null **std_math**: **constants**: - PI - E **functions**: - sin - cos - tan - sqrt - floor - ceil - abs - pow - log - exp **import**: allow "std/math" in as Math ### matrix **functions**: - **name**: matrix_dot **signature**: matrix_dot(a, b) -> array - **name**: matrix_transpose **signature**: matrix_transpose(matrix) -> array - **name**: matrix_add **notes**: A one-row b matrix broadcasts across rows of a. **signature**: matrix_add(a, b) -> array - **name**: matrix_sub **signature**: matrix_sub(a, b) -> array - **name**: matrix_mul_elements **signature**: matrix_mul_elements(a, b) -> array - **name**: matrix_scale **signature**: matrix_scale(matrix, scalar) -> array - **name**: matrix_sigmoid **signature**: matrix_sigmoid(matrix) -> array - **name**: matrix_dsigmoid **signature**: matrix_dsigmoid(matrix) -> array - **name**: matrix_sum_rows **signature**: matrix_sum_rows(matrix) -> array - **name**: matrix_mse **signature**: matrix_mse(a, b) -> number **validation**: Matrices must be non-empty rectangular arrays containing finite numbers. ### network - **name**: web_get **pipe_supported**: true **signature**: web_get(url, headers = {}) -> string - **name**: web_send **pipe_supported**: true **signature**: web_send(url, body, headers = {}) -> string - **description**: Starts a native HTTP server. **handler_request_fields**: - method - path - headers - body - query **handler_response**: **object_fields**: - status - headers - body **string**: HTTP 200 text/html response. **name**: listen **pipe_supported**: true **returns**: Server control object with close(). **signature**: listen(port, handler) -> object - **client_methods**: - send - close **description**: Starts a native WebSocket server. **handler_arguments**: - client - message **name**: api **pipe_supported**: true **returns**: Server control object with close(). **signature**: api(port, handler) -> object ### io - **description**: Prints values to standard output, separated by spaces. **name**: show **pipe_supported**: true **returns**: null **signature**: show(...args) -> null - **description**: Displays an optional terminal prompt and waits for user input. **name**: input **parameters**: **prompt**: Optional prompt text. **pipe_supported**: true **returns**: string **signature**: input(prompt = null) -> string ### filesystem - **modes**: **base64**: Reads raw bytes as Base64. **text**: Reads UTF-8 text. **name**: read_file **pipe_supported**: true **signature**: read_file(path, mode = "text") -> string|null - **encodings**: **base64**: Decodes Base64 content before writing bytes. **null**: Writes UTF-8 text. **name**: write_file **pipe_supported**: true **signature**: write_file(path, content, encoding = null) -> bool - **name**: append_file **pipe_supported**: true **signature**: append_file(path, content) -> bool - **name**: write_image **pipe_supported**: true **signature**: write_image(path, base64_content) -> bool - **description**: Opens a URL or local file through the OS default or selected app. **name**: open **options**: **browser**: Preferred browser application. **editor**: Preferred text editor. **image_viewer**: Alias for viewer. **mode**: auto, browser, editor, viewer, or image_viewer. **viewer**: Preferred image viewer. **pipe_supported**: true **signature**: open(target, options = null) -> bool - **name**: open_file **pipe_supported**: true **signature**: open_file(path, options = null) -> bool - **name**: list_dir **pipe_supported**: true **signature**: list_dir(path) -> array - **name**: make_dir **pipe_supported**: true **signature**: make_dir(path) -> bool - **name**: rename **pipe_supported**: true **signature**: rename(oldPath, newPath) -> bool - **default_destination**: .archive/ **description**: Copies a file or directory recursively. **name**: archive **pipe_supported**: true **signature**: archive(sourcePath, destPath = null) -> bool - **description**: Returns a lowercased extension without the leading dot; preserves compound archive extensions. **name**: get_ext **pipe_supported**: true **signature**: get_ext(path) -> string - **name**: exists **pipe_supported**: true **signature**: exists(path) -> bool - **name**: zip **operations**: - create - list - extract **signature**: zip(source, destination = null, operation = null) -> bool|array **supported_extensions**: - zip - jar - apk - docx - xlsx - pptx - epub - 7z - rar - tar - tar.gz - tgz - tar.bz2 - tbz2 - tar.xz - txz - **description**: Moves an item to .trash by default, or permanently deletes it with autoRemove=true. **name**: trash **pipe_supported**: true **signature**: trash(path, autoRemove = false) -> bool ### system - **description**: Runs a Sesi script as a concurrent background process and returns its PID. **name**: spawn **pipe_supported**: true **signature**: spawn(path) -> number - **description**: Creates a hot-reloading wrapper around an exported Sesi function. **name**: live **pipe_supported**: true **signature**: live(filePath, exportName = "handle") -> fn - **description**: Synchronously executes a shell command. **name**: exec **pipe_supported**: true **signature**: exec(command) -> string - **description**: Exact alias of exec(). **name**: run **pipe_supported**: true **signature**: run(command) -> string - **description**: Parses, compiles, and runs another Sesi file synchronously. **name**: sesi **pipe_supported**: true **signature**: sesi(filePath, local = false, checkOnly = false) -> string - **description**: Runs Python source with the host Python runtime. **name**: python **pipe_supported**: true **requirements**: Unavailable in Sesi safe mode. **signature**: python(code, args = null) -> string - **description**: Runs JavaScript source using the Sesi Node.js runtime. **name**: js **pipe_supported**: true **requirements**: Unavailable in Sesi safe mode. **signature**: js(code, args = null) -> string - **description**: Builds a complete HTML document. **name**: html **options**: **head**: Raw markup inserted in head. **lang**: Defaults to "en". **title**: Defaults to "Sesi". **pipe_supported**: true **signature**: html(body, options = null) -> string - **description**: Retrieves one environment variable or all environment variables. **name**: env **pipe_supported**: true **signature**: env(key = null, defaultValue = null) -> string|object - **description**: Returns the current Unix timestamp in milliseconds. **name**: time **signature**: time() -> number - **description**: Formats a Unix timestamp as readable text. **name**: format **pipe_supported**: true **signature**: format(timestamp, options = null) -> string - **description**: Runs function closures, built-ins, or async functions concurrently. **name**: multi_req **pipe_supported**: true **signature**: multi_req(fns) -> array - **description**: Runs a multi-step model workflow using previous step outputs as context. **name**: workflow **pipe_supported**: false **result_fields**: - input - steps - final **signature**: workflow(steps, input = "") -> object **step_fields**: **optional**: - model - temperature - max_tokens - top_k - top_p - thinkingLevel - cache - search **required**: - prompt - **description**: Registers a local alias resolved by model(), image(), and workflow(). **name**: set_alias **pipe_supported**: true **signature**: set_alias(alias, model) -> bool - **description**: Registers a custom Sesi function as an AI-accessible tool. **name**: define_tool **signature**: define_tool(name, fn, description = "") -> bool - **description**: Invokes a registered custom tool. **name**: tool_call **signature**: tool_call(name)(...args) -> any - **name**: list_tools **signature**: list_tools() -> array - **name**: error_type **pipe_supported**: true **returns_fields**: - type - message - data **signature**: error_type(type, message, data = null) -> object - **description**: Throws a typed error. **name**: raise_error **pipe_supported**: true **signature**: raise_error(type_or_error, message = "", data = null) -> never - **name**: retry **options**: **backoff_factor**: 2 **initial_delay**: 1000 **max_retries**: 3 **pipe_supported**: true **signature**: retry(action, options = null) -> any - **description**: Creates a memoized delayed computation. **name**: lazy **signature**: lazy(action, ...args) -> lazy - **description**: Resolves lazy values or promises; returns ordinary values unchanged. **name**: force **signature**: force(value) -> any - **description**: Runs an action with a deadline; returns fallback or throws TimeoutError. **name**: timeout **signature**: timeout(action, ms, fallback = unset) -> any - **name**: profile **signature**: profile(name, action) -> any - **name**: profile_start **signature**: profile_start(name) -> string - **name**: profile_end **signature**: profile_end(name) -> object - **name**: profile_report **signature**: profile_report(format = "object") -> array|string - **description**: Returns a random float in the range [0, 1). **name**: random **signature**: random() -> number ### conversion **supported_categories**: **document**: **fallback**: Gemini_document_conversion **local_conversions**: - markdown_to_html - html_to_markdown - html_to_text - svg_to_html - html_to_svg - svg_to_text - csv_to_json - tsv_to_json - json_to_csv - json_to_tsv - json_to_yaml - yaml_to_json **media**: **formats**: - images - SVG - audio - video **tools**: - sharp - Playwright - ImageMagick - ffmpeg ### conversion_media - **config**: **file_type**: Optional input extension; inferred for local file paths. **output_type**: Required target extension. **name**: convert **native_document_conversions**: - md_to_html - html_to_md - html_to_txt - csv_to_json - tsv_to_json - json_to_csv - json_to_tsv - json_to_yaml - yaml_to_json - svg_to_html - html_to_svg - svg_to_txt **signature**: convert(type) { config } { file } -> string **svg_media_conversions**: - svg_to_png - svg_to_jpg - svg_to_jpeg - png_to_svg - jpg_to_svg - jpeg_to_svg - gif_to_svg - webp_to_svg - bmp_to_svg - tiff_to_svg - avif_to_svg **types**: **audio**: Audio. **doc**: Documents and text. **media**: Images and video. - **name**: gif **options**: **fps**: 12 **loop**: 0 means loop forever. **overwrite**: true **timeout**: Milliseconds. **width**: Optional output width. **pipe_supported**: true **requirements**: Requires ffmpeg and is unavailable in safe mode. **signature**: gif(input, output, options = null) -> string - **ai_options**: - images - image - ratio - aspectRatio - duration - resolution - negative_prompt - audio - task **forms**: **ai_generation**: Returns base64 MP4 data. **local_ffmpeg**: Creates or transcodes a local video and returns output path. **local_options**: - fps - width - height - codec - crf - pixel_format - preset - audio - mute - overwrite - timeout **name**: video **pipe_supported**: true **signature**: video(input, output, options = null) -> string - **description**: Runs FFmpeg using a structured argument array; shell command strings are rejected. **name**: ffmpeg **pipe_supported**: true **requirements**: Requires ffmpeg and is unavailable in safe mode. **result_fields**: - ok - code - signal - stdout - stderr - args **signature**: ffmpeg(args, options = null) -> object ### media **audio**: **capabilities**: - Gemini_TTS - Gemini_transcription - local_whisper_transcription - local_system_speech **conversion**: **document**: **fallback**: Gemini **local**: - markdown_to_html - html_to_markdown - html_to_text - SVG_HTML_conversion - CSV_JSON_conversion - TSV_JSON_conversion - JSON_YAML_conversion **raster_and_media**: **tools**: - sharp - playwright - ImageMagick - FFmpeg - pandoc **image_generation**: **providers**: - Gemini - OpenAI **video_generation**: **modes**: - Veo_generateVideos - Omni_interactions_create **output**: base64_video_data **provider**: Gemini ### speech_translation - **description**: Speaks text through the system voice engine. **name**: speech **parameters**: **gemini_model**: Optional Gemini TTS model; returns base64 audio instead of playback. **text**: Text to speak. **voice**: Optional installed system voice. **pipe_supported**: true **returns**: true after playback, or base64 audio when using a Gemini TTS model. **signature**: speech(text, voice = null, gemini_model = null) -> bool|string - **description**: Transcribes an audio file using nodejs-whisper or Gemini. **name**: from_speech **parameters**: **audio_path**: Path to the audio file. **gemini_model**: Optional Gemini transcription model. **language**: Optional language code, such as en or fr. **pipe_supported**: true **requirements**: **whisper**: nodejs-whisper and a downloaded Whisper model are required for local transcription. **returns**: Transcript text. **signature**: from_speech(audio_path, language = null, gemini_model = null) -> string - **description**: Translates text using the translate package or an optional Gemini model. **name**: translate **parameters**: **from_language**: Source ISO code or English language name. **gemini_model**: Optional Gemini model. **text**: Source text. **to_language**: Target ISO code or English language name. **pipe_supported**: true **returns**: Translated text. **signature**: translate(text, to_language, from_language = "en", gemini_model = null) -> string --- ## Compiler & VM ### bytecode **chunk**: **fields**: **code**: Flat opcode and operand byte array **constants**: Constant pool **lines**: Source line table **compiler**: **builtin_optimization**: Known built-in calls compile to CALL_BUILTIN rather than resolving a function object dynamically. **class**: Compiler **output**: Chunk **supports**: - global_variables - local_variables - lexical_scopes - closures - upvalues - default_function_parameters - if_else - while_loops - for_range_loops - for_iterable_loops - break - continue - try_catch_finally - imports - memory_declarations - model_calls - image_calls - video_calls - structured_output - tool_calls - conversion_expressions **core_structures**: **Chunk**: **fields**: - code - constants - lines **FunctionProto**: **fields**: - name - arity - params - chunk - isAsync - upvalues **key_opcodes**: **flow_control**: - JUMP - JUMP_IF_FALSE - LOOP - TRY_START - TRY_END - FINALLY_START - FINALLY_END **operations**: - ADD - SUBTRACT - MULTIPLY - DIVIDE - MODULO - NEGATE - EQUAL - NOT_EQUAL - LESS - LESS_EQUAL - GREATER - GREATER_EQUAL - NOT **runtime**: - CALL - CALL_BUILTIN - CALL_MODEL - CALL_IMAGE - CALL_VIDEO - CONVERT - PRINT - IMPORT - ALLOW - INITIALIZE_MEMORY **stack**: - CONSTANT - NIL - true - false - POP **variables**: - DEFINE_GLOBAL - GET_GLOBAL - SET_GLOBAL - GET_LOCAL - SET_LOCAL - GET_UPVALUE - SET_UPVALUE - CLOSE_UPVALUE **module**: chunk **opcode_groups**: **AI**: - CALL_MODEL - CALL_IMAGE - CALL_VIDEO **ai**: - CALL_MODEL - CALL_IMAGE - CALL_VIDEO **arithmetic**: - ADD - SUBTRACT - MULTIPLY - DIVIDE - MODULO - NEGATE **collections**: - BUILD_ARRAY - BUILD_OBJECT - GET_INDEX - SET_INDEX - GET_PROPERTY - SET_PROPERTY **comparison**: - EQUAL - NOT_EQUAL - LESS - LESS_EQUAL - GREATER - GREATER_EQUAL - NOT **control_flow**: - JUMP - JUMP_IF_FALSE - LOOP - TRY_START - TRY_END - FINALLY_START - FINALLY_END **conversion**: - CONVERT **functions**: - CLOSURE - CALL - CALL_BUILTIN - RETURN - RETURN_VOID **memory**: - INITIALIZE_MEMORY **modules**: - IMPORT - ALLOW **output**: - PRINT **stack_and_constants**: - CONSTANT - NIL - TRUE - FALSE - POP - true - false **variables**: - DEFINE_GLOBAL - GET_GLOBAL - SET_GLOBAL - GET_LOCAL - SET_LOCAL - GET_UPVALUE - SET_UPVALUE - CLOSE_UPVALUE **source_modules**: - chunk.ts - compiler.ts **virtual_machine_model**: stack_based ### compiler **architecture**: **compiler**: Compiler **execution_engines**: - bytecode_vm - tree_walking_interpreter **lexer**: Lexer **parser**: Parser **source_language**: Sesi **built_in_call_opcode**: CALL_BUILTIN **bytecode**: **opcode_categories**: - constants_and_stack - global_variables - local_variables - arithmetic - comparison - logical_operations - control_flow - collections - functions_and_closures - builtins - AI_calls - output - exceptions - imports - memory - upvalues - conversions **structure**: - code - constants - lines **supports**: - closures - upvalues - local_slot_variables - default_parameters - loops - break - continue - try_catch_finally - imports - model_calls - image_calls - video_calls - conversion_calls - bytecode_disassembly **class**: Compiler **optimizations**: - slot_based_local_variables - closure_upvalue_capture - builtin_fast_path - short_circuit_logical_operators - bytecode_jump_patching **source**: Sesi_AST **supported_expressions**: - literals - identifiers - binary_operations - unary_operations - logical_operations - assignments - function_calls - member_access - index_access - arrays - objects - prompts - model_calls - image_calls - video_calls - structured_output - tool_calls - conversion - conditional_expressions - await_expressions **supported_statements**: - let - function - expression - block - if_else - while - for - return - break - continue - try_catch_finally - import - allow - memory - export **target**: bytecode_chunk ### compiler_and_vm **bytecode**: **core_types**: - Chunk - FunctionProto - OpCode **features**: - constant pool - source line table - bytecode disassembly - closures and upvalues - local stack slots - globals - arrays and objects - property and index access - function calls - built-in fast path - AI call opcodes - loops - conditional branches - try/catch/finally - imports and module permissions - memory initialization **compiler**: **class**: Compiler **supported_constructs**: - variables - functions - default parameters - blocks and lexical scopes - if/else - while loops - for loops - break and continue - return - try/catch/finally - imports - allow statements - memory declarations - prompt expressions - model, image, and video expressions - structured output - tool calls - conversion expressions - conditional expressions - await expressions **specialized_opcodes**: - CALL_MODEL - CALL_IMAGE - CALL_VIDEO - INITIALIZE_MEMORY - CONVERT - CALL_BUILTIN - CLOSURE - GET_UPVALUE - SET_UPVALUE - CLOSE_UPVALUE ### execution **engines**: **bytecode_vm**: **capabilities**: - globals - local_slots - closures - upvalues - control_flow - builtins_fast_path - model_call_opcodes - try_catch_finally **components**: - chunk - compiler - vm - disassembler **default**: true **tree_walking_interpreter**: **capabilities**: - lexical_scopes - async_functions - model_tool_orchestration - standard_module_loading - runtime_type_values **role**: fallback **safety**: **default_safe_mode**: true **environment_variables**: **SESI_LOCAL_FS**: Set to true to permit local filesystem access when safe mode is disabled. **SESI_SAFE_MODE**: Set to false to enable unsafe process and external-tool features. **restricted_in_safe_mode**: - exec - run - spawn - python - js - ffmpeg - gif - video - open - open_file - native_http_server - websocket_server - browser_module - non_zip_archive_access - external_media_processes ### semantic_analysis **checks**: - lexical declarations - function declarations - parameters - imports - loop variables - catch variables - memory and prompt declarations - tool declarations - builtin symbol recognition - scope nesting **component**: Dry-run scope checker **diagnostics**: **errors**: - undefined-symbol **warnings**: - unused-symbol ### semantic_checks **analysis**: - strips line and block comments while preserving strings - tokenizes source code - creates lexical scope hierarchy - tracks declarations and references - recognizes built-ins - validates imports and allow bindings - suppresses selected unused warnings for loop variables and underscore-prefixed names **diagnostics**: **undefined_symbol**: **code**: undefined-symbol **severity**: error **unused_symbol**: **code**: unused-symbol **severity**: warning **function**: runDryRunSemanticChecks **module**: dry-run-semantic-checks.ts ### static_analysis **dry_run**: **diagnostics**: - undefined_symbols - unused_symbols - type_mismatches - return_type_mismatches **type_system**: **supported_annotations**: - number - string - bool - any - array - object - optional_types - union_types ### validation_and_analysis **dry_run_checker**: **diagnostics**: - undefined-symbol - unused-symbol **profiler**: **metrics**: - invocation_count - total_duration_ms - average_duration_ms - minimum_duration_ms - maximum_duration_ms - last_duration_ms **type_checker**: **diagnostics**: - type-mismatch - return-type-mismatch - argument-type-mismatch ### types **assertions**: - **name**: is_array **signature**: is_array(value) -> bool - **name**: is_object **signature**: is_object(value) -> bool - **name**: is_string **signature**: is_string(value) -> bool - **name**: is_number **signature**: is_number(value) -> bool - **name**: is_bool **signature**: is_bool(value) -> bool - **name**: is_null **signature**: is_null(value) -> bool - **name**: is_function **signature**: is_function(value) -> bool **conversion**: - **name**: type **pipe_supported**: true **returns**: One of: number, string, bool, null, array, object, unknown. **signature**: type(value) -> string - **description**: Converts any value to a string representation. **name**: str **pipe_supported**: true **signature**: str(value) -> string - **description**: Converts an array or object to formatted valid JSON. **name**: to_json **pipe_supported**: true **returns**: Formatted JSON string. **signature**: to_json(value) -> string - **description**: Parses JSON into a native Sesi primitive, array, or object. **name**: from_json **pipe_supported**: true **returns**: Parsed value, or null when parsing fails. **signature**: from_json(string) -> any - **description**: Encrypts UTF-8 content with AES-256-CBC. **name**: encrypt **pipe_supported**: true **returns**: iv:ciphertext encrypted payload. **signature**: encrypt(content, password) -> string - **description**: Decrypts an AES-256-CBC iv:ciphertext payload. **name**: decrypt **pipe_supported**: true **returns**: Decrypted UTF-8 content. **signature**: decrypt(content, password) -> string - **description**: Converts a value to a number. **name**: num **pipe_supported**: true **returns**: Number or null when conversion fails. **signature**: num(value) -> number|null - **description**: Converts a value to a floating-point number. **name**: float **pipe_supported**: true **returns**: Number or null when conversion fails. **signature**: float(value) -> number|null - **description**: Converts a value using Sesi truthiness rules. **name**: bool **pipe_supported**: true **signature**: bool(value) -> bool **pipe_supported**: true --- ## CLI & Tools ### cli **entry_functions**: - runSesi - runSesiFile **options**: - safeMode - allowLocalFs - allowedPaths - encrypt - decrypt - password - raw - ast - tokens - dry - bytecode - bytecodeDump - treeWalker - timeoutMs - profile **output_modes**: - token_table - raw_ast_json - AST_tree - bytecode_disassembly - semantic_diagnostics - runtime_profile ### cli_runtime **diagnostics**: - token table output - AST tree output - bytecode disassembly - dry-run semantic checks - type checks - structured runtime errors **entrypoints**: - runSesi - runSesiFile **execution_modes**: **bytecode_vm**: **default**: true **tree_walker**: **option**: treeWalker **supported_options**: - safeMode - allowLocalFs - allowedPaths - encrypt - decrypt - password - raw - ast - tokens - dry - bytecode - bytecodeDump - treeWalker - timeoutMs - deadlineAt - profile ### developer_tools **bytecode**: **CLI_option**: bytecodeDump **disassembler**: disassemble **dry_run**: **diagnostics**: **errors**: - undefined-symbol **warnings**: - unused-symbol **semantic_checker**: runDryRunSemanticChecks **profiling**: **class**: SesiProfiler **report_formatter**: formatProfileReport **type_checking**: **function**: runTypeChecks ### debugging - **description**: Pauses execution and opens the interactive sesi-debug REPL. **name**: debug **pipe_supported**: true **repl_commands**: **continue**: Resumes normal execution. **env**: Shows active lexical scope variables. **eval**: Evaluates a Sesi expression in the active scope. **help**: Displays debugger help. **signature**: debug(message = null) -> null ### quality_tools **diagnostics**: - lexer_errors - parser_errors - compiler_errors - runtime_errors - structured_runtime_errors - stack_traces **dry_run**: **semantic_checks**: - undefined_symbol_detection - unused_symbol_detection - type_checks **profiling**: **class**: SesiProfiler **reports**: - object - text ### tool_orchestration **blocked_automated_tools**: - exec - run - spawn - python - js - ffmpeg - gif - video **custom_tools**: **automatic_execution**: supported **default_max_tool_calls**: 8 **discovery**: list_tools **discovery_builtin**: list_tools **listing_builtin**: list_tools **max_tool_calls_default**: 8 **registration**: define_tool **registration_builtin**: define_tool **schema_generation**: JSON_Schema **supported_parameter_types**: - number - string - bool - array - object - optional - union - any **supports**: - typed_parameter_schema_generation - required_parameter_detection - automatic_execution - multi_turn_model_tool_loops - tool_result_transcripts **execution**: **behavior**: - send_tool_schemas_to_model - parse_model_tool_call - validate_tool_arguments - execute_registered_tool - append_serialized_tool_result_to_follow_up_prompt - continue_until_final_model_response **blocked_automated_tools**: - exec - run - spawn - python - js - ffmpeg - gif - video **max_default_tool_calls**: 8 **protected_automated_tools**: - exec - run - spawn - python - js - ffmpeg - gif - video **schema_generation**: **format**: JSON_schema_function_declarations **source**: Sesi function parameters and type annotations **supported_tool_formats**: - OpenAI_function_tools - Gemini_functionDeclarations - local_transformers_function_tools ### package_management **installation**: **archive_extractors**: **unix**: - unzip - tar **windows**: - PowerShell Expand-Archive - tar **package_manager_user_agent**: Sesi-PackageManager/1.8.6 **source**: GitHub_zipball **source_archive**: GitHub ZIP archive **manifest**: sesi.json **modules_directory**: sesi_modules **supported_sources**: - owner/repository#ref - github:owner/repository#ref - https://github.com/owner/repository **supported_specifications**: - owner/repository - owner/repository#ref - github:owner/repository - https://github.com/owner/repository ### package_manager **installation**: **extraction_tools**: - unzip - tar - PowerShell_Expand_Archive **source**: GitHub_zipball **installation_method**: - download_GitHub_zipball - extract_archive - save_to_sesi_modules - update_manifest_dependencies **installation_source**: GitHub ZIP archives **manifest**: sesi.json **modules_directory**: sesi_modules **source**: pm.ts **source_support**: - owner/repository#ref - github:owner/repository#ref - https://github.com/owner/repository#ref **supported_sources**: - owner/repository#reference - github:owner/repository#reference - https://github.com/owner/repository#reference **supported_specifications**: - owner/repository - owner/repository#ref - github:owner/repository#ref - https://github.com/owner/repository ### pricing **currency**: USD **pricing_as_of**: 2026-07-27 **pricing_sources**: **gemini**: https://ai.google.dev/gemini-api/docs/pricing **openai**: https://developers.openai.com/api/docs/pricing **source**: token-pricing.ts **supported_model_families**: - gemini-2.5 - gemini-3.1 - gemini-3.5 - gemini-3.6 - gpt-5.6 **unit**: per_million_tokens ### token_pricing **currency**: USD **functionality**: - per_million_token_rates - long_context_rate_thresholds - input_output_cost_estimation - custom_rate_overrides **pricing_as_of**: 2026-07-27 **pricing_sources**: **gemini**: https://ai.google.dev/gemini-api/docs/pricing **openai**: https://developers.openai.com/api/docs/pricing **providers**: **gemini**: **configured_models**: - gemini-3.6-flash - gemini-3.5-flash-lite - gemini-3.5-flash - gemini-3.1-flash-lite - gemini-3.1-pro-preview - gemini-2.5-flash-lite - gemini-2.5-flash - gemini-2.5-pro **source**: https://ai.google.dev/gemini-api/docs/pricing **openai**: **configured_models**: - gpt-5.6-sol - gpt-5.6-terra - gpt-5.6-luna **source**: https://developers.openai.com/api/docs/pricing **sources**: **gemini**: https://ai.google.dev/gemini-api/docs/pricing **openai**: https://developers.openai.com/api/docs/pricing **supported_families**: - gemini - gpt-5.6 **supported_model_families**: - gemini-2.5 - gemini-3.1 - gemini-3.5 - gemini-3.6 - gpt-5.6 **unit**: per 1,000,000 tokens ### introspection - **name**: name **pipe_supported**: true **signature**: name(func) -> string|null - **name**: arity **pipe_supported**: true **signature**: arity(func) -> number|null ### error_return_reference **keys**: null **len**: null **num**: null **pop_empty_array**: null **str**: String representation, including null as string text. **type**: unknown **values**: null ### performance_notes - show() is unbuffered and flushes on every call. - Most array operations are O(n). - Object key access is generally O(1). - String concatenation with + is O(n). --- ## Configuration & Environment ### configuration **environment_variables**: **GEMINI_API_KEY**: **purpose**: Gemini_provider_authentication **OPENAI_API_KEY**: **purpose**: OpenAI_provider_authentication **SESI_DEBUG**: **purpose**: interpreter_execution_debug_logging **SESI_LOCAL_CACHE_DIR**: **purpose**: local_model_cache_directory **SESI_LOCAL_DEVICE**: **purpose**: local_model_device **SESI_LOCAL_DTYPE**: **purpose**: local_model_quantization_dtype **SESI_LOCAL_FS**: **purpose**: allow_local_filesystem_when_safe_mode_is_disabled **SESI_LOCAL_MODEL**: **purpose**: override_default_local_model **SESI_LOCAL_SYSTEM_PROMPT**: **purpose**: default_local_system_prompt **SESI_LOCAL_WARN_TOKENS**: **purpose**: local_model_input_warning_threshold **SESI_MEMORY_AUTO_SUMMARIZE**: **purpose**: enable_or_disable_memory_auto_summarization **SESI_MEMORY_MAX_TOKENS**: **purpose**: memory_maximum_token_budget **SESI_MEMORY_SUMMARY_MODEL**: **purpose**: memory_summary_model **SESI_MEMORY_TARGET_TOKENS**: **purpose**: memory_summary_target_token_budget **SESI_PASSWORD**: **purpose**: default_encryption_password **SESI_PATH**: **purpose**: additional_module_search_paths **SESI_SAFE_MODE**: **purpose**: control_default_safe_mode ### constants **DEFAULT_LOCAL_MODEL**: onnx-community/Qwen2.5-0.5B-Instruct **DEFAULT_LOCAL_MODEL_WARNING_TOKENS**: 2048 ### environment_variables **GEMINI_API_KEY**: Gemini API authentication key **OPENAI_API_KEY**: OpenAI API authentication key **SESI_DEBUG**: Set to 1 to print interpreter execution debug output **SESI_LOCAL_CACHE_DIR**: Local model cache directory **SESI_LOCAL_DEVICE**: Local inference device **SESI_LOCAL_DTYPE**: Local ONNX model quantization/dtype **SESI_LOCAL_FS**: Set to true to permit local filesystem access when safe mode is off **SESI_LOCAL_MODEL**: Default local Hugging Face model **SESI_LOCAL_SYSTEM_PROMPT**: Default local-model system prompt **SESI_LOCAL_WARN_TOKENS**: Local input token warning threshold **SESI_MEMORY_AUTO_SUMMARIZE**: Automatic memory summarization toggle **SESI_MEMORY_MAX_TOKENS**: Memory maximum token budget **SESI_MEMORY_SUMMARY_MODEL**: Memory summary model **SESI_MEMORY_TARGET_TOKENS**: Memory target token budget **SESI_PASSWORD**: Default encryption password **SESI_PATH**: Additional module search paths **SESI_SAFE_MODE**: Set to false to relax safe-mode restrictions ### global_variables **args**: **description**: Command-line script arguments excluding interpreter options and script filename. **type**: array ### security **additional_protections**: - prototype_pollution_blocking - archive_entry_traversal_validation - sensitive_automated_tool_blocking - constrained_local_module_imports **archive_protection**: - validates_archive_entry_paths - blocks_absolute_archive_paths - blocks_null_byte_paths - blocks_directory_traversal_entries **automated_tools**: **blocked_sensitive_tools**: - exec - run - spawn - python - js - ffmpeg - gif - video **blocked_automated_tools**: - exec - run - spawn - python - js - ffmpeg - gif - video **controls**: - path_traversal_protection - allowed_directory_enforcement - archive_entry_validation - prototype_pollution_protection - restricted_external_process_execution_in_safe_mode - restricted_browser_server_and_websocket_features_in_safe_mode - restricted_non_zip_archive_access_in_safe_mode - restricted_automated_sensitive_tool_execution **cryptography**: **algorithm**: AES-256-CBC **functions**: - encrypt - decrypt **implementation_details**: **initialization_vector**: Random 16-byte IV **key_derivation**: SHA-256 password hash **serialized_format**: : **default_mode**: safe **defaults**: **environment_override**: SESI_SAFE_MODE=false **local_filesystem_access**: false **local_filesystem_override**: SESI_LOCAL_FS=true **safe_mode**: true **environment_controls**: **local_filesystem**: SESI_LOCAL_FS **safe_mode**: SESI_SAFE_MODE **environment_variables**: **SESI_LOCAL_FS**: Set to "true" to permit local filesystem access **SESI_SAFE_MODE**: Set to "false" to disable safe mode **filesystem**: **allowed_paths**: - current_working_directory - script_directory - explicitly_configured_paths **default_allowed_directory**: current_working_directory **environment_variable**: SESI_LOCAL_FS **optional_local_access**: **environment_variable**: SESI_LOCAL_FS **required_value**: true **path_validation**: ensureSafePath **path_validation_function**: ensureSafePath **path_validator**: ensureSafePath **protected_against**: - path_traversal **protections**: - resolves paths against a controlled base directory - blocks traversal outside allowed paths - supports explicit allowed paths - supports script-directory access - validates archive extraction entries - path_traversal_prevention - archive_entry_validation - prototype_pollution_protection **path_protection**: **behavior**: - resolves_paths_against_allowed_directories - blocks_path_traversal - supports_interpreter_specific_allowed_paths - supports_script_directory_allowlisting **controls**: - path_traversal_prevention - allowed_directory_enforcement - script_directory_access - optional_local_filesystem_mode **function**: ensureSafePath **protections**: - path_traversal_prevention - allowed_directory_validation - safe_mode_local_filesystem_lockdown **protected_operations**: - local_filesystem_access - external_application_launching - shell_execution - Python_execution - JavaScript_execution - FFmpeg_execution - native_HTTP_server - WebSocket_server - non_ZIP_archive_access **protections**: - filesystem_path_traversal_validation - prototype_pollution_blocking - safe_mode_process_execution_restrictions - safe_mode_external_application_restrictions - safe_mode_http_server_restrictions - safe_mode_browser_restrictions - safe_mode_non_zip_archive_restrictions - safe_mode_ffmpeg_restrictions - sensitive_automated_tool_blocklist - archive_entry_path_validation **prototype_pollution**: **blocked_assignment_keys**: - __proto__ - prototype - constructor - __defineGetter__ - __defineSetter__ - __lookupGetter__ - __lookupSetter__ **protections**: - stripPrototypes - **blocked_property_names**: - __proto__ - prototype - constructor - __defineGetter__ - __defineSetter__ - __lookupGetter__ - __lookupSetter__ **sanitization_function**: stripPrototypes **restricted_in_safe_mode**: - exec - run - spawn - python - js - ffmpeg - gif - video - open - open_file - browser_module - native_http_server - websocket_server - external_archive_access - media_conversion - non_zip_archive_access - std/browser - std/game.run - command_line_media_conversion **safe_mode**: **default**: true **effects**: - disables unrestricted_local_filesystem_access - disables_external_application_launching - disables_process_spawning - disables_shell_execution - disables_python_execution - disables_javascript_execution - disables_ffmpeg_and_media_processing - disables_non_zip_archive_access - disables_native_http_and_websocket_servers - disables_browser_automation **environment_override**: SESI_SAFE_MODE=false **environment_variable**: SESI_SAFE_MODE **restrictions**: - local filesystem access is constrained to allowed directories - arbitrary shell execution is disabled - Python execution is disabled - JavaScript execution is disabled - external process spawning is disabled - opening external applications is disabled - native HTTP and WebSocket servers are disabled - FFmpeg and media processing are disabled - non-ZIP archive access is disabled - game preview server is disabled - local_filesystem_access - process_execution - browser_launching - native_http_server - websocket_server - external_archive_access - media_processing **safe_mode_default**: true **sensitive_tools**: - exec - run - spawn - python - js - ffmpeg - gif - video ### security_notes **archive_protection**: - Validates extraction entry paths. - Blocks non-ZIP archive access in safe mode. **model_tool_protection**: - Prevents automated execution of sensitive process-control tools. **path_protection**: - Resolves paths against configured allowed directories. - Blocks traversal outside allowed roots in safe mode. **prototype_pollution_protection**: - Uses prototype-free objects for external or parsed data. - Blocks assignments to dangerous prototype-related property names. ### memory - **description**: Searches a memory binding using semantic similarity. **embedding_models**: **fallback**: gemini-embedding-2 **primary**: gemini-embedding-001 **name**: memory_search **pipe_supported**: true **result_fields**: - text - score **signature**: memory_search(name, query, top_k = 3) -> array - **description**: Configures automatic memory summarization. **name**: memory_config **options**: **enabled**: true **max_tokens**: 900000 **summary_model**: gemini-3.5-flash-lite **target_tokens**: 60% of max_tokens by default. **pipe_supported**: true **signature**: memory_config(name, options = {}) -> object - **description**: Compacts a memory binding when it exceeds its token budget. **name**: memory_trim **pipe_supported**: true **signature**: memory_trim(name, max_tokens = 900000) -> string --- ## Dependencies & Exports ### dependencies **AI**: - @huggingface/transformers - @google/genai - openai **ai**: - @huggingface/transformers - @google/genai - openai **archives**: - adm-zip **browser**: - playwright **conversion**: - html-to-text - js-yaml - marked - turndown - sharp - image-size **external_tools**: - ffmpeg - tar - 7z - 7zz - unrar - rar - cmake - Python - espeak-ng - macOS say - PowerShell System.Speech **media**: - sharp - image-size - playwright **networking**: - ws **node_standard_library**: - fs - os - path - http - crypto - child_process - vm **optional_or_feature_specific**: - playwright - translate - nodejs-whisper - smart-whisper - node-wav - ffmpeg - ImageMagick - pandoc - fluidsynth - espeak-ng **optional_or_platform_dependent**: - ffmpeg - tar - 7zip - unrar - rar - nodejs-whisper - smart-whisper - node-wav - translate - espeak-ng - cmake **parsing_and_conversion**: - js-yaml - marked - turndown - html-to-text **required_or_core**: - @google/genai - openai - @huggingface/transformers - js-tiktoken - adm-zip - ws **required_or_optional**: - @huggingface/transformers - @google/genai - openai - js-tiktoken - adm-zip - ws - translate - html-to-text - nodejs-whisper - smart-whisper - node-wav **required_or_primary**: - @google/genai - @huggingface/transformers - openai - js-tiktoken - js-yaml - marked - turndown - sharp - html-to-text - image-size - adm-zip - ws **runtime**: - @huggingface/transformers - @google/genai - openai - js-tiktoken - adm-zip - ws - translate - nodejs-whisper - smart-whisper - node-wav **tokenization**: - js-tiktoken **transcription_optional**: - nodejs-whisper - smart-whisper - node-wav **utilities**: - js-tiktoken - adm-zip - ws - smart-whisper - nodejs-whisper - node-wav ### key_dependencies **ai**: - @google/genai - @huggingface/transformers - openai **conversion**: - sharp - marked - turndown - html-to-text - image-size - js-yaml **runtime**: - js-tiktoken - adm-zip - ws - playwright **speech**: - nodejs-whisper - smart-whisper - node-wav ### external_dependencies **ai**: - @huggingface/transformers - openai - @google/genai - js-tiktoken **archives**: - adm-zip **browser_automation**: - playwright **document_conversion**: - html-to-text - js-yaml - marked - turndown **image_processing**: - sharp - image-size **optional_system_tools**: - ffmpeg - pandoc - ImageMagick - tar - 7zip - unrar - fluidsynth - cmake - espeak-ng **transcription**: - nodejs-whisper - smart-whisper - node-wav **websocket**: - ws ### entry_points **execution**: - runSesi - runSesiFile **exports**: - Lexer - Parser - Interpreter - Environment - Compiler - VM - disassemble - runInstall **supported_options**: - safeMode - allowLocalFs - allowedPaths - encrypt - decrypt - password - raw - ast - tokens - args - dry - bytecode - bytecodeDump - treeWalker - timeoutMs - deadlineAt - profile ### exported_instances **aiRuntime**: **singleton**: true **type**: AIRuntime ### imports - @huggingface/transformers - ./types - fs - os - path - crypto - openai - @google/genai ### interfaces **MemorySummaryConfig**: **fields**: **enabled**: boolean **maxTokens**: number **summaryModel**: string **targetTokens**: number **MemoryTrimResult**: **fields**: **error**: string? **summarized**: boolean **summaryModel**: string? **text**: string **tokensAfter**: number **tokensBefore**: number **usage**: AIResponse.usage? **VideoGenerationRequest**: **fields**: **audio**: boolean? **duration**: number? **images**: string[]? **model**: string **negativePrompt**: string? **pollInterval**: number? **prompt**: string **ratio**: string? **resolution**: string? **task**: string? ### module **language**: TypeScript **name**: ai-runtime **purpose**: Provides a unified AI runtime supporting local Hugging Face Transformers, OpenAI GPT, Google Gemini, image generation, video generation, speech, embeddings, caching, tool calls, streaming, and memory summarization. ### module_resolution **search_order**: - Directory containing the running Sesi script. - Current working directory. - Directories specified by SESI_PATH. - ~/.sesi/lib global library directory. **sesi_path_separator**: **unix**: : **windows**: ; ### utility_functions **cosineSimilarity**: **edge_cases**: - Returns 0 for empty vectors - Returns 0 for unequal vector lengths - Returns 0 when either vector has zero magnitude **purpose**: Calculates cosine similarity between equally sized numeric vectors. **importTransformersModule**: **async**: true **behavior**: - Dynamically imports "@huggingface/transformers" - Enables remote model downloads through transformers.env.allowRemoteModels - Returns the Transformers module **stripPrototypes**: **purpose**: Recursively converts objects into prototype-free plain objects. ### analysis_and_diagnostics **bytecode**: **opcode_categories**: - stack_and_constants - globals_and_locals - arithmetic - comparisons - control_flow - collections - closures - builtins - ai_calls - exceptions - imports - memory **supports_disassembly**: true **dry_run**: **checks**: - undefined_symbols - unused_symbols - declared_type_mismatches - return_type_mismatches **profiling**: **implementation**: SesiProfiler **measurements**: - count - total_ms - average_ms - min_ms - max_ms - last_ms ---