"""pydantic-deep: Deep agent framework built on pydantic-ai. This library provides a deep agent framework with: - Planning via TodoToolset - Filesystem operations via create_console_toolset (from pydantic-ai-backend) - Subagent delegation via SubAgentToolset - Multiple backend options for file storage - Structured output support via output_type - History processing/summarization for long conversations Example: ```python from pydantic import BaseModel from pydantic_deep import ( create_deep_agent, DeepAgentDeps, LocalBackend, create_console_toolset ) from pydantic_ai_summarization import create_summarization_processor # Create agent with file tools agent = create_deep_agent( model="anthropic:claude-sonnet-4-6", instructions="You are a helpful coding assistant", ) # With structured output and summarization class Analysis(BaseModel): summary: str issues: list[str] agent = create_deep_agent( output_type=Analysis, history_processors=[ create_summarization_processor( trigger=("tokens", 100000), keep=("messages", 20), ) ], ) # Create dependencies with LocalBackend deps = DeepAgentDeps(backend=LocalBackend(root_dir=".")) # Run agent result = await agent.run("Create a Python script", deps=deps) print(result.output) ``` """ from pydantic_ai_backends import ( BUILTIN_RUNTIMES, AsyncBaseSandbox, BackendProtocol, BaseSandbox, CompositeBackend, ConsoleDeps, DockerSandbox, EditResult, ExecuteResponse, FileData, FileInfo, GrepMatch, LocalBackend, RuntimeConfig, SandboxProtocol, SessionManager, StateBackend, WriteResult, create_console_toolset, get_console_system_prompt, get_runtime, is_async_backend, ) from pydantic_ai_shields import ( BudgetExceededError, CostInfo, CostTracking, InputBlocked, OutputBlocked, ToolBlocked, ) from pydantic_ai_summarization import ( ContextManagerCapability, LimitWarnerCapability, SlidingWindowCapability, SlidingWindowProcessor, SummarizationCapability, SummarizationProcessor, create_sliding_window_processor, create_summarization_processor, ) from pydantic_ai_todo import create_todo_toolset as TodoToolset from subagents_pydantic_ai import SubAgentToolset from pydantic_deep._text import NUM_CHARS_PER_TOKEN, create_content_preview from pydantic_deep.agent import create_deep_agent, create_default_deps, run_with_files from pydantic_deep.deps import DEFAULT_USAGE_LIMITS as DEFAULT_USAGE_LIMITS from pydantic_deep.deps import DeepAgentDeps, unwrap_backend from pydantic_deep.features.browser import BrowserCapability, BrowserToolset from pydantic_deep.features.checkpointing import ( Checkpoint, CheckpointMiddleware, CheckpointStore, CheckpointToolset, FileCheckpointStore, InMemoryCheckpointStore, RewindRequested, fork_from_checkpoint, ) from pydantic_deep.features.context import ( DEFAULT_CONTEXT_FILENAMES, DEFAULT_MAX_CONTEXT_CHARS, SUBAGENT_CONTEXT_ALLOWLIST, ContextFile, ContextFilesCapability, ContextToolset, discover_context_files, format_context_prompt, load_context_files, ) from pydantic_deep.features.eviction import ( BINARY_PRUNED_TEMPLATE, DEFAULT_EVICTION_PATH, DEFAULT_MAX_BINARY_CONTENT, DEFAULT_TOKEN_LIMIT, EVICTION_MESSAGE_TEMPLATE, EvictionCapability, ) from pydantic_deep.features.forking import ( JUDGE_SYSTEM_PROMPT, BranchOverlay, ForkBranchLimitError, ForkCoordinator, ForkDepthLimitError, ForkStateStore, InMemoryForkStateStore, JudgeAgent, build_diff_report, clone_for_branch, compute_confidence, count_retry_parts, count_stuck_loop_hits, create_fork_toolset, ) from pydantic_deep.features.forking.capability import LiveForkCapability from pydantic_deep.features.forking.types import ( BranchChange, BranchCost, BranchDiffAgreement, BranchDiffOperation, BranchDiffReport, BranchIsolation, BranchOutcome, BranchSpec, BranchState, BranchStatus, ConfidenceSignals, DiffSummary, FileChange, FlushError, FlushReport, ForkCostSummary, ForkHandle, JudgeVerdict, MergeResult, MergeStrategy, PathDiff, PendingApprovalRequest, ResolveOutcome, ) from pydantic_deep.features.hooks import ( DEFAULT_BLOCKED_COMMANDS, DEFAULT_BLOCKED_READ_PATHS, DEFAULT_BLOCKED_WRITE_PATHS, DEFAULT_SECRET_PATTERNS, EXIT_ALLOW, EXIT_DENY, Hook, HookEvent, HookInput, HookResult, HooksCapability, default_security_hook, ) from pydantic_deep.features.liteparse import ( PARSE_DOCUMENT_DESCRIPTION, SCREENSHOT_DOCUMENT_DESCRIPTION, LiteparseToolset, ) from pydantic_deep.features.memory import ( DEFAULT_MAX_MEMORY_LINES, DEFAULT_MEMORY_DIR, DEFAULT_MEMORY_FILENAME, DEFAULT_PIN_END_MARKER, AgentMemoryToolset, MemoryAccessError, MemoryCapability, MemoryFile, format_memory_prompt, get_memory_path, load_memory, ) from pydantic_deep.features.monitoring import ( MonitorEvent, MonitorInfo, MonitorManager, create_monitor_toolset, ) from pydantic_deep.features.patch import ( CANCELLED_MESSAGE, PatchToolCallsCapability, patch_tool_calls_processor, ) from pydantic_deep.features.periodic_reminder import ( LLMReminderGenerator, PeriodicReminderCapability, PeriodicReminderConfig, ReminderGenerator, make_config_for_mode, ) from pydantic_deep.features.plan import PlanOption, create_plan_toolset from pydantic_deep.features.skills import ( BackendSkillResource, BackendSkillScript, BackendSkillScriptExecutor, BackendSkillsDirectory, CallableSkillScriptExecutor, FileBasedSkillResource, FileBasedSkillScript, LocalSkillScriptExecutor, Skill, SkillException, SkillNotFoundError, SkillResource, SkillResourceLoadError, SkillResourceNotFoundError, SkillsCapability, SkillScript, SkillScriptExecutionError, SkillsDirectory, SkillsToolset, SkillValidationError, SkillWrapper, ) from pydantic_deep.features.stuck_loop import StuckLoopDetection, StuckLoopError from pydantic_deep.features.teams import ( AgentTeam, SharedTodoItem, SharedTodoList, TeamMember, TeamMemberHandle, TeamMemberSpec, TeamMessage, TeamMessageBus, create_team_toolset, ) from pydantic_deep.goal import ( DEFAULT_GOAL_MODEL, GoalEvaluation, GoalEvaluator, GoalState, build_goal_transcript, format_goal_status, goal_continue_directive, parse_goal_command, ) from pydantic_deep.mcp import ( BUILTIN_MCP_NAMES, MCPAuth, MCPConfigError, MCPNotInstalledError, MCPProbeResult, MCPRegistry, MCPServerConfig, auth_satisfied, build_mcp_server, builtin_mcp_servers, create_mcp_resources_toolset, parse_mcp_servers, probe_mcp_server, ) from pydantic_deep.prompts import BASE_PROMPT from pydantic_deep.spec import DeepAgent, DeepAgentSpec from pydantic_deep.styles import ( BUILTIN_STYLES, OutputStyle, discover_styles, format_style_prompt, load_style_from_file, resolve_style, ) from pydantic_deep.types import ( BrowseResult, CompiledSubAgent, ResponseFormat, SubAgentConfig, Todo, UploadedFile, ) try: from importlib.metadata import version __version__ = version("pydantic-deep") except Exception: # pragma: no cover __version__ = "0.0.0" __all__ = [ # Main entry points "BASE_PROMPT", "create_deep_agent", "create_default_deps", "run_with_files", "DeepAgentDeps", "DEFAULT_USAGE_LIMITS", "unwrap_backend", "DeepAgent", "DeepAgentSpec", # Goal-completion loop engine "DEFAULT_GOAL_MODEL", "GoalEvaluation", "GoalEvaluator", "GoalState", "build_goal_transcript", "format_goal_status", "goal_continue_directive", "parse_goal_command", # MCP (Model Context Protocol) client support "MCPServerConfig", "MCPAuth", "MCPRegistry", "MCPProbeResult", "MCPConfigError", "MCPNotInstalledError", "build_mcp_server", "create_mcp_resources_toolset", "probe_mcp_server", "auth_satisfied", "builtin_mcp_servers", "BUILTIN_MCP_NAMES", "parse_mcp_servers", # Backends (from pydantic-ai-backend) "BackendProtocol", "SandboxProtocol", "LocalBackend", "StateBackend", "CompositeBackend", "BaseSandbox", "AsyncBaseSandbox", "is_async_backend", "DockerSandbox", # Runtimes "RuntimeConfig", "BUILTIN_RUNTIMES", "get_runtime", # Session Management "SessionManager", # Capabilities (pydantic-ai AbstractCapability) "BrowserCapability", "SkillsCapability", "ContextFilesCapability", "MemoryCapability", "StuckLoopDetection", "StuckLoopError", "PeriodicReminderCapability", "PeriodicReminderConfig", "ReminderGenerator", "LLMReminderGenerator", "make_config_for_mode", "HooksCapability", "ContextManagerCapability", "SummarizationCapability", "SlidingWindowCapability", "LimitWarnerCapability", "CostTracking", # Toolsets "BrowserToolset", "LiteparseToolset", "PARSE_DOCUMENT_DESCRIPTION", "SCREENSHOT_DOCUMENT_DESCRIPTION", "TodoToolset", "create_console_toolset", "get_console_system_prompt", "ConsoleDeps", "SubAgentToolset", "SkillsToolset", "create_plan_toolset", # Skills types "Skill", "SkillResource", "SkillScript", "SkillWrapper", "SkillsDirectory", # Skills local "FileBasedSkillResource", "FileBasedSkillScript", "LocalSkillScriptExecutor", "CallableSkillScriptExecutor", # Skills backend "BackendSkillResource", "BackendSkillScript", "BackendSkillScriptExecutor", "BackendSkillsDirectory", # Skills exceptions "SkillException", "SkillNotFoundError", "SkillValidationError", "SkillResourceNotFoundError", "SkillResourceLoadError", "SkillScriptExecutionError", # Context files "ContextToolset", "ContextFile", "load_context_files", "discover_context_files", "format_context_prompt", "DEFAULT_CONTEXT_FILENAMES", "DEFAULT_MAX_CONTEXT_CHARS", "SUBAGENT_CONTEXT_ALLOWLIST", # Memory (persistent agent memory) "AgentMemoryToolset", "MemoryAccessError", "MemoryFile", "load_memory", "get_memory_path", "format_memory_prompt", "DEFAULT_MEMORY_DIR", "DEFAULT_MEMORY_FILENAME", "DEFAULT_MAX_MEMORY_LINES", "DEFAULT_PIN_END_MARKER", # Eviction "EvictionCapability", "create_content_preview", "NUM_CHARS_PER_TOKEN", "DEFAULT_TOKEN_LIMIT", "DEFAULT_EVICTION_PATH", "DEFAULT_MAX_BINARY_CONTENT", "EVICTION_MESSAGE_TEMPLATE", "BINARY_PRUNED_TEMPLATE", # Processors (from summarization-pydantic-ai) "SummarizationProcessor", "SlidingWindowProcessor", "create_summarization_processor", "create_sliding_window_processor", # Hooks (Claude Code-style lifecycle hooks) "DEFAULT_BLOCKED_COMMANDS", "DEFAULT_BLOCKED_READ_PATHS", "DEFAULT_BLOCKED_WRITE_PATHS", "DEFAULT_SECRET_PATTERNS", "Hook", "HookEvent", "HookInput", "HookResult", "EXIT_ALLOW", "EXIT_DENY", "default_security_hook", # Checkpointing "Checkpoint", "CheckpointStore", "InMemoryCheckpointStore", "FileCheckpointStore", "CheckpointMiddleware", "CheckpointToolset", "RewindRequested", "fork_from_checkpoint", # Teams (parallel subagents and agent teams) "AgentTeam", "SharedTodoItem", "SharedTodoList", "TeamMember", "TeamMemberHandle", "TeamMemberSpec", "TeamMessage", "TeamMessageBus", "create_team_toolset", "MonitorManager", "MonitorEvent", "MonitorInfo", "create_monitor_toolset", "PlanOption", # Output styles "OutputStyle", "BUILTIN_STYLES", "resolve_style", "load_style_from_file", "discover_styles", "format_style_prompt", # Patch tool calls processor "PatchToolCallsCapability", "patch_tool_calls_processor", "CANCELLED_MESSAGE", # Shields (from pydantic-ai-shields) "CostInfo", "BudgetExceededError", "InputBlocked", "ToolBlocked", "OutputBlocked", # Types "BrowseResult", "FileData", "FileInfo", "WriteResult", "EditResult", "ExecuteResponse", "GrepMatch", "Todo", "SubAgentConfig", "CompiledSubAgent", "UploadedFile", "ResponseFormat", # Live Run Forking - coordinator, isolation, diff, and public types "LiveForkCapability", "ForkCoordinator", "ForkStateStore", "InMemoryForkStateStore", "BranchOverlay", "build_diff_report", "clone_for_branch", "create_fork_toolset", "ForkBranchLimitError", "ForkDepthLimitError", "BranchSpec", "BranchIsolation", "BranchState", "BranchStatus", "BranchCost", "ForkCostSummary", "ForkHandle", "MergeStrategy", "MergeResult", "FlushError", "FlushReport", "PendingApprovalRequest", "FileChange", "BranchDiffReport", "PathDiff", "BranchChange", "DiffSummary", "BranchDiffAgreement", "BranchDiffOperation", # Live Run Forking - autonomous judge for unattended merge resolution "JudgeAgent", "JudgeVerdict", "ConfidenceSignals", "BranchOutcome", "ResolveOutcome", "JUDGE_SYSTEM_PROMPT", "compute_confidence", "count_retry_parts", "count_stuck_loop_hits", ]