# ANTRI Code (`antri_cli`) **An intelligent, terminal-first AI coding chatbot, autonomous meta-agent, desktop control plane, and mobile ecosystem inspired by OpenCode.** [![npm version](https://img.shields.io/npm/v/antri_cli.svg?style=flat-square&color=8b5cf6)](https://www.npmjs.com/package/antri_cli) [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg?style=flat-square)](LICENSE) [![Node: >=18](https://img.shields.io/badge/node-%3E%3D18.0.0-emerald.svg?style=flat-square)](https://nodejs.org)
ANTRI Code Banner
--- ## Why ANTRI Code? ANTRI Code turns your developer workspace into an integrated, autonomous coding ecosystem across the **Terminal CLI**, **Desktop Control Plane**, and **Mobile App**. It doesn't just wait for prompts โ€” it breaks tasks down step-by-step, asks clarifying questions on architecture trade-offs, runs multi-iteration self-improvement loops, debates solutions against itself to hunt edge cases, and writes its own custom dynamic tools. --- ## Visual Interfaces & Multi-Surface Ecosystem ### 1. Terminal CLI Interface The lightning-fast, keyboard-driven terminal REPL with interactive `@` file attachment picking, mode badges, and instant response streaming.
ANTRI Code Terminal Home Interface

### 2. Desktop Control Plane A minimalist cream-white desktop control center with visual agent studio, live dialectic debate cards, iterative goal pipelines, thinking profile editors, memory graph inspectors, **`+` file/image uploader**, and **interactive prompt toolkit autocomplete menu**.
ANTRI Desktop Control Plane

### 3. Mobile App & PWA A standalone, touch-optimized mobile control center designed for iOS and Android. Features a modern **Gemini-style sliding side menu (Drawer)** with a `+ New chat` pill button, active session highlighting, **click-and-hold (long-press) chat options** to delete or rename sessions, standalone thinking profiles, camera/gallery `+` upload button, silent dialectic debates, and local memory. --- ## Quick Start & One-Shot Setup ### Global Installation Install globally from npm with a single command: ```bash npm install -g antri_cli ``` ### Launch Interactive CLI Run `antri` anywhere in your terminal: ```bash antri ``` ### Launch Desktop Control Plane Open the lightweight desktop control plane in app mode: ```bash antri desktop ``` *(or type `/desktop` from inside the CLI)* ### Launch Mobile App Server Start the standalone mobile app server accessible on phone or local network: ```bash antri mobile ``` *(or type `/mobile` from inside the CLI)* ### Zero-Lockfile Self-Updating Update globally to the latest version anytime: ```bash antri update ``` --- ## Key Highlights & Core Architecture ### 1. Plan Mode vs. Vibe Mode - **Plan Mode (`/plan` or `antri --mode plan`)**: The agent enters an interactive blueprinting session. It analyzes architecture, drafts roadmaps, specifies file changes, and asks clarifying questions before making code edits. - **Vibe Mode (`/vibe` or `antri --mode vibe`)**: Continuous conversational coding. The agent immediately writes code, creates files, executes tools, and builds solutions in real time. ### 2. Multi-File & Multimodal Uploads (`+` Button & `@` Mentions) - **Desktop & Mobile `+` Attachment Button**: Upload screenshots, code files, architecture diagrams, or take live camera photos directly in chat. - **Interactive `@` File Attachment Picker**: Type `@` in the prompt box to trigger an interactive folder & file explorer. Navigate with arrow keys, hit Enter, and inject full file contexts directly into your conversation. - **Interactive `/` Prompt Toolkit Palette**: Type `/` in the Desktop and Terminal prompt box for live command autocomplete with full keyboard navigation. ### 3. Comprehensive Workspace & Coding Tools (CLI & Desktop) ANTRI includes surgical code manipulation and repository tools built directly into the agent: - `edit_file`: Precise search-and-replace block editing preserving indentation, comments, and structure without full rewrites. - `create_directory` & `delete_file`: File and folder scaffolding and cleanup. - `find_files`: Glob and extension file discovery (e.g. `*.ts`, `src/**/*.tsx`), filtering out noise. - `grep_search`: Codebase-wide regex and literal pattern search with file paths, line numbers, and matching snippets. - `file_info` & `git_diff`: File metadata inspection and uncommitted git diff analysis. - `read_file`, `write_file`, `list_dir`, `search_files`, `run_command`, `execute_python`. ### 4. Autonomous Silent Debate & Goal Engines (Background Synthesis) When asked for deep research, architectural tradeoffs, or multi-step goal execution, ANTRI runs reasoning engines **secretly in the background** without polluting your terminal or chat logs with intermediate chatter: - **Silent Dialectic Debate (`/silent-debate [query]`, `/sdebate`, `run_silent_debate`)**: The agent secretly pits a *Proposer (Thesis)*, *Adversary (Antithesis)*, and *Researcher (Verification)* against each other before delivering the final consensus under a badge: ```markdown > โš”๏ธ [Dialectic Debate Synthesized] [Authoritative conclusion and hardened recommendations] ``` - **Silent Goal Loop (`/silent-goal [task]`, `/sgoal`, `run_silent_goal`)**: The agent runs a 3-stage optimization loop (Draft -> Adversarial Critique & Score -> Hardened Synthesis) and returns: ```markdown > ๐ŸŽฏ [Goal Loop Plan Synthesized] [Production-ready implementation and architecture plan] ``` ### 5. Claude-Style Interactive Artifacts, Visual Graphs & Mind Maps (`/imagine`, `/mindmap`, `/view`, Artifacts Hub) ANTRI allows you to generate, preview, and interact with rich visual applications, plans, diagrams, and hierarchical concept mind maps directly inside the interface: - **`๐Ÿง  /mindmap [topic/concept]`**: Automatically generates rich, interactive, and beautifully styled hierarchical **Visual Mind Maps** (rendered with Mermaid 10 and dark cyber neon styling) complete with Zoom In/Out, Reset, Pan gestures, and code export on both Desktop and Mobile! - **`๐Ÿ“Š /imagine [topic/code]`**: Automatically crafts an interactive visual code architecture diagram and flowchart graph (rendered with Mermaid/SVG nodes). - **`๐ŸŒ /view [plan/app]`**: Generates a complete, self-contained interactive HTML/CSS/JS application or plan (e.g. *"generate me a 7-day workout plan"*, interactive habit tracker, dashboard, or calculator). - **"๐Ÿ‘๏ธ View Artifact" In-Chat Action**: Every generated artifact in chat displays a sleek interactive card. Tapping **"View Artifact"** launches the live, interactive app/diagram directly inside the app interface without exposing raw code! - **Dedicated "Artifacts" Hub (Side Drawer)**: Located directly **above the `+ New chat` button** in the sliding navigation drawer. Displays all generated artifacts neatly separated and grouped by **Chat Title / Session** (e.g. `Chat - 1` with its 2 artifacts) with one-click interactive viewing and management. ### 6. Tasteful Emojis Directive (Max 2 Emojis) ANTRI strictly adheres to a clean, professional communication standard with a hard limit of **maximum 2 emojis per response**, ensuring readability without clutter. ### 7. 100% Silent & Nuanced Philosophy Note Capture ANTRI continuously listens and adapts to your thinking style without interruptive banners or snackbars: - **Completely Silent**: Note capture happens 100% in the background. - **Minute Nuance & Philosophy Extraction**: Automatically recognizes philosophical worldviews (*stoicism, pragmatism, existentialism, ethics, first principles, epistemic humility*), core values, thinking habits, and subtle personal preferences, compounding them into your active thinking profile (`~/.antri/profiles/profile_1.md`). - View captured insights anytime with `/notes`. ### 8. Multi-Profile Thinking System (`/profile`, `/notes`) - Maintains distinct Markdown thinking profiles (`~/.antri/profiles/profile_1.md`, `profile_2.md` on Desktop/CLI, and on-device profiles on Mobile). - **Interactive Profile Selector**: Switch profiles on the fly with `/profile`. - **RAG Active Context**: Profile rules and notes are seamlessly recalled into conversations. ### 9. Multi-Tiered Persistent Memory (Lifelong Learning) - **Episodic Store**: Session transcripts and debate histories. - **Semantic Memory**: Dense 128-dimensional vector store with cosine similarity. - **Workspace Conventions**: Remembers repository-specific patterns (`.antri/conventions.md`). - **Autonomous Self-Recall**: Queries past solutions before answering new prompts. - `/memory` (inspect store) ยท `/consolidate` (run reflection) ยท `/learn ` (save rule). ### 10. The Meta-Agent (Autonomous Self-Evolution) - **Sandboxed Python Runtime (`execute_python`)**: Safe execution of code snippets. - **Autonomous Self-Debugging**: Intercepts stack traces, analyzes root causes, creates patches, and retries automatically. - **Dynamic Skill Synthesis (`synthesize_skill`)**: The agent writes its own Python/JS tools, verifies them via dry-run, and persists them in `~/.agent-cli/skills/`. - **Meta-Optimization (`/meta`)**: Tracks tool latency, success rates, and refines prompt heuristics. --- ## Supported AI Providers & Model Suites ANTRI defaults to **Google Gemini 3.7** via the official **Google GenAI SDK (`@google/genai`)**, while also supporting multi-provider fallback: | Provider | Description | Notable Models | |---|---|---| | **Google Gemini** *(Default Flagship)* | Next-gen hybrid reasoning, coding & 1M context via `@google/genai` | `gemini-3.7-flash`, `gemini-3.7-pro`, `gemini-3.5-pro`, `gemini-2.5-flash` | | **OpenAI** | Frontier reasoning & omni models | `gpt-4o`, `gpt-4o-mini`, `o1`, `o3-mini`, `gpt-4.5-preview` | | **Anthropic** | Industry standard coding models | `claude-3-7-sonnet`, `claude-3-5-sonnet`, `claude-3-5-haiku` | | **DeepSeek** | Frontier open-weights & reasoning | `deepseek-chat` (V3), `deepseek-reasoner` (R1), `deepseek-v4-flash` | | **Cerebras** | Ultra-fast ~2,000 tok/sec CS-3 inference | `llama-3.3-70b`, `llama3.1-70b`, `llama3.1-8b`, `qwen-2.5-coder-32b` | | **Cohere** | Enterprise reasoning & citations | `command-r-plus-08-2024`, `command-r-08-2024`, `command-r7b-12-2024` | | **Vortex API** | High-throughput GPU cluster | `vortex-llama-3.3-70b`, `vortex-deepseek-r1-full`, `vortex-deepseek-v3` | | **OpenCode** | Dedicated coding & architecture | `opencode/deepseek-coder-v2.5`, `opencode/qwen2.5-coder-32b-instruct` | | **NVIDIA NIM** | 100+ Live NIM API cloud functions | `meta/llama-3.2-11b-vision-instruct`, `stepfun-ai/step-3.7-flash` | | **Ollama** | Local offline GPU inference | `llama3.3:70b`, `qwen2.5-coder:32b`, `deepseek-r1:latest` | | **Custom** | Custom endpoints | Any OpenAI-compatible server (`vLLM`, `LocalAI`, `LM Studio`) | --- ## CLI Commands & Slash Shortcuts ### Terminal CLI Options ```bash antri # Launch interactive session antri desktop # Launch desktop control plane antri mobile # Launch standalone mobile PWA server antri --mode plan # Launch in Plan Mode (collaborative architecture) antri --mode vibe # Launch in Vibe Mode (direct coding flow) antri -alwaysallow # Launch with tool permission prompts bypassed antri update # Self-update to latest release without lockfile churn antri -g, --goal "" # Run autonomous multi-iteration refinement loop antri -d, --debate "" # Run Dialectic multi-persona self-debate antri --depth # Set debate depth antri -p, --prompt "" # Run one-shot query antri -m, --model # Specify model antri --provider # Specify provider ``` ### Interactive REPL Slash Commands | Command | Description | |---|---| | `/plan` | Switch to **Plan Mode** (collaborative architecture before coding) | | `/vibe` | Switch to **Vibe Mode** (direct fast coding flow) | | `/desktop` | Launch the **Desktop Control Plane** in standalone window | | `/imagine [code]` | Create **Visual Code Architecture Diagram & Graph** artifact | | `/mindmap [topic]` | Create **Interactive Hierarchical Visual Mind Map** artifact | | `/view [plan]` | Generate **Interactive HTML/JS Application & Plan** artifact | | `/artifacts` | List all generated interactive HTML, graph, and mind map artifacts | | `/mobile` | Launch the **Mobile App PWA server** | | `/alwaysallow` | Toggle permission prompts for sensitive tools | | `/goal [task]` | Run autonomous goal execution & refinement loop with live stages | | `/silent-goal [task]` | Run goal loop optimization **silently in background** with final plan | | `/loop [task]` | Alias for `/goal` loop | | `/debate [query]` | Launch Dialectic multi-persona self-debate with live argument streams | | `/silent-debate [q]` | Run Dialectic debate **silently in background** with final consensus | | `/depth ` | Set debate depth (`quick`, `deep`, `rigorous`) | | `/profile [name]` | Open interactive profile switcher or switch thinking profile | | `/notes` | View active profile notes & recorded style insights | | `/update` | Self-update ANTRI Code CLI | | `/connect` | Open interactive AI Provider selector | | `/models` | Search and select from available models | | `/tools` / `Ctrl+O` | Inspect recently executed workspace tools | | `/skills` | List built-in and dynamically synthesized custom skills | | `/meta` | View meta-optimization metrics and self-healing statistics | | `/memory` | View persistent memory status across all tiers | | `/consolidate` | Run reflection & knowledge compounding loop | | `/learn ` | Manually record a rule or persistent insight | | `/history` | View session conversation history | | `/export [file]` | Export conversation transcript to Markdown | | `/clear` | Start a clean new session | | `/help` | Show command reference manual | | `/exit` | Exit ANTRI Code | --- ## โš™๏ธ Google Cloud & Gemini Configuration ANTRI is built natively around **Google Gemini 3.7** and the **Google GenAI SDK (`@google/genai`)**, with backend deployment ready for **Google Cloud Run**. ### 1. Free Gemini API Key (Default Engine) Get a free API key from [Google AI Studio](https://aistudio.google.com/): ```bash # In the interactive CLI: antri > /key gemini AIzaSy... # Or set as an environment variable: export GEMINI_API_KEY=AIzaSy... ``` ### 2. Google Cloud Run Backend Deployment The ANTRI backend runs on **Google Cloud Run** (`*.run.app`) with built-in health probes (`/api/health`), Cloud Firestore state persistence, and zero-config browser access for judges: ```bash # Deploy ANTRI backend to Google Cloud Run in 1 command: gcloud run deploy antri-backend \ --source . \ --region us-central1 \ --allow-unauthenticated \ --set-env-vars GEMINI_API_KEY=AIzaSy... ``` --- ## ๐Ÿงช Reproducible Testing & Evaluation Guide (For Hackathon Judges) Follow these step-by-step instructions to reproduce and evaluate all core agentic features deterministically: ### 1. Run the Full Automated Test Suite (100% Offline & Isolated) ANTRI includes **47 comprehensive automated unit tests** covering configuration, vector memory, goal loops, dialectic pipelines, file manipulation tools, artifact parsers, mobile servers, and echo interceptors: ```bash npm test ``` *Expected Result: `47 passing tests (0 failures)` completing in ~2โ€“4 seconds.* --- ### 2. Test Autonomous Dialectic Debate (`/debate`) Observe multi-persona background debate (Thesis $\rightarrow$ Antithesis $\rightarrow$ Empirical Verification $\rightarrow$ Consensus): ```bash # In the interactive CLI: antri > /debate "Should we use PostgreSQL or Redis for high-throughput session caching?" ``` *Expected Result: Live streaming personas clash and synthesize into a final consensus.* --- ### 3. Test Claude-Style Live Interactive Multi-Page SPA Artifacts (`/view`) Test on-the-fly generation of rich, runnable web applications rendered natively with zero raw code exposure: ```bash > /view "7-day football conditioning and stretching routine with interactive timers and progress checklist" ``` *Expected Result: Renders an interactive **"๐Ÿ‘๏ธ View Artifact"** card. Opening it launches a multi-tab web application with working countdown stopwatch, dynamic progress percentage checklists, and bottom stepper navigation.* --- ### 4. Test Reference-Style Collapsible Mind Maps (`/mindmap`) Test the Markmap hierarchical tree visualization engine: ```bash > /mindmap "Distributed Systems Architecture and Consensus Protocols" ``` *Expected Result: Renders a collapsible horizontal mind map tree. Click the `<` / `>` circular handles to fold/unfold subtrees, pan/drag the canvas, and toggle light/poster-dark themes.* --- ### 5. Test Persistent Thinking Profiles & Silent Note Capture (`/profile`, `/notes`) Test lifelong cognitive memory and persona extraction: ```bash > /profile # Navigate and switch between thinking profiles with arrow keys > /notes # View automatically extracted coding conventions and philosophy rules ``` --- ### 6. Test Native Flutter Mobile Client (`antri_flutter`) Run the cross-platform mobile client with floating Prompt Toolkit and Gemini-style navigation drawer: ```bash cd antri_flutter flutter test # Run mobile widget tests flutter run # Launch mobile client on device/emulator ``` --- ## Local Development & Setup ```bash # 1. Clone the repository git clone https://github.com/ashu90-prog/antri_cli.git cd antri_cli # 2. Install dependencies npm install # 3. Build TypeScript & assets npm run build # 4. Run automated test suite (47 unit tests) npm test # 5. Run in dev mode npm run dev ``` --- ## License Distributed under the **MIT License**. See `LICENSE` for more information.