DocBot

The technical GPS for any codebase.

---

Documentation is almost always outdated or non-existent. DocBot solves this by autonomously mapping your system architecture in seconds, not days. It parallelizes codebase exploration using specialized agents to extract deep semantic understanding—from high-level components to individual symbols.

Features

• Interactive Architecture Maps: Dynamically renders a whiteboard-style graph of your entire system that you can actually chat with.
• Git-Integrated Workflow: Built-in git hooks (post-commit/merge) and CLI tools (docbot status, docbot diff) to keep docs in perfect sync with your code.
• Autonomous Explorer Agents: Uses Tree-sitter and LLMs to investigate symbols, trace dependencies, and uncover the "heart" of your logic while you code.
• One-Click Generation: Zero-config needed. Run docbot init on any repository and get a fully functional documentation site instantly.
• Live Pipeline Visualization: Includes a built-in telemetry server to watch the agentic exploration process in real-time.

Quick Start

1. Install

   git clone <repo-url> && cd docbot
   uv pip install -e .

2. Configure Create a .env file with your BACKBOARD_API_KEY.

3. Initialize & Generate

   docbot init
   docbot generate --viz

4. Explore

   docbot serve

Commands

Command	Description
docbot init	Initialize .docbot/ directory and config.
docbot generate	Run the full documentation pipeline.
docbot update	Incremental update for changed files.
docbot status	Show documentation lag and affected scopes.
docbot diff	Compare two architecture snapshots.
docbot serve	Launch the interactive explorer UI.
docbot hook install	Install git hooks for auto-documentation.

Options

Flags for docbot generate:

Flag	Description
--no-llm	Skip LLM enrichment (extraction only).
--model / -m	Model ID (default: openai/gpt-4o-mini).
--concurrency / -j	Parallel explorer workers (default: 4).
--timeout / -t	Per-scope timeout in seconds (default: 120).
--max-scopes	Maximum number of documentation scopes (default: 20).
--agent-depth	Max recursion depth for agents (default: 4).
--serve	Start live webapp during generation for visualization.
-p / --port	Webapp port when using --serve (default: 8000).

Notes:

• Agent exploration is default-on in current pipeline paths.
• MIMO (xiaomi/mimo-v2-flash) uses a bounded delegation strategy tuned for stability and broader deep-search coverage.

How It Works

DocBot uses a high-concurrency Map-Reduce pipeline:

1. Scan: Multi-language discovery and entrypoint mapping.
2. Plan: Dependency-aware grouping into logical "scopes".
3. Explore: Parallel AST extraction (Tree-sitter) and LLM-agent reasoning.
4. Reduce: Cross-scope graph construction and anomaly detection.
5. Render: Generation of Markdown, Mermaid diagrams, and Search Index.

Architecture

Project Structure

• src/docbot/pipeline/: Core Map-Reduce logic (Scanner, Planner, Explorer, etc.)
• src/docbot/git/: Git-aware logic (Snapshotting, Diffing, Hooks)
• src/docbot/viz/: Pipeline telemetry and replay server
• src/docbot/extractors/: Language-specific AST analysis
• webapp/: React + Vite + ReactFlow + Tailwind frontend

Stack

• Backend: Python, FastAPI, Typer, Pydantic, Tree-sitter
• Frontend: React, Vite, React Flow (interactive graphs), Tailwind