Documentor

An AI documentation suite. It ingests a codebase, parses it semantically, generates accurate documentation using LLMs, and serves it via CLI, Web UI, and CI/CD pipelines.

Installation

Recommended Method (pipx)

Because Documentor is a standalone CLI tool, the best way to install it on modern Linux/macOS systems (which enforce PEP 668) is using pipx. This installs Documentor in an isolated environment while exposing the CLI globally.

# If you don't have pipx installed: python3 -m pip install --user pipx
pipx install documentor-ai

Alternative Methods

If you are inside an active virtual environment (like .venv/), you can use standard pip:

pip install documentor-ai

(Note: If you attempt this globally on newer Linux distributions without pipx, you will get an externally-managed-environment error. You can bypass this by appending --break-system-packages, though pipx is heavily preferred).

Upgrading an Existing Installation

To get the latest version (including new features like Interactive Chat), run the upgrade command corresponding to how you installed it:

# If you used pipx
pipx upgrade documentor-ai

# If you used standard pip
pip install --upgrade documentor-ai

BYO-LLM (Bring Your Own LLM)

Documentor uses LiteLLM under the hood, allowing you to use your preferred model (OpenAI, Anthropic, Gemini, DeepSeek, local models via Ollama, etc.).

Run the interactive setup command to configure your default model and API keys:

documentor configure

(You will be prompted for MODEL_NAME, GEMINI_API_KEY, etc. You can just press Enter to leave them empty or fall back to the default gemini/gemini-3.6-flash)

Usage

Commands Overview

Here is a quick summary of all the available commands:

• documentor --version: Show the application's version and exit.
• documentor configure: Configure API keys for the LLM providers.
• documentor generate <path>: Triggers the Core Engine to parse the codebase and generate documentation for the given path. Options include --model, --regenerate, --resume, and --file.
• documentor chat [question]: Ask questions against the codebase using Retrieval-Augmented Generation (RAG). Start without a question for an interactive session.
• documentor serve: Spin up the beautiful Web UI locally (defaults to port 8000).

1. Command Line Interface (CLI)

To generate documentation for a repository, run:

documentor generate /path/to/your/repo

Note on Parsing: Documentor automatically ignores node_modules, dist, .env files, logs, and all .git ignored files by default. If you want to force Documentor to ignore specific files or folders, just create a .docignore file in the root of your project!

More Examples: Generate for the current directory (.) using your configured model:

documentor generate .

Single/Multiple Specific Files: If you only want to generate or regenerate documentation for specific files (for instance, if you only updated a few files in a large project), you can specify them using the --file or -f flag. This will skip ARCHITECTURE.md and QUICKSTART.md and only document the targeted files:

documentor generate . -f src/main.py -f src/utils.py

Generate using a specific model (by overriding your default model):

documentor generate . --model gemini/gemini-3.6-flash

2. Intelligent AI Assistant (Terminal)

Once the repository is indexed, you can converse with an intelligent AI assistant about your codebase directly from your terminal.

Interactive Chat Mode (Recommended): If you run chat without providing a specific question, it will drop you into an interactive terminal where you can chat continuously!

documentor chat --path .

Single Question Mode: If you just want a quick answer, you can provide the question directly:

documentor chat "How does the authentication system work?" --path .

Note on Intelligence: The AI assistant is context-aware and highly intelligent! While it won't directly edit your files, you can ask it to explain code in simple terms, provide setup guides for beginners, or suggest code improvements.

3. Web UI (Playground & Visual Chat)

Prefer a visual interface? Spin up the beautifully designed, glassmorphic Web UI:

documentor serve --port 8000

Then open http://localhost:8000 in your browser. The Web UI not only loads the documentation for your current directory (.), but it also features a Full-screen Conversational AI Chat!

https://github.com/user-attachments/assets/470ee45a-a252-4eb3-8c10-ac33e60230bb

The Web UI chat includes:

• Conversational Memory: The AI remembers the context of your ongoing chat.
• Mermaid Diagram Support: Ask the AI to draw architectural diagrams, and it will render them as beautiful SVG graphics right in the chat!
• Clean Layout: Features a light/dark theme toggle, a collapsible sidebar, and a distraction-free fullscreen chat mode.

3. GitHub Action (CI/CD)

Documentor comes packaged as a Docker Action. You can automate documentation generation on your Pull Requests by creating .github/workflows/documentor.yml in your target repository:

name: Generate AI Docs
on:
  pull_request:
    branches: [ main ]
jobs:
  docs:
    runs-on: ubuntu-latest
    permissions:
      contents: write
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0
      - uses: nirajmatere/documentor@main
        env:
          # Define your model and map secrets appropriately
          MODEL_NAME: 'gemini/gemini-3.6-flash'
          GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
      - run: |
          git config --global user.name 'github-actions[bot]'
          git config --global user.email 'github-actions[bot]@users.noreply.github.com'
          git add documentor_docs/
          git diff --quiet && git diff --staged --quiet || git commit -m "docs: Auto-update AI documentation"
          git push

FAQ

What models does Documentor support?
Documentor uses LiteLLM, so it supports virtually any major model: OpenAI, Anthropic, Gemini, DeepSeek, and even local models via Ollama.

What programming languages are supported?
We currently parse code using Tree-sitter, which allows us to natively understand Python, JavaScript, TypeScript, Go, Java, Rust, C, C++, and C#.

How do I skip certain files or directories from being documented?
By default, Documentor automatically ignores .git, node_modules, dist, .env files, logs, and git-ignored files. If you need to ignore specific custom directories, create a .docignore file in your repository root.

Can I only regenerate docs for specific files?
Yes, you can use the -f flag to target specific files (e.g., documentor generate . -f src/utils.py).

Is my codebase safe?
Your code is only sent to the LLM you configure in documentor configure. If you choose to configure a local LLM through Ollama, your code never leaves your machine.