Sylva 🌲

Sylva is an automated AI tool to auto-generate comprehensive AGENTS.md and CODEBASE_CONVENTIONS.md documentation by reading your entire repository recursively using Ax-LLM.

It extracts the folder structures, code conventions, state management patterns, and specific architecture styles embedded in your code directly into a format that future AI Coding Assistants (or human engineers) can perfectly follow.

📖 Full Documentation

• Support for OpenAI, Google Gemini, Anthropic, and Azure LLMs.
• Direct --github-repository fetching using Git Clone.
• Direct --local-repository analysis.
• Multi-LLM model configuration with automatic mini-model fallback for Codebase Analysis Iterations.
• Strict Type-Safety achieved with Ax-LLM's internal framework tools.

Install dependencies:

npm install

Configure your environment variables:

cp .env-example .env
# Edit .env and supply your API Keys (GEMINI_API_KEY, ANTHROPIC_API_KEY, OPENAI_API_KEY, etc)

Build the project:

npm run build

The project is published to the NPM registry and can be executed natively anywhere!

Option 1: Run seamlessly without installing

# Analyze the current working directory natively
npx @thelogicatelier/sylva

# Provide a specific GitHub repo
npx @thelogicatelier/sylva --github-repository https://github.com/expressjs/express

Option 2: Install globally on your system

npm install -g @thelogicatelier/sylva

# Once installed globally, you can execute it anywhere!
sylva
sylva --help

Alternatively, if running from source:

npm start -- --github-repository https://github.com/expressjs/express -m openai/gpt-4o

List Supported Models:

npx @thelogicatelier/sylva --list-models

The iteration count (-i) and model choice (-m) dramatically affect output accuracy. Here's a battle-tested guide:

Real-world example: Analyzing a React/Tailwind frontend + Python/FastAPI backend + Wix API monorepo:

npx @thelogicatelier/sylva --local-repository . -m openai/gpt-5.2 -i 25

For detailed guidance, see the Choosing the Right Model and Iteration Depth Guide docs.

Sylva now includes deterministic framework detection and source code integration scanning that scans the entire repository (including nested subprojects and monorepos) before invoking the LLM. This prevents framework hallucination and produces more accurate AGENTS.md output.

What it detects via Manifests:

• OpenClaw (openclaw.json) — treated as the primary orchestrator
• Node.js/JS/TS — React, Angular, Vue, Next.js, Express, NestJS, etc. from package.json
• Python — Django, Flask, FastAPI from requirements.txt, pyproject.toml, Pipfile
• Java/JVM — Spring Boot, Quarkus from pom.xml, build.gradle
• .NET — ASP.NET Core from *.csproj, global.json
• Go — Gin, Echo, Fiber from go.mod
• Rust — Actix, Axum, Tokio from Cargo.toml

What it detects via Source Scanning (Integrations & Deployments):

• Deployment Platforms — Fly.io, Railway, Render, AWS, GCP, Azure, DigitalOcean via fly.toml, railway.json, app.yaml, or even Dockerfile base image inspection.
• External APIs — Stripe, Wix, AWS, Instagram Graph API via raw API URL regex matching or SDK imports directly in the source code.
• Strict Security — It automatically ignores .env files entirely to prevent secret leakage and strictly respects the .gitignore.

Version certainty: Versions are only reported when explicitly found in manifest/lockfiles. Never assumed.

Web grounding: When BRAVE_API_KEY is set, Sylva fetches official docs for detected frameworks (version-specific when exact versions are known, latest fallback otherwise) to feed into the architecture constraints. Note: It intelligently rate-limits itself to respect the Brave Free Tier (1 req/s).

Debug output: awareness.json and grounding.json are saved alongside AGENTS.md for full transparency.

• AUTOSKILL_MODEL: Set this to gemini or anthropic or openai to change the default execution provider globally without providing -m on every execution.
• GITHUB_REPO_URL: Hardcode a repository for the CLI parser to utilize if no explicit configuration flags are provided.
• BRAVE_API_KEY: Set this to enable web-grounded framework documentation retrieval via Brave Search. If not set, Sylva will still detect frameworks deterministically but will skip web reference gathering.

Unit tests are located at tests/ which mirror the /src folder structure separating logical test groupings. Execute tests by running:

npm test

We strictly follow a feature-branch workflow. Please ensure your contributions maintain high code quality and test coverage.

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
3. Commit your Changes (git commit -m 'Add some AmazingFeature') - Our Husky pre-commit hooks will automatically lint and test your code.
4. Push to the Branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

Please report all issues and feature requests using our GitHub Issues tracker. When reporting bugs, please provide a minimally reproducible example along with the console output (if applicable).

Built by The Logic Atelier