GitHub Repository Analysis

You've got a repo. You want an architecture diagram. The old way: spend an hour reading code, figuring out the structure, manually drawing boxes and arrows.

The Codelit way: paste the GitHub URL. Get a diagram.

The GitHub feature analyzes your repository's structure, detects your tech stack, and generates an interactive architecture diagram that reflects what you've actually built, not what you think you built.

When you paste a GitHub URL, Codelit's parser examines key files in the repository:

Package manifests: package.json, requirements.txt, go.mod, Cargo.toml, pom.xml, Gemfile. These reveal frameworks (Next.js, Django, FastAPI, Spring Boot), databases (pg, mongoose, prisma), queues (bull, celery), and cache layers (ioredis, redis-py).

Infrastructure files: Dockerfile, docker-compose.yml, terraform/*.tf, .github/workflows/*.yml, kubernetes/*.yaml. These show how services are deployed, what infrastructure exists, and how CI/CD is configured.

Configuration files: .env.example, config/, environment variable references. These hint at external services. Stripe keys mean a payment integration, AWS credentials mean cloud services, SendGrid keys mean email service.

Directory structure: monorepo patterns (packages/, apps/, services/) reveal multi-service architectures. A frontend/ and backend/ split implies a client-server architecture.

The parser doesn't read your business logic. It reads your dependency declarations and infrastructure definitions, the parts that define your architecture.

The detection is organized into categories:

Databases: PostgreSQL, MongoDB, Redis, DynamoDB, and the rest. Detected from ORM configs, connection strings in .env.example, and database drivers in package manifests.

Message Queues: Kafka, RabbitMQ, SQS, Bull/BullMQ, Celery. Detected from queue client libraries and worker configurations.

Cloud Services: AWS (S3, Lambda, SQS, SNS, CloudFront), GCP (Cloud Run, Pub/Sub, Cloud Storage), Azure (App Service, Cosmos DB). Detected from SDK imports, Terraform providers, and infrastructure configs.

Frontend Frameworks: React, Next.js, Vue, Nuxt, Angular, Svelte, SvelteKit. Detected from framework dependencies and config files.

Backend Frameworks: Express, Fastify, Django, FastAPI, Spring Boot, Gin, Actix. Detected from framework imports and route definitions.

Third-Party Services: Stripe, Auth0, Firebase, SendGrid, Twilio, Sentry, Datadog. Detected from SDK dependencies and API key references.

For public repos, Codelit uses the GitHub API directly. No authentication needed.

For private repos, you connect your GitHub account through OAuth. Codelit requests read-only access to repository contents, nothing more. No write access, no access to other repos you haven't explicitly authorized.

The OAuth flow takes about 10 seconds. After that, paste any of your private repo URLs and the analysis works the same way.

Once the tech stack is detected, Codelit feeds the information into the AI generation pipeline. But instead of generating from a vague description, it's working from concrete facts: you're running Next.js with Prisma and PostgreSQL, deployed on Vercel with a Redis cache and Stripe integration.

The result is a diagram that actually matches your codebase. It's not a generic "web app architecture." It's your architecture, with your specific technologies, your specific service boundaries, and your specific data flows.

The parser is good, but it's not telepathic. Things it can miss:

• Runtime dependencies that aren't declared in package manifests (rare, but possible)
• Custom infrastructure that doesn't use standard tools (if you're deploying with shell scripts, the parser won't catch that)
• Internal service communication patterns that aren't obvious from config files

That's why the diagram is editable. The GitHub analysis gives you a strong starting point, usually 80-90% accurate, and you refine from there.

Head to codelit.io, click the GitHub import option, and paste a URL. Public repos work immediately. Private repos need a quick OAuth connection.

More details on the GitHub feature page.