GitHub (Repo to Canvas)

Output quality varies by repo size and structure. Feedback welcome at ask@collapseindex.org.

The GitHub tab (H key, then click GitHub) lets you connect a public or private GitHub repository and have Chinilla AI generate an architecture diagram by reading the repo’s file tree and config files.

How it works

1. Open the AI panel with H
2. Click the GitHub tab
3. Paste a repo URL (e.g. https://github.com/owner/repo) or pick from your linked repos
4. Click Analyze Repository (10 credits)

Chinilla AI fetches the repo’s file tree, README, and configuration files (docker-compose.yml, package.json, pyproject.toml, etc.) to identify services, components, and connections. Source code is not stored. A diagram appears on the canvas when analysis completes.

What gets analyzed

• File tree structure (directories, file types, language detection)
• README.md (truncated to 10,000 chars)
• Manifest files: package.json, pyproject.toml, Cargo.toml, go.mod, requirements.txt
• Container configs: Dockerfile, docker-compose.yml, compose.yaml
• CI/CD configs: .github/workflows/*.yml

Source files are never sent to the LLM or stored on our servers.

Linking your GitHub account

To analyze private repos or pick from your repo list, link your GitHub account:

1. Open the GitHub tab
2. Click Pick from your repos
3. Authorize Chinilla via GitHub OAuth

You can analyze any public repo by URL without linking an account.

Re-analyzing and auto-check

Once a repo is connected, the panel shows:

• Check for updates (free) — polls GitHub for new commits since the last analyze
• Re-analyze (10 credits) — runs a fresh analysis and updates the diagram
• Auto-check toggle — automatically checks for new commits when you open the project (throttled to once per 5 minutes per project, never costs credits)

When Check for updates finds new commits, the panel shows a “New changes detected” banner with a one-click Sync now button.

Caching

Repeat analyses of the same commit return instantly from cache and cost 0 credits. The cache is keyed by the repo’s tree SHA, so any new commit busts the cache and triggers a fresh analysis.

Privacy: The cache is shared across users for public repos only (so two people analyzing the same open-source commit both pay 0 credits the second time). Analyses of private repos are never cached — every re-analyze re-runs the AI pass and costs the full 10 credits, but your private code structure is never exposed to other users via the cache.

Publishing

Once analyzed, you can publish your diagram as a public link or embed it in your README:

1. In the GitHub panel, toggle Publish diagram on
2. Copy the public URL, Markdown embed, or iframe embed

The Markdown embed includes a live PNG of your architecture, perfect for repo READMEs.

Limits

• Public repos and connected private repos
• Max ~150 files analyzed per pass (large repos use a curator pre-pass to select the most relevant services)
• 10 credits per analyze
• Pro plan required