Sourcebee
GitHub collaboration app for engineering organizations - org-wide visibility for ICs and leaders, with AI analysis built in
Work together...better. Sourcebee is a GitHub collaboration app for engineering orgs: connect a GitHub org and get real-time, org-wide visibility into pull requests, CI/CD pipelines, and releases, plus AI-generated analysis of how the org is performing, in plain English. Built for both day-to-day work (ICs) and the bigger picture (leaders), from the same underlying data. Currently in beta.
- Real-time pull request tracking across the whole org - flag priorities, track review status, catch what's going stale before it blocks a release
- Team insights and benchmarks - org metrics measured against industry benchmarks, with goals you can set and track over time
- CI/CD visibility - full check-run history and pipeline health without digging through GitHub Actions logs
- Release view - what shipped, when, and which PRs went into each deploy
- Zero-setup GitHub App install - connects to an existing org in minutes with no config files or new workflows required
- AI-generated org analysis included out of the box, no credits or per-query charges, plus an MCP server add-on for wiring org data into Claude, Cursor, or any MCP-compatible agent
- Reads only repo/PR/commit metadata via the GitHub App - no write access to code, no file contents ever touched, and all analysis is team/repo-level rather than tracking individual developers
- Self-distributed browser extension for Chromium and Gecko browsers, surfacing Sourcebee context right inside GitHub's UI
Multi-service platform split across four repos: buzzcatcher (Go API backend) on DigitalOcean App Platform with a managed Postgres cluster, deployed by pushing a Docker image to GHCR and explicitly triggering a new deployment rather than relying on git-integration auto-deploy; sourcebee-app, housebee, and pollinator each as their own direct-upload Cloudflare Pages project (wrangler pages deploy from each app's own CI), kept separate specifically because a direct-upload deploy replaces a project's entire file tree and would otherwise clobber a shared project. The three frontends are Svelte, designed by me and implemented with AI coding agents.
All shared/rarely-changing infrastructure - DNS, DO app specs, Pages project shells, Cloudflare zone settings - lives in one central repo (wagglebee) managed with OpenTofu, with dev and prod as fully separate applies: separate state files (in R2), separate variables, separate CI environments, so a dev-scoped CI job has zero access to prod secrets and vice versa. Secrets are managed through Doppler and loaded into each environment's GitHub Actions deploys.
Third-party integrations wired through the backend: GitHub App (webhooks + OSS token) for repo/PR/commit data, Clerk for auth, Paddle for billing (sandbox in dev), and the Anthropic API for the AI analysis and MCP features - each surfaced as scoped secrets per environment rather than shared across dev/prod.
Per-hostname Cloudflare Web Analytics (manually embedded snippet per app, not zone-wide auto-install) so each app's traffic is measured independently instead of blending into one zone total.