We document and share
everything we build.
At Bubble, build-in-public is not a slogan. The tools we develop, we publish. Every repo below reflects a real need we encountered in production. MIT license, contributions welcome.
The repos we maintain
Four projects published under MIT. Each one addresses a concrete need we encountered while building our agent system.
Encrypted secrets manager (SOPS) for multi-agent projects. Stores and distributes environment credentials without exposing plaintext values in configs or logs.
Supervision framework for orchestrating multiple Claude Code agents in parallel. Handles task dependencies, automatic retries, and status reports via Telegram.
Configuration and deployment scripts for a VPS infrastructure hosting autonomous agents. Includes systemd configs, monitoring, automated backups, and update scripts.
Reusable skills library for Claude Code: web search, Telegram messaging, Google Workspace access, audio transcription, and more. One command to install in any agent project.
Why we publish in open source
Transparency
We don't claim to master what we hide. Publishing the code means accepting to be read, critiqued, improved. It's also the best proof that the architecture actually holds up.
Sharing know-how
The tooling ecosystem around AI agents is still very young. Every pattern we document saves time for others building the same things on their end.
Accountability
Publishing under MIT is a commitment: the code must be readable, maintainable, documented. It forces us to do things properly, not just make them work for ourselves.
A question about a repo, or want to contribute?
Open an issue directly on GitHub, or reach out to talk it over.
Want to deploy this kind of stack in your company?
We implement AI agents for professionals. Let's talk about your needs.