Bluebag - Run AI agent skills in production
by•
Bluebag is a production skills runtime for AI agents. Use Anthropic-style agent skills in your own app without custom infra. Load public skills, sync with a CLI, and run them safely across any LLM using Vercel AI SDK or LangChain.
Replies
When Anthropic published their Skills system, it immediately clicked for me.
The idea is simple but powerful: give agents a filesystem-based skill library (instructions, scripts, reference files) and let them progressively load only what they need.
The problem? In my own projects, I wasn’t using Claude. Most workloads ran on Gemini for cost and flexibility, which meant I couldn’t use Anthropic’s Skills directly, even though I really wanted that architecture for remote agents built with Vercel’s AI SDK.
So we built a model-agnostic, Agent Skills runtime that works with any remote agent.
Bluebag handles the hard parts of running skills at scale: sandbox lifecycle, skill reusability, dependency installation, concurrent file access, and secure execution.
Right now, you can:
- Package instructions, metadata, scripts, and references into a Skill directory
- Run Python or JS inside skills (with automatic dependency installs)
- Let agents create, read, and download files via a files API
- Manage everything with a CLI (push/pull), a TypeScript SDK, and a lightweight web app
Playground: https://www.bluebag.ai/playground
Would love to see what people build with it. The free tier is intentionally generous 🙂