Registry
The Gem registry distributes composable Gems over the same archive format AgentGem builds. It's GitHub-backed: Gems live in a repository, so distribution rides on infrastructure you already trust — no separate service to run.
The operations
- Publish — push a built Gem to the registry under a name and version. Because the Gem is already secret-safe, publishing never leaks credentials.
- Resolve — look up a Gem reference and fetch its archive (manifest + lock).
- Merge — combine multiple Gems into one. Manifests are reconciled and a single lock is re-resolved, so the result is a coherent, reproducible Gem rather than a pile of overlapping config.
- Install — pull a Gem back down into a local testbed so you can run or extend it.
Why composition matters
The manifest/lock split (see Concepts) exists precisely so Gems can be assembled. A small, focused Gem — say, a research skill set plus one MCP server — can be merged with another to build a larger agent, without hand-editing config or re-introspecting every machine. Publish the pieces once; compose them many times.
A typical flow
- Build a focused Gem and publish it to the registry.
- On another machine (or in CI), resolve the Gems you need.
- Merge them into the agent you want.
- Install the merged Gem into a testbed, then run it or send it to a target.
Back to the documentation index.