AgentGem

Pack your agent into a Gem.

Start from a project you've built with Claude, Codex, or Hermes — its skills, MCP servers, hooks, and instructions, layered with your user settings. Iterate until it's right, then crystallize it into a secret-safe, composable Gem to publish and deploy. Built on AgentBack: one Zod contract, exposed as a REST endpoint and an MCP tool.

Get started → How it works

The Gem workflow

Project in. Reusable Gem out.

Discover Find the projects you've built with your agent, from session history
Compose Project skills, MCP, hooks & instructions, layered with your user settings
Iterate Run tasks & workflows in the project to refine the setup
Crystallize Capture it as a secret-safe, reusable Gem — secrets redacted
Publish / deploy To the registry, or to Eve, Flue, OpenAI, AgentCore & Claude

Why AgentGem

One contract, every boundary.

One Zod contract

Every operation is defined once. AgentBack exposes it as a REST endpoint, an MCP tool, and an OpenAPI 3.1 doc — with runtime validation for free.

RESTMCPOpenAPI

Redacted at capture

MCP secrets are stripped the moment config is read — by value and by key name. No response, preview, or Gem ever carries a raw secret.

trust boundary

Composable registry

Publish Gems to a GitHub-backed registry, then resolve, merge, and install them — assembling bigger agents out of smaller, shareable pieces.

publishmergeinstall

Deploy anywhere

The same neutral Gem feeds multiple code-gen targets and managed backends — pick a target, publish, and undeploy when you're done.

EveFlueOpenAIAgentCoreClaude Agents

What you get

A Gem, with secrets sealed.

gem.json
// built from your agent config — secrets never leave your machine
{
  "name": "research-assistant",
  "skills": ["deep-research", "make-pdf"],
  "mcpServers": {
    "firecrawl": {
      "command": "npx",
      "env": { "FIRECRAWL_API_KEY": "<redacted>" }
    }
  },
  "claudeMd": "included"
}