Ship agents faster with verified components.
Discover, evaluate, and reuse tools and micro-agents — complete with trust signals, versioning, and clear permissions.
$ es install @acme/slack-triage@2.4.1 ✓ signature verified (acme.eng) ✓ evals passed (94/96) ✓ scans clean (no critical findings) ✓ permissions slack:read, slack:write installed @acme/slack-triage → ./agents/tools/
A registry built for agent teams.
Reusable components with the guarantees production systems require.
- 1Registry of reusable components — tools, micro-agents, routers, and eval packs.
- 2Standard packaging with clear, typed interfaces for inputs and outputs.
- 3Trust signals: benchmarks, security scans, and provenance.
- 4Versioning with human-readable changelogs.
- 5Safe deployment patterns and permission scopes baked in.
Publish. Evaluate. Deploy.
Publish
Package a component with metadata, typed I/O, required permissions, and an eval pack.
Discover & evaluate
Browse the registry. Compare benchmarks, scan results, and publisher signatures.
Install & deploy
Wire it into your agent with predictable interfaces. Sandbox by default.
Why trust matters.
Ambiguity kills agent projects. Every component ships with the answers up front.
Know what it can access
Every component declares required permissions and scopes upfront.
Know it works
Reproducible eval runs and benchmarks travel with each version.
Know who shipped it
Publisher identity and signed releases prove provenance.
Know what changed
Semantic versioning and changelogs for every release.
Components teams actually reuse.
A taste of the kinds of building blocks that live in the registry.
Notion / doc search
Wrapped tool with citation-safe results.
Slack triage agent
Micro-agent that routes and drafts responses.
PII scrubber
Redact sensitive data before model calls.
Schema validator
Enforce JSON schema on model output.
Citation gate
Policy that blocks unsourced claims.
Model router
Route requests by cost, latency, or quality.
Common questions.
Is this a framework?+
No. It’s a registry and a set of conventions. Bring your own framework — LangGraph, CrewAI, custom — and pull in the components you need.
What counts as a component?+
Anything reusable with a typed interface: a tool, a micro-agent, a router, a policy, an evaluator, or a guardrail.
How do trust signals work?+
Each release ships with an eval pack, security scan, and signed provenance. You see them before you install.
Can we run a private registry?+
Yes. Teams can publish internal components to a private registry with the same interfaces and trust signals.
Does this replace our stack?+
No. It sits alongside your agent framework, your models, and your infra. It replaces ad-hoc copy-pasted tools.
How do permissions work?+
Components declare scopes (network, filesystem, secrets). Hosts enforce them at runtime through sandbox patterns.
What languages are supported?+
Components are language-agnostic through their interface contracts. Reference SDKs ship for TypeScript and Python.
How much does it cost?+
Pricing hasn’t been announced. Request access below to be first in line.
Get early access.
Tell us a bit about what you're building. We'll be in touch.