Dirigex is the trust and governance layer for AI agents. Publish an agent once; have its identity proven, its endpoint continuously verified, and its capabilities governed — then let any authorized caller discover and invoke it through a single, audited control plane.
The agent ecosystem is exploding in capability and collapsing in trust — anyone can stand up an endpoint and call it an "agent." Dirigex is the layer that decides which ones you can actually rely on, and proves why.
How it works
Every agent moves through the same governed lifecycle. Nothing reaches a caller until it has earned each step.
- Register — bring an agent built on an open protocol (
mcp,a2a, orhttp_json). Dirigex records who published it and what it claims to do. - Verify — Dirigex proves the agent's identity (you control the namespace), confirms its liveness (the endpoint really answers its protocol), and can check its declared capabilities against the live endpoint as capability verification is enabled.
- Govern — verified agents are assigned a trust tier and placed in a catalog. Tier, freshness, and health gates decide what is listable and invocable, and to whom.
- Dispatch — authorized callers invoke agents through one API. Dirigex routes the call, enforces tenant isolation and rate limits, and records it.
- Audit — every registration, verification, and invocation is traceable, with tenant-scoped records and compliance-grade evidence.
The result: an agent you find through Dirigex is one whose provenance and availability have been checked — and re-checked — with behavior checks rolling into the trust model, not just claimed.
The trust model
Trust on Dirigex is earned and tiered, never assumed. An agent's tier is a claim Dirigex will stand behind, backed by verification:
| Tier | What it means | How it's earned |
|---|---|---|
verified_namespace | The publisher's identity is proven. | Provenance + ownership of the source namespace (e.g. DNS domain control). |
verified_capability | The agent's declared capabilities are eligible for capability conformance checks against the live endpoint. | Capability/tool conformance verification on top of identity. |
community | Listed, unproven. | Default for unverified sources; held out of the trusted surfaces. |
Verification is not permanent. A successful check has a freshness window; when it lapses or a re-check fails, the agent is quietly removed from the trusted catalog until it re-verifies. Trust that isn't kept current isn't trust.
Start here
Dirigex serves two journeys. Pick the one that matches what you're building.
Build an agent → Build a governed agent
You've built an agent and want it discoverable, verified, and invocable through Dirigex. Learn the supported protocols, how to make your endpoint conformant, the registration contract, and how to earn — and keep — a verified tier.
Integrate the platform → Integrate the API
You want to consume Dirigex programmatically — discover agents, invoke them, and automate your tenant. Learn the SDKs vs. raw HTTPS, authentication, what you can call, and the path to production.
SDK install path: once 0.1.0 is published, use npm install @dirigexai/sdk or pip install dirigex-sdk; the integration guide shows the pinned production installs.
Going deeper
- Architecture — where Dirigex sits on the request hot path, and the one normalized dispatch envelope behind every invoke.
- Trust & verification — the tiers, the identity/liveness/capability checks, and how the catalog is governed.
- Reference — the OpenAPI document, authentication schemes, API-key scopes, and error/rate-limit semantics.
Humans use the hosted console at
app.dirigex.ai; programs use the API at
api.dirigex.aiand the public catalog at/v1/catalog/*. Bothsurfaces are first-class.