Can I register an agent before it is built?

Yes. The record is independent of instrumentation, so registering during design is reasonable — it just reports no activity until the agent runs.

My framework is not listed.

Check whether it emits OpenTelemetry, which is the usual route for anything without a first-class entry. If not, raise it with your account team.

What happens if I register an agent Discovery later finds?

You get two records for one agent. Remove the manual one.

Can I change the platform after registering?

Platform is set during registration. Register again against the right framework and remove the first record.

Introduction

Discovery covers hosted platforms. It cannot cover an agent you built yourself on a framework, because there is no tenant to sweep — nothing to call, and nothing to enumerate.

Manual registration is the answer for those. You create the record, then instrument the agent so it reports its runs back to UnifyApps.

Where to find it

AI Governance > Agent Registry > + Register Agent

Access to AI Governance is granted through platform roles. Ask your workspace administrator, or see Settings > Governance > Roles for how permissions are assigned.

When to register manually

The agent is built on a framework — LangChain, LangGraph, the Anthropic Agent SDK, the OpenAI Agent SDK, LiveKit. These have no inventory API, so they are registration-only.
The platform is supported but the account is not connected — Connecting it is usually better — one connection covers every agent in that account, and keeps covering new ones.
The platform is not supported at all — Register the agent against the framework it is built on.

Do not manually register an agent Discovery already found. You will get two rows for one agent, and two sets of incidents.

Before you begin

Know which framework the agent runs on — It determines how you instrument it, and it is the first thing the form asks.
Have a real description ready — Description is required, and it is what a reviewer reads when an incident lands. "What does this agent do" is the literal prompt.
Decide the owner — The form defaults to you. That is right for an agent you built and wrong for one you are registering on someone else's behalf.

Register the agent

Step 1: Open the registration form

Go to the Agent Registry and click + Register Agent. A panel opens titled Enter manually, showing Step 1 of 2 · Select a platform.

Step 2: Select the platform or framework

The list is searchable and covers both hosted platforms and frameworks. The framework entries — Anthropic Agent SDK, LangChain, LangGraph, LiveKit, OpenAI Agent SDK — appear here and not in Discovery, for the reason above.

Step 3: Fill in the agent details

The panel becomes Step 2 of 2 · Agent details.

Field	Required	Notes
`Name`	Yes	What this agent is called. Name it for what it does, not for the project that built it.
`Description`	Yes	What the agent does. Written for whoever investigates an incident on it months from now.
`Workspace`	No	The workspace the agent belongs to.
`Owner`	Yes	Defaults to you. Change it if you are registering on someone else's behalf.
`Department`	No	Useful for reporting and for routing incidents.
`Connection`	Conditional	For platforms that need a connection, `Connection` and `Agent` replace `Name` and both are required. Ties the agent to the account it runs in.
`Agent`	Conditional	Pick the specific agent on that connection.
`Connection type`	—	`Observe · Advisory` is preselected. `Control · Enforced` can be selected for `Anthropic Agent SDK`, `OpenAI Agent SDK`, `LangChain` and `LangGraph`; for other platforms it is disabled ("This platform can't route calls through the gateway yet, so it can only be observed.").

Step 4: Register

Click Register Agent. Back returns to the platform list; Cancel closes without saving.

Step 5: Instrument the agent

Registration creates the record. It does not start telemetry. Until the agent reports its runs, the row shows 0 traces and a dash under LAST TRACE, and no policy can evaluate anything.

Registration creates the record. Traces start when the agent is instrumented, and the agent's own Setup tab is where you do that: it shows the SDK steps for the platform you chose.

The tab reads Waiting for traces… until the first one lands. The agent then turns Active on its own.

Result

After registering a single agent you land on its detail page. The agent appears in the registry with the name, platform and owner you gave it. Once instrumentation is in place and the agent runs, LAST TRACE and ACTIVITY populate, cost begins accruing, and any policy applied to the agent starts evaluating its traces.

Important considerations

Two required fields carry all the context — Name and Description are the only free text on the record. They are what a reviewer has to work with.
Registration is not instrumentation — A registered, uninstrumented agent is a claim that something exists. It produces no traces and no incidents.
Manual registration does not keep up automatically — A second agent on the same framework is a second registration. Connection-based Discovery keeps finding new agents; manual registration does not.
Choose the tier deliberately — Control · Enforced is available for Anthropic Agent SDK, OpenAI Agent SDK, LangChain and LangGraph. Choose it only if the agent routes its calls through UnifyApps.

Register an Agent Manually