Bring every agent under management.Dome is GA.

Start free
Dome Systems

Use case

Onboard agents

Using Dome's Agent Registry

Onboarding an agent is four resources. An Agent record carrying the credential, the Callers it may act for, the Rules that bound what it reaches, and the Audits that record what it did. Register once and every call after that has a name on it.

Composed from

Four resources, one record

Agents, Callers, Rules and Audits. The registry is the name for those four working together, and the agents your teams build and the agents your customers point at you both run on them.

Agents

The record and the credential. Name, owner, declared capabilities, and a lifecycle that runs provisioned, active, suspended, revoked.

Callers

The person the agent is acting for. Asserted per request and verified by OIDC or HMAC before it counts as an identity.

Rules

What this agent may reach. Written once, evaluated on every call, versioned in a repo.

Audits

What happened. Agent, caller, rule, result, in one vocabulary across every call.

Nothing here belongs only to the Agent Registry. One policy language, one audit vocabulary, one access layer, across every composition. See the whole resource set.

Scenario

The agents your teams already shipped

A product manager wired an agent into Salesforce and the warehouse on Tuesday. By Thursday two other teams had forked it. Nothing in your logs tells any of them apart from the apps they run inside.

What makes it hard

Registration has to beat not registering

A review queue loses to a cron job every time. The governed path has to be the fast one.

Discovery arrives with the incident

The first record of an ungoverned agent is usually a post-mortem, reconstructed from logs that were never designed to name agents.

Blocking relocates the work

Restrict one tool and the next attempt routes around you. The agent still ships.

Exceptions compound

Per-team fixes get more fragile with every agent added. The count grows faster than the fixes.

The work, in resources

  1. Agents. Register from the SDK, the CLI, or a CI job. The credential issued here is the root of trust for every call that follows.

  2. Callers. Point Dome at your identity provider. The person behind the agent travels with the request instead of stopping at the front door.

  3. Rules. Attach the tools and models this agent needs. Anything unattached is unreachable, and it does not appear in discovery either.

  4. Audits. Every call lands with the agent, the caller, the rule and the result. Nothing left to reconstruct.

Scenario

The agents your customers point at your product

An integration partner ships an agent that drives your support API. Traffic triples on Wednesday, the call pattern changes, and a customer opens a ticket. You need to name whose agent, which version, acting for which end user.

What makes it hard

An API key names a company

It does not name the agent, the version, or the person the agent is working for.

Isolation is the product

One scoping mistake is a cross-customer data incident. Nobody files that as a bug report.

Support asks for the record

Answering what a partner's agent did at 14:20 needs a per-tenant trail. API access logs do not produce one.

Attribution comes before billing

You cannot charge for, or contain, what you cannot attribute to a tenant.

The work, in resources

  1. Agents. Each customer agent registers under its own tenant and gets its own credential. Suspend one without touching the others.

  2. Callers. The partner asserts their end user per request. Your backend learns whose request it really is, not just which agent forwarded it.

  3. Rules. Write once at tenant scope. Per-customer exceptions go in as overrides rather than forks.

  4. Audits. One trail per tenant, filterable and exportable. Hand it to support, or stream it to the customer.

What changes

When registration is the fast path

The same agents still get built and the same teams still ship them. What changes is the route they take to production, and who sees them on the way.

Before

After

Discovery after the fact

Ungoverned agents surface through outages, compliance reviews, or incidents. Each one becomes a reconstruction project from logs that were never designed to track agents.

Discovery as a consequence

Every agent surfaces the moment it tries to call. Registered, identified, attributable. The platform team sees what is running without asking around.

Audit assembled retroactively

Compliance evidence is rebuilt from access logs, interview notes, and best guesses. The bill comes due before every audit window, and it grows with the agent count.

Audit as a byproduct

Every governed action lands in the trail with who, what, under which rule, and with what result. Evidence is produced by running, not before audits.