Every agent action reaches your systems through Tier Two.

And on the way through, it picks up an identity, a policy decision, a just-in-time approval where it matters, and a signed record.

Decision ledger

Follow every agent action from identity to outcome.

Filter the decision timeline by actor, agent, connection, action, identity, or result. Open any denial to see what policy stopped it before a downstream credential was used.

acme · governance control plane

Audit / Action ledger

Every decision in one timeline

live
Last 24 hoursAll connectionsAll results

Dana

research-agent

query_dataset

finance/kpis · Employee

Allowed

Dana

research-agent

refresh_schema

production · Elevated

Approval

Morgan

build-agent

drop_database

production · Organization

Denied

Morgan

build-agent

list_projects

engineering · Organization

Allowed

Selected denial

drop_database · production

No allow rule matched this action. The call was blocked before any downstream credential was used.

Actor
Morgan
Agent
build-agent
Connection
Metabase
Event ID
evt_7ca1
Tenancy & enrollment

Know every agent in your company.

Your org is the tenant

Teams, members, and admins map to your organization. Admins decide; owners approve.

Agents bound to owners

Every agent belongs to an owner — a person or a team. Ownership routes its approvals and anchors its audit trail.

Devices pair and revoke

Agents sign from paired devices with local keys. Revoke a device and every key on it stops working.

A directory, not a mystery

Every agent in one list: live sessions, latency, denied-call trend, device trust, and what it's allowed to reach.

Enrolling a fleet? Push the client by MDM or hand out an admin-minted enroll secret. Machines enroll without a human, land unassigned, and fail closed — no governed call succeeds until someone binds the device at first sign-in.

Connections

Connect your systems. Keep your credentials.

Register the systems agents may reach. For each connection, choose per-user credentials (brokered OAuth or a linked key, so downstream permissions stay the employee's) or one org-provisioned credential that avoids credential setup for each employee.

your side

Agent

device signs the request; no downstream credential on the laptop

governance plane

Tier Two

policy → identity lane → credential from the sealed vault

downstream

Your connection

receives the call with the actor attached

On org-provisioned connections, calls are forwarded server-side: the credential is decrypted only when needed, used in-process to forward the call, and never reaches a device.

Policy

Every rule reads as a plain sentence.

Rules are authored with schema-driven pickers over real connections, tools, and resources, and every rule is rendered back as plain prose, so review means reading, not decoding.

Default-deny

No rule, no call. Rules match by connection, tool, and resource; the most specific rule wins, and explicit denies beat everything.

Simulate before you ship

Test a call against your rules before an agent ever makes it, and lint catches rules that can't fire.

Approvals & access

Grant routine access. Approve sensitive actions just in time.

Just-in-time approvals

Sensitive actions can require a fresh, one-use approval. Standing grants do not bypass step-up rules, and every decision is recorded.

Standing grants

For the routine, agents request access; an admin decides once and the grant stands until revoked. Requests route to the agent's owner or team.

Audit

Know exactly what every agent did.

The action ledger

Every call, allowed or denied, with who, which agent, what action, what resource, and which identity it ran as. Filterable by all of it.

The governance ledger

Every change to the rules themselves: rule edits, grants, credential rotations, and device pairings, each with the actor who made it.

Both ledgers stream to your SIEM: HMAC-signed webhooks or Splunk HEC, three versioned streams, delivered in order and at least once, minutes behind live.

Who it's for

Both sides of the rollout.

For security & platform teams

  • One place to see, allow, and revoke everything agents can do.
  • Default-deny policy with simulation, lint, and prose review.
  • A signed, append-only record for every decision, including the denials.

For the engineers whose agents do the work

  • Pair once; your agent works everywhere the org allows.
  • No credentials to paste into config files; the gateway carries them.
  • When the agent needs more, it asks in-line and a human approves.

Give your agents an identity you control.

Self-serve and free while in beta. Start with one agent and one connection.