Identity and access
Know whose authority is behind the action. Each agent acts under its own identity and its own role. For mediated connections, the credential stays in Oxagen.
The mandates table on the Guardrails page, with illustrative data. The agent invoice-bot holds a mandate for monthly infrastructure invoices under PO-4471, active until Dec 31, 2026. A mandate for vendor payouts under PO-4390 is revoked.
Agent identity
When three engineers run the same agent under one shared token, the audit log shows the token and nobody behind it. In Oxagen an agent is a principal in its own right, with its own identity and its own role. It does not borrow yours.
Oxagen decides every capability the agent asks for against that role, records the call as the agent's, and bounds it by the person it acts for. An agent cannot receive a role that grants more than the person who assigns it holds. Oxagen refuses that assignment and names the capabilities that overreached.
The Equipment page with illustrative data. Each agent has its own card: Release manager runs on Claude Code on the laptop mbpro and has 214 runs, and billing-fixer is suspended. The GitHub, Stripe, and Linear servers sit beside them.
What an agent request carries
The Waiting on a person card from the Runs page, with illustrative data. Two calls wait: Push to release/4.11, and Run the staging migration, which the rule migrations routed to you, with a button to approve the request.
Request decisions
- The agent asks
When a task needs a system, a scope, or an action its mandate does not already cover, the agent requests it at the moment of use.
- A rule answers
A workspace rule allows the request, denies it, or routes it to a named person. Stella holds a routed call until the person answers. Codex and Cursor receive a refusal that gives the reason.
- The record keeps the answer
Oxagen records the answer and the rule behind it beside the call, so a reviewer can trace the action back to its authority.
Connection credentials
For mediated connections, Oxagen uses the connection credential on the agent's behalf, and the agent does not receive it. The agent's identity says who acted. The connection credential says which account the action used.
Identity and enrollment paths
| Path | For | How it starts | How it ends |
|---|---|---|---|
| Agent enrollment | A machine running Claude Code, Codex, Cursor, or Stella | A person runs oxagen agent enroll with their CLI login | Revoke one machine, or every machine an agent is enrolled on, with oxagen agent unenroll |
| Machine registration | A machine set up from a coding agent's MCP connection | A one-time enrollment token that expires unused after 30 minutes | The same revocation as agent enrollment |
| Single sign-on | People in the organization | OIDC or SAML 2.0 sign-in, with SCIM 2.0 to create people, on the Enterprise plan | SCIM removal from Okta or Microsoft Entra ID |
Default for a new agent
In scope
- Every new agent starts with the Agent Contributor role. It reads freely, makes low and medium risk changes, and has high-risk requests routed to a person first.
- MCP tools need first-use consent under that role.
- Agent role checks run on every plan.
Out of scope
- A capability the person assigning the role does not hold.
- A credential for a connection Oxagen does not mediate, which the agent may still hold itself.
Product walkthrough
Review one access request and the authority behind its answer.