Project operator

An agent that works inside the project boundary.

Investigate deploys and project state in persistent threads. Proposed mutations carry evidence, approval state, and an audit trail instead of disappearing into a chat transcript.

Automation with a control plane underneath it

The operator uses typed project operations. It does not receive an unbounded shell or bypass the roles and billing checks enforced by the API.

Durable threads

Keep investigation context, runs, events, evidence, and decisions attached to the project for later review.

Evidence before action

Read deployment state, logs, configuration, and supported project resources before proposing a change.

Approval gates

Sensitive actions can wait for a human decision; execution rechecks current authorization before proceeding.

Scoped identity

Agent authorization uses explicit scopes and project membership rather than a copied long-lived account token.

Verification events

A run records what happened after an action so success is based on observed state, not merely an accepted request.

One operation contract

The same resource operations back the dashboard, Rust CLI, API, MCP, and infrastructure manifest workflows.

Workflow

Investigate, approve, verify

  1. 01

    Start with project state

    Create or continue a project thread and ask the operator to inspect a deployment or operational problem.

    Evidence retained
  2. 02

    Review the proposed action

    See the intended tool, target, preview, and risk before approving or rejecting a sensitive mutation.

    Decision explicit
  3. 03

    Execute and verify

    The run uses the current user authority, records events, and checks the resulting project state.

    Outcome recorded

Give your agent a cloud it can operate.

No free tier. No seats. Pay for what you run.