Product
Industries
Roles
Research
Resources

Developers

You run three agents while you work on something else, each in its own terminal, each with its own scrollback. Oxagen records each run as it happens, so you can read it, steer it, and answer it from one place without changing your editor or your harness.

Setup

Setup runs on the computer where the agent works.

  1. Install the Oxagen app

    It puts the oxagen command on your PATH.

  2. Enroll your harness

    oxagen agent enroll --harness claude-code adds Oxagen's hooks to Claude Code's settings. Codex, Cursor, and Stella enroll the same way.

  3. Check what is active

    oxagen agent status lists each wrapped harness and whether its hooks are present. Which controls work depends on the harness.

  4. Start a run

    The next run reports to Oxagen.

Steering and skills

Turn a useful correction into guidance the team can reuse. Your team's rules, instructions, and skills live in the workspace's steering repository. Oxagen delivers them to each run when it starts.

The run keeps the steering versions it received. oxagen run show <run-id> prints them, so you can tell what this run was told apart from what the repository says today.

Live steering

Send a running agent a message from the run page or the CLI: oxagen steer "Leave the migrations alone and run the tests first" --agent billing-bot. On a host that supports it, the message lands after the tool call in progress. Otherwise it waits for the agent's next prompt. You can cut the model call in flight on runs whose model traffic passes through Oxagen.

The record shows when you sent it and whether the run took it. Cursor takes a message at the end of its turn.

A release agent's run still in flight. A rule routed its push to Sam Chen, who approved it three minutes later. Priya Shah sent a correction, which reached the agent at its next turn, and the agent is now editing the release notes. It has cost $0.66 so far.

One live run

Cut the 4.11.0 release notesRelease agent run by Priya ShahRunning$0.66Cost so far
  1. Run started on Claude CodeMandate release. Operator Priya Shah.
  2. Read 22 files and 31 merged pull requestsAllowed by rule repo.read+$0.62
  3. Asked to push to release/4.11Routed to Sam Chen by rule release-branch+$0.04
  4. Sam Chen approved the pushAnswered in 3 minutes
  5. Priya Shah sent a correctionReached the agent at its next turn
  6. Editing docs/releases/v4.11.0.mdxAllowed by rule docs.write
Illustrative A run in flight with its correction delivered

Approvals and questions

When a rule sends a call to a person on an agent that waits for the answer, everyone allowed to answer is notified, and the run page shows the call with the rule that sent it. When an agent stops to ask you a question, the run page shows it and you answer there.

Steering a run needs an organization Owner, or a workspace Admin or Member.

Run records

Each run's page shows the transcript the way the harness prints it, with each step's time and cost. It lists the files the run changed, its commits, and the pull requests they landed in, with each pull request's checks.

Oxagen locks the run's record when the agent's host ends the run. Locked means any change after that point shows. A run that starts again is locked again when it ends.

Agent setup

Choose your agent and follow its current install guide.