Skip to content

Connect coding agents

wheeljack launches coding agents already installed on your machine. It does not bundle subscriptions, provider credentials, model access, or provider billing.

Adapter Executable Structured mode Authentication outside wheeljack
Claude Code claude Yes Install and authenticate Claude Code.
Codex CLI codex Yes Run codex login or use the CLI’s normal sign-in flow.
OpenCode opencode Yes Run opencode auth login or configure its normal provider credentials.
Pi pi Yes Install @earendil-works/pi-coding-agent, then run pi /login.
Generic shell Platform shell No No provider authentication.

Provider installation commands change independently of wheeljack. Use each provider’s official documentation, then confirm that the CLI starts successfully in an ordinary terminal before diagnosing it inside wheeljack.

  1. Install and authenticate the CLI outside wheeljack.
  2. Restart wheeljack if the executable was added to PATH while wheeljack was already running.
  3. Open Settings → Agents.
  4. Run detection, then verify the adapter.
  5. Choose the adapter as the default or select it when creating a new agent pane.

Verification checks executable resolution, the adapter’s structured protocol, and the launch profile wheeljack will use. A changed executable or protocol profile invalidates stale verification instead of trusting an old result.

A generic shell displays terminal output only. A structured adapter additionally exposes chat messages, reasoning, tool activity, approvals, questions, cancellation, model selection, Plan routing, autonomy requests, and review evidence.

wheeljack fails a structured launch when the required persistent-session contract is unavailable. It does not silently downgrade a Plan task to a one-shot or unstructured agent.

  • Agent default preserves the CLI’s normal sandbox and approval policy.
  • Full access explicitly asks the CLI to allow internet access and local files without its ordinary approval boundary.

Managed agents can request bounded coordination actions such as messaging a peer, starting a child, handing off work, or requesting review. Configure each action as Allow automatically, Ask every time, or Deny in agent settings. wheeljack enforces the selected policy and records the request and result in Autonomy history.