Agent Clients

An Agent Client connects Pontia to a coding agent’s native session, control channel, and conversation history. The agent still runs locally with its own credentials, configuration, tools, and filesystem permissions.

Pontia currently includes two Agent Clients:

Choose Clients during initialization

Run:

pontia init

The initializer checks the selected Clients before changing Pontia’s configuration. pi is selected by default; Codex must be selected explicitly. You may select both or neither, although Pontia cannot start useful agent Sessions without an available Client.

Running pontia init again preserves the existing External API token and Workspace Browser roots. It can also install or repair a selected Client integration.

Client capabilities differ

Pontia uses one Session model across Clients, but it does not pretend that every Client supports the same controls. The Dashboard enables actions such as interruption, model changes, history branching, and queued delivery only when the active Client reports the corresponding capability.

A Pontia Session is durably associated with one client-native conversation identity after binding. Reopening the same native conversation continues the existing Session; a new native conversation belongs to a new Session.

Client credentials and permissions

Pontia does not replace the Client’s own authentication or sandbox configuration. Sign in to pi or Codex and verify that it can perform the intended work before starting the integration. The agent can access only what its local process and configuration permit, even if the Workspace is visible in Pontia.

See Troubleshooting when a Client is missing, cannot start, or stops reporting output.