Start a NEW task in the local pi agent — a separate CLI coding agent with its own read/bash/edit/write tools and its own context window. Blocks until pi settles, then returns only its final answer plus stats, prefixed [session: ]; continue that session later with pi_reply. Good for: a second opinion from a different model, work kept out of this context, or parallel investigation. Caution: pi has no permission system. With its default tools it edits files and runs shell commands as your user inside cwd. For analysis-only work pass tools or no_tools.

Send a new turn to an existing pi session that is not executing right now — including one that timed out or was cancelled: the session survives, so resume it here instead of restarting with pi. pi still has its prior turns (but never this conversation), so the follow-up can be short. Survives restarts of this server. For a turn still running under 'rpc', use pi_send instead.

List the models pi can actually reach right now — provider, model id, context window, max output, thinking and image support — read from the live catalog. Use it to pick model and thinking values for pi / pi_reply. Starts no session, runs no task.

Deliver a message into a pi turn that is executing right now. Works only on runs started with transport 'rpc' — 'print' runs cannot be reached, and a session that already finished takes pi_reply, not pi_send. Returns immediately; pi's reaction appears in the answer of the pi/pi_reply call still waiting on that turn. pi_running lists reachable sessions.

List pi turns executing at this moment — the ones pi_send can reach — with session id, working directory, elapsed time, and messages already sent in. Only rpc-transport runs appear; 'print' runs are unreachable mid-run. For past sessions use pi_sessions.

List all pi sessions started through this server, newest first, with their working directory — running or finished, including runs that timed out. Use it to recover an id for pi_reply. For turns still executing (pi_send targets), use pi_running.