Struct SessionRecorder 

pub struct SessionRecorder { /* private fields */ }

Records every AgentEvent into a structured tree of Sessions and LoopRecords.

Call on_event for every event emitted on the agent’s tx channel, then flush before shutdown or saving.

Session grouping

Sessions are keyed by session_id. Every AgentStart event that carries a session_id the recorder has not seen before opens a new Session; all subsequent loops with the same session_id are appended to that session.

The recorder never rotates sessions on its own. If you want a new session to start after a period of inactivity, call BasicAgent::check_and_rotate (or BasicAgent::new_session) before the next prompt. The next AgentStart will carry the new session_id and the recorder will open a fresh Session automatically, with SessionFormation::InactivityTimeout or SessionFormation::FirstLoop as the recorded reason.

Example

use phi_core::session::{SessionRecorder, SessionRecorderConfig};
use phi_core::AgentEvent;

let mut recorder = SessionRecorder::new(SessionRecorderConfig::default());
// Feed events as they arrive:
// recorder.on_event(event);
recorder.flush();

Implementations

impl SessionRecorder

pub fn new(config: SessionRecorderConfig) -> Self

Create a new recorder with the given configuration.

pub fn on_event(&mut self, event: AgentEvent)

Feed one event into the recorder.

Must be called for every event emitted on the agent’s tx channel.

pub fn flush(&mut self)

Finalize all open LoopRecords (status → LoopStatus::Aborted) and move them into their sessions.

Call before saving or on process shutdown.

pub fn checkpoint(&mut self) -> usize

Promote sessions that have no remaining open loops to the completed list, without aborting any running loops.

A session is eligible when every loop belonging to it has already received an AgentEnd event (i.e. it has no entry in the internal open-loops map). Sessions that still have active loops are left in place.

This is intended for periodic checkpointing in production: save finished sessions to disk while leaving in-flight agent runs untouched. In contrast, flush first aborts all open loops and then promotes everything.

Returns the number of sessions that were promoted.

pub fn drain_completed(&mut self) -> Vec<Session>

Drain all completed sessions out of the recorder (consuming them).

Useful for periodic checkpointing. Call flush first if you want to include in-progress sessions, or checkpoint to drain only fully-finished sessions without aborting active loops.

pub fn sessions(&self) -> impl Iterator<Item = &Session>

All sessions known to this recorder (completed and in-progress).

pub fn get_session(&self, session_id: &str) -> Option<&Session>

Look up a session by session_id.

pub fn current_loop(&self, loop_id: &str) -> Option<&LoopRecord>

Look up an in-progress LoopRecord by loop_id.