@ai-presence/core

Core presence state runtime for AI interfaces.

This package is intentionally tiny in the prototype: it defines the canonical interaction states, transition events, a small runtime that renderers can subscribe to or poll, and shared controller inputs that stay renderer-agnostic.

The face demo consumes this layer as a browser global so the current static prototype still works without a build step. Package consumers can use either the CommonJS entry or the ESM export:

import { PresenceEvent, createPresenceRuntime } from "@ai-presence/core";

const presence = createPresenceRuntime();
presence.send(PresenceEvent.SUBMIT);

For debugging and demos, createPresenceTrace records a bounded transition timeline and summarizePresenceTrace turns it into compact integration evidence:

import {
  PresenceEvent,
  createPresenceRuntime,
  createPresenceTrace,
  summarizePresenceTrace,
} from "@ai-presence/core";

const trace = createPresenceTrace({ limit: 32 });
const presence = createPresenceRuntime();
const detach = trace.attach(presence);

presence.send(PresenceEvent.SUBMIT);
presence.send(PresenceEvent.STREAM_OPEN);
presence.send(PresenceEvent.TOKEN);

const summary = summarizePresenceTrace(trace);
console.log(trace.getEntries().map((entry) => entry.state));
console.log(summary.presenceBeforeOutputMs);
console.log(summary.interrupted);
detach();

Trace entries include elapsedMs and sincePreviousMs, and the summary exposes facts such as firstTokenMs, firstOutputMs, presenceBeforeOutputMs, interruptMs, interrupted, finalState, hasOutput, and complete without coupling core to any renderer.

Renderers can derive shared interaction-posture inputs from the same snapshot before mapping them into renderer-specific controllers:

import { PresenceEvent, createPresenceRuntime, presenceControlInputsForSnapshot } from "@ai-presence/core";

const presence = createPresenceRuntime();
const snapshot = presence.send(PresenceEvent.STREAM_OPEN);
const inputs = presenceControlInputsForSnapshot(snapshot);

console.log(inputs.latencyPhase); // "before-output"
console.log(inputs.attentionTarget); // "response"
console.log(inputs.transitionEvent); // "stream-open"

These values describe observable interaction posture such as attention target, tension, speech activity, interruption, latency phase, recovery, and compact transition context. They are not emotion detection or private emotion inference.