AgentOS

import { agent } from '@framers/agentos';

// Personality is six 0-1 trait values. The runtime appends a trait-derived
// directive to the system prompt and modulates three cognitive-memory
// mechanisms (involuntary recall, consolidation, schema encoding) based on
// honesty / emotionality / openness. Default is neutral (0.5) on every axis.
const tutor = agent({
  provider: 'openai',
  model: 'gpt-4o',
  instructions: 'You are a patient programming tutor.',
  personality: {
    honesty:           0.85,  // direct, transparent, no flattery
    emotionality:      0.65,  // tone-aware without being clinical
    extraversion:      0.50,
    agreeableness:     0.75,  // warm, encouraging
    conscientiousness: 0.90,  // structured, thorough, follow-through
    openness:          0.85,  // creative, exploratory framing
  },
  memory: {
    enabled:    true,         // session history persists automatically
    cognitive:  true,         // Ebbinghaus decay + reconsolidation + 6 more
  },
});

// Sessions scope conversation history by ID. Same agent, multiple users,
// no cross-talk — each session has its own memory bag.
const session = tutor.session('user-42');

// The agent remembers across turns. Context from the first message is
// recalled automatically in the second.
await session.send('My exam is on distributed systems next Thursday.');
await session.send('I struggle with consensus algorithms.');
const reply = await session.send('What should I focus on this week?');

console.log(reply.text);
// => "Given Thursday's exam and your block on consensus, lock in Paxos
//     and Raft this week. Start with the leader-election proof…"

// Inspect what the session actually carries — full message history +
// token usage. Useful for debugging memory recall or cost.
console.log(session.messages());
const usage = await session.usage();
console.log(`Total tokens: ${usage.totalTokens}`);