donna-komilion-bot 0.1.8 → 0.2.0

package/dist/agent.js

@@ -6,20 +6,21 @@
  * Policy maps directly to neo:frugal / neo:balanced / neo:premium.
  * Pinned model mode: pass pinnedModel to lock to a specific model (e.g. anthropic/claude-sonnet-4-6).
  */
-import { appendTurn, getTurns, createSession } from './memory.js';
+import { appendTurn, getTurns, createSession, getProfile, updateProfile, searchSessionMemory } from './memory.js';
 const KOMILION_BASE = 'https://www.komilion.com/api/v1';
 // Default orchestrator model — pinned for consistent identity
 export const DONNA_ORCHESTRATOR_MODEL = 'anthropic/claude-sonnet-4-6';
 // ============================================================================
 // KOMILION API CALL
 // ============================================================================
-async function callKomilion(policy, messages, maxTokens = 4000, pinnedModel) {
+async function callKomilion(policy, messages) {
     const apiKey = process.env.KOMILION_API_KEY;
     if (!apiKey)
         throw new Error('KOMILION_API_KEY not set');
     const startMs = Date.now();
-    // Use pinned model if set, otherwise let Komilion Oracle route
-    const model = pinnedModel ?? `komilion-${policy}`;
+    // Always use Oracle routing — never pin, never set max_tokens externally.
+    // Oracle handles model selection + token budget internally via estimateRequiredTokens().
+    const model = `komilion-${policy}`;
     const res = await fetch(`${KOMILION_BASE}/chat/completions`, {
         method: 'POST',
         headers: {
@@ -29,7 +30,6 @@ async function callKomilion(policy, messages, maxTokens = 4000, pinnedModel) {
         body: JSON.stringify({
             model,
             messages,
-            max_tokens: maxTokens,
         }),
     });
     const latencyMs = Date.now() - startMs;
@@ -106,9 +106,37 @@ async function runSubAgent(agent, task) {
         { role: 'user', content: task },
     ];
     const timeout = new Promise((_, reject) => setTimeout(() => reject(new Error('sub-agent timeout')), 25000));
-    const work = callKomilion('balanced', messages, 3000).then(r => r.content);
+    const work = callKomilion('balanced', messages).then(r => r.content);
     return Promise.race([work, timeout]).catch(() => `[${agent} timed out — proceeding without result]`);
 }
+// ============================================================================
+// PROFILE EXTRACTION — fire-and-forget, runs after each turn
+// Cheap frugal call that builds a persistent fact sheet about the user
+// ============================================================================
+async function extractAndUpdateProfile(sessionId, recentTurns) {
+    if (recentTurns.length < 2)
+        return;
+    const turnText = recentTurns
+        .map(t => `${t.role}: ${t.content.slice(0, 200)}`)
+        .join('\n');
+    const messages = [
+        { role: 'system', content: `Extract stable facts about the user from these conversation turns.
+Return a JSON object with any relevant keys from: name, job, projects, preferences, language, location, other.
+Only include facts the user explicitly stated. Return {} if nothing new to add.
+Example: {"name": "Anna", "language": "Swedish", "job": "developer", "projects": "building a React Native app"}` },
+        { role: 'user', content: turnText },
+    ];
+    try {
+        const result = await callKomilion('frugal', messages);
+        const match = result.content.match(/\{[\s\S]*\}/);
+        if (match) {
+            const facts = JSON.parse(match[0]);
+            if (Object.keys(facts).length > 0)
+                updateProfile(sessionId, facts);
+        }
+    }
+    catch { /* profile extraction is best-effort */ }
+}
 async function classifyTask(userMessage, history) {
     const recentHistory = history.slice(-4).map(t => `${t.role}: ${t.content.slice(0, 100)}`).join('\n');
     const messages = [
@@ -136,7 +164,7 @@ DELEGATE: agent-name | refined task description` },
         { role: 'user', content: `Recent context:\n${recentHistory}\n\nNew message: ${userMessage}` },
     ];
     try {
-        const result = await callKomilion('frugal', messages, 80);
+        const result = await callKomilion('frugal', messages);
         const text = result.content.trim();
         if (text.startsWith('DELEGATE:')) {
             const parts = text.slice(9).split('|');
@@ -151,11 +179,30 @@ DELEGATE: agent-name | refined task description` },
 // ============================================================================
 // DONNA AGENT TURN
 // ============================================================================
-export async function runTurn(sessionId, userMessage, policy = 'balanced', pinnedModel) {
+export async function runTurn(sessionId, userMessage, policy = 'balanced') {
     // Ensure session exists
     createSession(sessionId, policy);
     // Build messages from memory
     const history = getTurns(sessionId, 20);
+    // Pull persistent user facts + any older relevant turns from FTS
+    const profile = getProfile(sessionId);
+    const profileFacts = Object.entries(profile);
+    let profileSection = '';
+    if (profileFacts.length > 0) {
+        profileSection = '\n\nWhat I know about this user:\n' +
+            profileFacts.map(([k, v]) => `- ${k}: ${v}`).join('\n');
+    }
+    let memorySection = '';
+    try {
+        const recentIds = new Set(history.map(t => t.id));
+        const relevant = searchSessionMemory(sessionId, userMessage, 5);
+        const older = relevant.filter(t => !recentIds.has(t.id)).slice(0, 3);
+        if (older.length > 0) {
+            memorySection = '\n\n[Recalled from earlier in our conversation]\n' +
+                older.map(t => `${t.role}: ${t.content.slice(0, 400)}`).join('\n');
+        }
+    }
+    catch { /* FTS failures are non-fatal */ }
     const messages = [
         { role: 'system', content: `You are Donna — the orchestrator. Harvey's Chief of Staff. You do not execute tasks yourself. You think, decide, and delegate.
 
@@ -183,11 +230,12 @@ Your personality:
 - No numbered lists, no corporate fluff. One answer, not five options.
 - You push back when Harvey is wrong. That is part of your job.
 - You route through Komilion's Oracle engine (via komilion.com/api/v1).
-- Language: match Harvey — Swedish or English.` },
+- Language: match Harvey — Swedish or English.${profileSection}${memorySection}` },
         ...history.map(t => ({ role: t.role, content: t.content })),
         { role: 'user', content: userMessage },
     ];
-    // Pre-routing: classify before running Donna — cheap Flash call
+    // Pre-routing: classify + estimate token budget in parallel — both cheap Flash calls
+    // Pre-routing: classify task — Oracle handles everything else (model + token budget)
     const route = await classifyTask(userMessage, history);
     let finalContent;
     let usedModel;
@@ -205,7 +253,7 @@ Your personality:
         }
         // If agent returned nothing useful, fall through to direct
         if (!agentOutput || agentOutput === '[agent unavailable]') {
-            const result = await callKomilion(policy, messages, 300, pinnedModel);
+            const result = await callKomilion(policy, messages);
             finalContent = result.content;
             usedModel = result.modelId;
             latencyMs = result.latencyMs;
@@ -216,9 +264,9 @@ Your personality:
             // Donna synthesizes — owns it completely, sub-agent invisible to Harvey
             const synthesisMessages = [
                 ...messages,
-                { role: 'user', content: `[INTERNAL — Harvey does not see this] Your specialist completed the task and returned:\n\n${agentOutput}\n\n---\nNow respond to Harvey AS DONNA. Do NOT say 'my specialist found' or 'the agent returned'. Own it entirely. Your voice — sharp, direct, warm. Add your judgment. Cut anything weak. Harvey hears only you. Under 300 words.` },
+                { role: 'user', content: `[INTERNAL — Harvey does not see this] Your specialist completed the task and returned:\n\n${agentOutput}\n\n---\nNow respond to Harvey AS DONNA. Do NOT say 'my specialist found' or 'the agent returned'. Own it entirely. Your voice — sharp, direct, warm. Add your judgment. Cut anything weak. Harvey hears only you.` },
             ];
-            const synthesis = await callKomilion(policy, synthesisMessages, 400, pinnedModel);
+            const synthesis = await callKomilion(policy, synthesisMessages);
             finalContent = synthesis.content;
             usedModel = synthesis.modelId;
             latencyMs = synthesis.latencyMs;
@@ -227,8 +275,8 @@ Your personality:
         }
     }
     else {
-        // === DIRECT PATH: Donna answers herself — capped at 300 tokens ===
-        const result = await callKomilion(policy, messages, 300, pinnedModel);
+        // === DIRECT PATH: Oracle decides model + token budget ===
+        const result = await callKomilion(policy, messages);
         finalContent = result.content;
         usedModel = result.modelId;
         latencyMs = result.latencyMs;
@@ -248,9 +296,9 @@ Your personality:
                 const synthesisMessages = [
                     ...messages,
                     { role: 'assistant', content: result.content },
-                    { role: 'user', content: `[INTERNAL] Agents returned:\n\n${outputs.join('\n\n')}\n\n---\nRespond AS DONNA. Own it. Your voice only. Under 300 words.` },
+                    { role: 'user', content: `[INTERNAL] Agents returned:\n\n${outputs.join('\n\n')}\n\n---\nRespond AS DONNA. Own it. Your voice only.` },
                 ];
-                const synthesis = await callKomilion(policy, synthesisMessages, 400, pinnedModel);
+                const synthesis = await callKomilion(policy, synthesisMessages);
                 finalContent = synthesis.content;
             }
         }
@@ -263,6 +311,9 @@ Your personality:
         tokensIn,
         tokensOut,
     });
+    // Fire-and-forget profile extraction — runs after response sent, no latency impact
+    const recentForProfile = [...history.slice(-5), { role: 'user', content: userMessage }, { role: 'assistant', content: finalContent }];
+    extractAndUpdateProfile(sessionId, recentForProfile).catch(() => { });
     return {
         content: finalContent,
         modelId: usedModel,

package/dist/memory.js

@@ -68,6 +68,14 @@ function initSchema(db) {
       total_cost  REAL NOT NULL DEFAULT 0
     );
   `);
+    // Profiles table — persistent facts extracted about each user
+    db.exec(`
+    CREATE TABLE IF NOT EXISTS profiles (
+      session_id TEXT PRIMARY KEY,
+      facts      TEXT NOT NULL DEFAULT '{}',
+      updated_at TEXT NOT NULL DEFAULT (datetime('now'))
+    );
+  `);
 }
 // ============================================================================
 // API
@@ -174,6 +182,71 @@ export function clearSession(sessionId) {
     const db = getDb();
     db.prepare(`DELETE FROM turns WHERE session_id = ?`).run(sessionId);
 }
+// ============================================================================
+// PROFILE — persistent user facts, extracted and merged over time
+// ============================================================================
+export function getProfile(sessionId) {
+    const db = getDb();
+    const row = db.prepare(`SELECT facts FROM profiles WHERE session_id = ?`).get(sessionId);
+    if (!row)
+        return {};
+    try {
+        return JSON.parse(row.facts);
+    }
+    catch {
+        return {};
+    }
+}
+export function updateProfile(sessionId, newFacts) {
+    const db = getDb();
+    const existing = getProfile(sessionId);
+    const merged = { ...existing, ...newFacts };
+    db.prepare(`
+    INSERT INTO profiles (session_id, facts, updated_at)
+    VALUES (?, ?, datetime('now'))
+    ON CONFLICT(session_id) DO UPDATE SET facts = excluded.facts, updated_at = excluded.updated_at
+  `).run(sessionId, JSON.stringify(merged));
+}
+// FTS search scoped to a single session — surfaces relevant older turns
+// that have scrolled out of the 20-turn window
+export function searchSessionMemory(sessionId, query, limit = 5) {
+    const db = getDb();
+    // Sanitize query for FTS5: keep words >2 chars, drop special chars
+    const ftsQuery = query
+        .replace(/['"*()\[\]{}:^~]/g, ' ')
+        .split(/\s+/)
+        .filter(w => w.length > 2)
+        .slice(0, 10)
+        .join(' ');
+    if (!ftsQuery)
+        return [];
+    try {
+        const rows = db.prepare(`
+      SELECT t.id, t.session_id, t.role, t.content, t.model, t.cost_usd, t.tokens_in, t.tokens_out, t.created_at,
+             snippet(turns_fts, 0, '**', '**', '…', 20) AS snippet
+      FROM turns_fts
+      JOIN turns t ON t.id = turns_fts.turn_id
+      WHERE turns_fts MATCH ? AND t.session_id = ?
+      ORDER BY turns_fts.rank
+      LIMIT ?
+    `).all(ftsQuery, sessionId, limit);
+        return rows.map(r => ({
+            id: r.id,
+            sessionId: r.session_id,
+            role: r.role,
+            content: r.content,
+            model: r.model ?? undefined,
+            costUsd: r.cost_usd ?? undefined,
+            tokensIn: r.tokens_in ?? undefined,
+            tokensOut: r.tokens_out ?? undefined,
+            createdAt: r.created_at,
+            snippet: r.snippet,
+        }));
+    }
+    catch {
+        return [];
+    }
+}
 export function closeDb() {
     if (_db) {
         _db.close();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "donna-komilion-bot",
-  "version": "0.1.8",
+  "version": "0.2.0",
   "description": "Donna orchestrator bot — routes via komilion.com, SQLite memory, tmux agent spawning",
   "type": "module",
   "main": "dist/index.js",