osborn 0.9.168 → 0.9.169

package/dist/claude-llm.d.ts

@@ -79,6 +79,14 @@ export declare const NAMED_AGENTS: {
         prompt: string;
     };
 };
+export declare const FAST_MODEL = "haiku";
+/**
+ * Apply turbo override to an agents roster.
+ * When turbo is true returns a DEEP COPY with every agent's model set to
+ * FAST_MODEL. When false returns the input object unchanged (zero allocation).
+ * NEVER mutates the input — NAMED_AGENTS and DB-sourced rows are untouched.
+ */
+export declare function applyTurbo(agents: Record<string, any>, turbo: boolean): Record<string, any>;
 /**
  * Claude LLM - Wraps Claude Agent SDK for LiveKit
  * Research mode: reads anything, writes only to session workspace
@@ -133,6 +141,13 @@ export declare class ClaudeLLM extends llm.LLM {
      * it was created with.
      */
     setAgents(agents: ClaudeLLMOptions['agents']): void;
+    /**
+     * Enable/disable Turbo mode. When on, the main model and every sub-agent
+     * spawn on FAST_MODEL. Takes effect at the next query cold start.
+     */
+    setTurbo(on: boolean): void;
+    /** Read-only accessor used by ClaudeLLMStream (private fields are class-scoped). */
+    get turbo(): boolean;
     /**
      * Set session ID to resume a specific conversation
      * Call this before sending the first message to resume from a previous session

package/dist/claude-llm.js

@@ -581,6 +581,24 @@ export const NAMED_AGENTS = {
         ].join('\n'),
     },
 };
+// Turbo mode — single fast model override. ONE place to update for future
+// Gemini/OpenAI fast model swap. Keep this immediately after NAMED_AGENTS.
+export const FAST_MODEL = 'haiku';
+/**
+ * Apply turbo override to an agents roster.
+ * When turbo is true returns a DEEP COPY with every agent's model set to
+ * FAST_MODEL. When false returns the input object unchanged (zero allocation).
+ * NEVER mutates the input — NAMED_AGENTS and DB-sourced rows are untouched.
+ */
+export function applyTurbo(agents, turbo) {
+    if (!turbo)
+        return agents;
+    const out = {};
+    for (const [name, agent] of Object.entries(agents)) {
+        out[name] = { ...agent, model: FAST_MODEL };
+    }
+    return out;
+}
 const RESEARCH_TOOLS = [
     'Read', 'Write', 'Edit', 'Glob', 'Grep',
     'Bash', 'WebSearch', 'WebFetch',
@@ -664,6 +682,9 @@ export class ClaudeLLM extends llm.LLM {
     // Dedup guard — prevents double-firing reviewer/gate if SubagentStop fires
     // more than once for the same agent_id (e.g. retry edge cases).
     #dispatchedFor = new Set();
+    // Turbo mode — when true, every spawned agent (main + sub-agents) runs on
+    // FAST_MODEL regardless of individual model config. Default off = no-op.
+    #turbo = false;
     constructor(opts = {}) {
         super();
         // Session resume/continue options
@@ -823,6 +844,16 @@ export class ClaudeLLM extends llm.LLM {
         this.#opts.agents = agents;
         console.log(`🤖 Named agents ${agents ? `set (${Object.keys(agents).join(', ')})` : 'reset to built-ins'} — applies at next query cold start`);
     }
+    /**
+     * Enable/disable Turbo mode. When on, the main model and every sub-agent
+     * spawn on FAST_MODEL. Takes effect at the next query cold start.
+     */
+    setTurbo(on) {
+        this.#turbo = on;
+        console.log(`⚡ Turbo mode ${on ? 'ON' : 'OFF'} — main model → ${on ? FAST_MODEL : (this.#opts.model || 'claude-opus-4-8[1m]')}; applies at next query cold start`);
+    }
+    /** Read-only accessor used by ClaudeLLMStream (private fields are class-scoped). */
+    get turbo() { return this.#turbo; }
     /**
      * Set session ID to resume a specific conversation
      * Call this before sending the first message to resume from a previous session
@@ -1563,7 +1594,8 @@ class ClaudeLLMStream extends llm.LLMStream {
                 permissionMode: this.#opts.permissionMode,
                 allowedTools,
                 // model: this.#opts.model || 'haiku', // haiku for speed with limited tools, sonnet for full research capabilities (including tool use trace in response)
-                model: this.#opts.model || 'claude-opus-4-8[1m]', // Opus 4.8 + [1m] → 1M context (see get model() note); prevents ~153k early compaction
+                // Turbo: when on, override main model to FAST_MODEL regardless of config.
+                model: this.#llmRef.turbo ? FAST_MODEL : (this.#opts.model || 'claude-opus-4-8[1m]'), // Opus 4.8 + [1m] → 1M context (see get model() note); prevents ~153k early compaction
                 enableFileCheckpointing: true,
                 settingSources: ['project', 'user'],
                 extraArgs: { 'replay-user-messages': null },
@@ -1994,7 +2026,13 @@ class ClaudeLLMStream extends llm.LLMStream {
                 // Named sub-agents — the orchestrator delegates to these specialists.
                 // Built-in definitions live at module level (NAMED_AGENTS); per-user
                 // DB-backed definitions (opts.agents, via set_agents) take precedence.
-                agents: this.#opts.agents ?? NAMED_AGENTS,
+                // TURBO BYPASS: when turbo is on and no custom agents were provided,
+                // opts.agents is undefined — the ?? NAMED_AGENTS fallback would skip
+                // the override entirely. Explicitly apply applyTurbo(NAMED_AGENTS)
+                // so built-in agents always get FAST_MODEL when turbo is on.
+                agents: this.#llmRef.turbo
+                    ? applyTurbo(this.#opts.agents ?? NAMED_AGENTS, true)
+                    : (this.#opts.agents ?? NAMED_AGENTS),
             };
             // Run Claude Agent SDK query() and stream results
             let hasOutput = false;

package/dist/index.js

@@ -30,7 +30,7 @@ const __dirname = dirname(__filename);
 import { createPatch } from 'diff';
 import { loadConfig, getMcpServers, getEnabledMcpServerNames, getVoiceMode, getRealtimeConfig, getDirectConfig, listAllClaudeSessions, getMostRecentSessionId, sessionExists, getSessionSummary, getConversationHistory, ensureSessionWorkspace, getSessionWorkspace, getMcpServerStatusList, buildMcpServersForKeys, listWorkspaceArtifacts } from './config.js';
 import { createSTT, createTTS, createRealtimeModelFromConfig, DIRECT_MODE_STT, DIRECT_MODE_TTS } from './voice-io.js';
-import { createClaudeLLM, NAMED_AGENTS } from './claude-llm.js';
+import { createClaudeLLM, NAMED_AGENTS, applyTurbo } from './claude-llm.js';
 import { clearPipelineFastBrainSession, prewarmBM25Index } from './pipeline-fastbrain.js';
 import { ensureClaudeAuth } from './claude-auth.js';
 import { createSmitheryProxy, destroySmitheryProxy, parseSmitheryUrl, isSmitheryUrl, SmitheryAuthorizationError } from './smithery-proxy.js';
@@ -2239,6 +2239,9 @@ async function main() {
     // switch/resume) reuse them without waiting for a frontend resend. Merged
     // OVER the built-in NAMED_AGENTS (same-name user rows shadow built-ins).
     let userNamedAgents = null;
+    // Turbo mode — read from set_agents handshake (data.turbo). When true every
+    // agent (main + sub-agents) runs on FAST_MODEL. Default false = perfect no-op.
+    let turboMode = false;
     // Built-ins the user has removed (tombstoned in user_agents with enabled=false).
     // A removed built-in disappears from the effective set but stays re-addable
     // from the frontend's "Available to add" list — default agents are never
@@ -3266,14 +3269,21 @@ async function main() {
             voiceMode: 'direct',
             skipTTSQueue: true,
             onCompactionEvent: buildOnCompactionEvent(),
-            // Per-user named agents survive LLM recreations (session switch/resume)
+            // Per-user named agents survive LLM recreations (session switch/resume).
+            // applyTurbo runs post-merge so custom models are also overridden when turbo is on.
+            // When neither custom agents nor turbo are active, pass undefined so the
+            // ClaudeLLM cold-start uses NAMED_AGENTS directly (the ?? NAMED_AGENTS path).
+            // When turbo IS on with no custom agents, undefined would skip the override —
+            // ClaudeLLM handles that internally via the turbo bypass in ClaudeLLMStream.
             agents: (userNamedAgents || userRemovedAgents.length) ? (() => {
                 const base = { ...NAMED_AGENTS };
                 for (const r of userRemovedAgents)
                     delete base[r];
-                return { ...base, ...(userNamedAgents || {}) };
+                return applyTurbo({ ...base, ...(userNamedAgents || {}) }, turboMode);
             })() : undefined,
         });
+        // Apply turbo to the freshly created LLM so the main model also picks it up.
+        directLLM.setTurbo(turboMode);
         currentLLM = directLLM;
         // Reset the session always-allow list for each new direct session
         sessionAlwaysAllowPaths = new Set();
@@ -6008,11 +6018,15 @@ async function main() {
                 userNamedAgents = Object.keys(validated).length ? validated : null;
                 // Removed built-ins (tombstones) — excluded from the effective set.
                 userRemovedAgents = Array.isArray(data.removed) ? data.removed.map(String) : [];
+                // Turbo preference from the frontend (sourced from user_settings DB row).
+                turboMode = !!data.turbo;
                 const base = { ...NAMED_AGENTS };
                 for (const r of userRemovedAgents)
                     delete base[r];
-                const merged = { ...base, ...(userNamedAgents || {}) };
-                const customized = !!userNamedAgents || userRemovedAgents.length > 0;
+                // applyTurbo wraps AFTER the base+custom merge so custom models also get overridden.
+                const merged = applyTurbo({ ...base, ...(userNamedAgents || {}) }, turboMode);
+                const customized = !!userNamedAgents || userRemovedAgents.length > 0 || turboMode;
+                currentLLM?.setTurbo?.(turboMode);
                 currentLLM?.setAgents?.(customized ? merged : undefined);
                 console.log(`🤖 set_agents: ${Object.keys(validated).length} user agent(s), ${userRemovedAgents.length} removed → effective [${Object.keys(merged).join(', ')}]`);
                 await sendToFrontend({

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "osborn",
-  "version": "0.9.168",
+  "version": "0.9.169",
   "description": "Voice AI coding assistant - local agent that connects to Osborn frontend",
   "type": "module",
   "bin": {