groove-dev 0.27.143 → 0.27.145

package/node_modules/@groove-dev/gui/src/stores/slices/providers-slice.js

@@ -0,0 +1,897 @@
+// FSL-1.1-Apache-2.0 — see LICENSE
+
+import { api } from '../../lib/api';
+import { loadJSON, persistJSON } from '../helpers.js';
+
+export const createProvidersSlice = (set, get) => ({
+  // ── Providers ────────────────────────────────────────────
+  _providerRefreshTick: 0,
+
+  // ── Local Models (Ollama) ─────────────────────────────────
+  ollamaStatus: { installed: false, serverRunning: false, hardware: null },
+  ollamaInstalledModels: [],
+  ollamaRunningModels: [],
+  ollamaCatalog: [],
+  ollamaPullProgress: {},
+
+  // ── Provider Setup (Settings) ──────────────────────────────
+  providerInstallProgress: {},
+
+  // ── Model Lab ──────────────────────────────────────────────
+  labRuntimes: loadJSON('groove:labRuntimes', []),
+  labActiveRuntime: null,
+  labModels: [],
+  labActiveModel: null,
+  labPresets: loadJSON('groove:labPresets', []),
+  labActivePreset: null,
+  labSessions: [],
+  labActiveSession: null,
+  labMetrics: {
+    ttft: null, tokensPerSec: null, tokensPerSecHistory: [], ttftHistory: [],
+    memory: null, peakMemory: null, totalTokens: 0, promptTokens: 0, completionTokens: 0,
+    generationTime: null, generationCount: 0, sessionStartTime: null,
+  },
+  labParameters: loadJSON('groove:labParameters', {
+    temperature: 0.7, topP: 0.9, topK: 40, minP: 0, repeatPenalty: 1.1,
+    maxTokens: 2048, frequencyPenalty: 0, presencePenalty: 0,
+    thinking: false, seed: null, stopSequences: [], jsonMode: false,
+  }),
+  labSystemPrompt: localStorage.getItem('groove:labSystemPrompt') || '',
+  labStreaming: false,
+  labAbortController: null,
+  labLocalModels: [],
+  labLaunching: null,
+  labLlamaInstalled: null,
+  labLaunchPhase: null,
+  labLaunchError: null,
+  labAssistantAgentId: localStorage.getItem('groove:labAssistantAgentId') || null,
+  labAssistantMode: false,
+  labAssistantBackend: localStorage.getItem('groove:labAssistantBackend') || null,
+
+  // ── Provider Actions ──────────────────────────────────────
+
+  async fetchProviders() {
+    return api.get('/providers');
+  },
+
+  // ── Local Models (Ollama) ─────────────────────────────────
+
+  async fetchOllamaStatus() {
+    try {
+      const check = await api.post('/providers/ollama/check');
+      const updates = {
+        ollamaStatus: { installed: check.installed, serverRunning: check.serverRunning, hardware: check.hardware },
+      };
+      if (check.installed) {
+        try {
+          const models = await api.get('/providers/ollama/models');
+          updates.ollamaInstalledModels = models.installed || [];
+          updates.ollamaCatalog = models.catalog || [];
+        } catch {}
+      }
+      if (check.serverRunning) {
+        try {
+          const running = await api.get('/providers/ollama/running');
+          updates.ollamaRunningModels = running.models || [];
+        } catch {
+          updates.ollamaRunningModels = [];
+        }
+      } else {
+        updates.ollamaRunningModels = [];
+      }
+      set(updates);
+      return updates.ollamaStatus;
+    } catch {
+      return get().ollamaStatus;
+    }
+  },
+
+  async startOllamaServer() {
+    try {
+      const result = await api.post('/providers/ollama/serve');
+      if (result.ok) {
+        get().addToast('success', 'Ollama server started');
+        await new Promise((r) => setTimeout(r, 2000));
+        await get().fetchOllamaStatus();
+      }
+      return result;
+    } catch (err) {
+      get().addToast('error', 'Could not start server', err.message);
+      throw err;
+    }
+  },
+
+  async stopOllamaServer() {
+    try {
+      const result = await api.post('/providers/ollama/stop');
+      if (result.ok) {
+        get().addToast('info', 'Ollama server stopped');
+        set((s) => ({
+          ollamaStatus: { ...s.ollamaStatus, serverRunning: false },
+          ollamaRunningModels: [],
+        }));
+      }
+      return result;
+    } catch (err) {
+      get().addToast('error', 'Stop failed', err.message);
+      throw err;
+    }
+  },
+
+  async restartOllamaServer() {
+    try {
+      const result = await api.post('/providers/ollama/restart');
+      if (result.ok) {
+        get().addToast('success', 'Ollama server restarted');
+        await new Promise((r) => setTimeout(r, 2000));
+        await get().fetchOllamaStatus();
+      }
+      return result;
+    } catch (err) {
+      get().addToast('error', 'Restart failed', err.message);
+      throw err;
+    }
+  },
+
+  async pullOllamaModel(modelId) {
+    try {
+      set((s) => ({ ollamaPullProgress: { ...s.ollamaPullProgress, [modelId]: { status: 'pulling', progress: '' } } }));
+      await api.post('/providers/ollama/pull', { model: modelId });
+      set((s) => {
+        const progress = { ...s.ollamaPullProgress };
+        delete progress[modelId];
+        return { ollamaPullProgress: progress };
+      });
+      get().addToast('success', `${modelId} ready to use`);
+      get().fetchOllamaStatus();
+    } catch (err) {
+      set((s) => {
+        const progress = { ...s.ollamaPullProgress };
+        delete progress[modelId];
+        return { ollamaPullProgress: progress };
+      });
+      get().addToast('error', `Pull failed: ${err.message}`);
+    }
+  },
+
+  async deleteOllamaModel(modelId) {
+    try {
+      await api.delete(`/providers/ollama/models/${encodeURIComponent(modelId)}`);
+      set((s) => ({ ollamaInstalledModels: s.ollamaInstalledModels.filter((m) => m.id !== modelId) }));
+      get().addToast('success', `Removed ${modelId}`);
+    } catch (err) {
+      get().addToast('error', `Delete failed: ${err.message}`);
+    }
+  },
+
+  async loadOllamaModel(modelId) {
+    try {
+      await api.post('/providers/ollama/load', { model: modelId });
+      get().addToast('success', `${modelId} loaded into memory`);
+      get().fetchOllamaStatus();
+    } catch (err) {
+      get().addToast('error', `Could not load model: ${err.message}`);
+    }
+  },
+
+  async unloadOllamaModel(modelId) {
+    try {
+      await api.post('/providers/ollama/unload', { model: modelId });
+      set((s) => ({ ollamaRunningModels: s.ollamaRunningModels.filter((m) => m.name !== modelId) }));
+      get().addToast('info', `${modelId} unloaded`);
+    } catch (err) {
+      get().addToast('error', `Unload failed: ${err.message}`);
+    }
+  },
+
+  spawnFromModel(modelId) {
+    get().openDetail({ type: 'spawn', presetProvider: 'ollama', presetModel: modelId });
+  },
+
+  // ── Onboarding ────────────────────────────────────────────
+
+  async fetchOnboardingStatus() {
+    try {
+      const data = await api.get('/onboarding/status');
+      if (data?.complete) {
+        set({ onboardingComplete: true });
+        localStorage.setItem('groove:onboardingComplete', 'true');
+      }
+      return data;
+    } catch {
+      return null;
+    }
+  },
+
+  dismissOnboarding() {
+    set({ onboardingComplete: true });
+    localStorage.setItem('groove:onboardingComplete', 'true');
+    api.post('/onboarding/dismiss').catch(() => {});
+  },
+
+  // ── Provider Setup (Settings) ──────────────────────────────
+
+  async installProvider(providerId) {
+    const update = (patch) => set((s) => ({
+      providerInstallProgress: {
+        ...s.providerInstallProgress,
+        [providerId]: { ...s.providerInstallProgress[providerId], ...patch },
+      },
+    }));
+
+    update({ installing: true, percent: 0, message: 'Starting install...', error: null, done: false });
+
+    try {
+      const res = await fetch(`/api/providers/${encodeURIComponent(providerId)}/install`, {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+      });
+      if (!res.ok) {
+        const err = await res.text();
+        throw new Error(err || `Install failed (${res.status})`);
+      }
+
+      let body;
+      try {
+        body = await res.text();
+      } catch (e) {
+        throw new Error(`Failed to read response: ${e.message}`);
+      }
+
+      let lastError = null;
+      let completed = false;
+      for (const line of body.split('\n')) {
+        if (!line.trim()) continue;
+        try {
+          const ev = JSON.parse(line);
+          const isError = ev.status === 'error';
+          const isDone = ev.status === 'complete';
+          if (isError) lastError = ev.output || 'Install failed';
+          if (isDone) completed = true;
+          update({
+            percent: ev.progress ?? get().providerInstallProgress[providerId]?.percent ?? 0,
+            message: ev.output || get().providerInstallProgress[providerId]?.message,
+            error: isError ? (ev.output || 'Install failed') : null,
+            done: isDone,
+            installing: !isDone && !isError,
+          });
+        } catch { /* skip malformed line */ }
+      }
+
+      if (lastError) throw new Error(lastError);
+      if (!completed) throw new Error(body.slice(0, 500) || 'Install ended without confirmation');
+
+      update({ installing: false, percent: 100, message: 'Installed', error: null, done: true });
+      set({ _providerRefreshTick: Date.now() });
+      get().addToast('success', `${providerId} installed`);
+    } catch (err) {
+      update({ installing: false, percent: 0, message: null, error: err.message, done: false });
+      get().addToast('error', `Install failed: ${providerId}`, err.message);
+      throw err;
+    }
+  },
+
+  async loginProvider(providerId, body) {
+    try {
+      const data = await api.post(`/providers/${encodeURIComponent(providerId)}/login`, body);
+      if (data?.url && !data?.browserOpened) window.open(data.url, '_blank');
+      return data;
+    } catch (err) {
+      get().addToast('error', `Login failed`, err.message);
+      throw err;
+    }
+  },
+
+  async setProviderPath(providerId, path) {
+    try {
+      await api.post(`/providers/${encodeURIComponent(providerId)}/set-path`, { path });
+      get().addToast('success', `Custom path set for ${providerId}`);
+    } catch (err) {
+      get().addToast('error', 'Failed to set path', err.message);
+      throw err;
+    }
+  },
+
+  async verifyProvider(providerId) {
+    try {
+      const data = await api.post(`/providers/${encodeURIComponent(providerId)}/verify`);
+      return data;
+    } catch (err) {
+      get().addToast('error', `Verification failed`, err.message);
+      throw err;
+    }
+  },
+
+  async setDefaultProvider(provider, model) {
+    try {
+      await api.post('/onboarding/set-default', { provider, model });
+      get().addToast('success', `Default set to ${provider} (${model})`);
+    } catch (err) {
+      get().addToast('error', 'Failed to set default', err.message);
+      throw err;
+    }
+  },
+
+  // ── Model Lab Actions ──────────────────────────────────────
+
+  setLabParameter(key, value) {
+    const params = { ...get().labParameters, [key]: value };
+    set({ labParameters: params });
+    persistJSON('groove:labParameters', params);
+  },
+
+  setLabSystemPrompt(text) {
+    set({ labSystemPrompt: text });
+    localStorage.setItem('groove:labSystemPrompt', text);
+  },
+
+  async fetchLabRuntimes() {
+    try {
+      const raw = await api.get('/lab/runtimes');
+      const data = raw.map((rt) => ({
+        ...rt,
+        status: rt.online === true ? 'connected' : rt.online === false ? 'error' : rt.status,
+      }));
+      set({ labRuntimes: data });
+      persistJSON('groove:labRuntimes', data);
+      if (data.length > 0 && !get().labActiveRuntime) {
+        get().setLabActiveRuntime(data[0].id);
+      } else if (get().labActiveRuntime) {
+        get().fetchLabModels(get().labActiveRuntime);
+      }
+    } catch { /* backend may not have lab endpoints yet */ }
+  },
+
+  async fetchLabLocalModels() {
+    try {
+      const data = await api.get('/lab/local-models');
+      set({ labLocalModels: data });
+    } catch { set({ labLocalModels: [] }); }
+  },
+
+  async checkLlamaStatus() {
+    try {
+      const data = await api.get('/llama/status');
+      set({ labLlamaInstalled: !!data.installed });
+    } catch { set({ labLlamaInstalled: false }); }
+  },
+
+  async launchLocalModel(modelId) {
+    set({ labLaunching: modelId, labLaunchPhase: 'starting', labLaunchError: null });
+    try {
+      const result = await api.post('/lab/launch-local', { modelId });
+      const raw = await api.get('/lab/runtimes');
+      const runtimes = raw.map((rt) => ({
+        ...rt,
+        status: rt.online === true ? 'connected' : rt.online === false ? 'error' : rt.status,
+      }));
+      set({ labRuntimes: runtimes });
+      persistJSON('groove:labRuntimes', runtimes);
+      get().setLabActiveRuntime(result.runtime.id);
+      set({ labActiveModel: result.model, labLaunching: null, labLaunchPhase: 'ready' });
+      get().addToast('success', `Launched ${result.model}`);
+      setTimeout(() => { if (get().labLaunchPhase === 'ready') set({ labLaunchPhase: null }); }, 3000);
+      return result;
+    } catch (err) {
+      set({ labLaunching: null, labLaunchPhase: 'error', labLaunchError: err.message });
+      get().addToast('error', 'Failed to launch model', err.message);
+      throw err;
+    }
+  },
+
+  async addLabRuntime(runtime) {
+    try {
+      const created = await api.post('/lab/runtimes', runtime);
+      const runtimes = [...get().labRuntimes, created];
+      set({ labRuntimes: runtimes });
+      persistJSON('groove:labRuntimes', runtimes);
+      get().setLabActiveRuntime(created.id);
+      get().addToast('success', `Runtime "${runtime.name}" added`);
+      return created;
+    } catch (err) {
+      get().addToast('error', 'Failed to add runtime', err.message);
+      throw err;
+    }
+  },
+
+  async startLabRuntime(id) {
+    try {
+      get().addToast('info', 'Starting server...');
+      await api.post(`/lab/runtimes/${id}/start`);
+      await get().fetchLabRuntimes();
+      get().setLabActiveRuntime(id);
+      get().addToast('success', 'Server started');
+    } catch (err) {
+      get().addToast('error', 'Failed to start server', err.message);
+    }
+  },
+
+  async stopLabRuntime(id) {
+    try {
+      await api.post(`/lab/runtimes/${id}/stop`);
+      const runtimes = get().labRuntimes.map((r) =>
+        r.id === id ? { ...r, status: 'error', latency: null } : r,
+      );
+      set({ labRuntimes: runtimes });
+      persistJSON('groove:labRuntimes', runtimes);
+      get().addToast('success', 'Server stopped');
+    } catch (err) {
+      get().addToast('error', 'Failed to stop server', err.message);
+    }
+  },
+
+  async removeLabRuntime(id) {
+    try {
+      await api.delete(`/lab/runtimes/${id}`);
+      const runtimes = get().labRuntimes.filter((r) => r.id !== id);
+      const active = get().labActiveRuntime === id ? null : get().labActiveRuntime;
+      set({ labRuntimes: runtimes, labActiveRuntime: active, labModels: active ? get().labModels : [] });
+      persistJSON('groove:labRuntimes', runtimes);
+      get().addToast('success', 'Runtime removed');
+    } catch (err) {
+      get().addToast('error', 'Failed to remove runtime', err.message);
+    }
+  },
+
+  async testLabRuntime(id) {
+    try {
+      const result = await api.post(`/lab/runtimes/${id}/test`);
+      const runtimes = get().labRuntimes.map((r) =>
+        r.id === id ? { ...r, status: result.ok ? 'connected' : 'error', latency: result.latency } : r,
+      );
+      const updates = { labRuntimes: runtimes };
+      if (result.ok && result.models && get().labActiveRuntime === id) {
+        updates.labModels = result.models;
+      }
+      set(updates);
+      persistJSON('groove:labRuntimes', runtimes);
+      return result;
+    } catch (err) {
+      const runtimes = get().labRuntimes.map((r) =>
+        r.id === id ? { ...r, status: 'error' } : r,
+      );
+      set({ labRuntimes: runtimes });
+      persistJSON('groove:labRuntimes', runtimes);
+      return { ok: false, error: err.message };
+    }
+  },
+
+  setLabActiveRuntime(id) {
+    set({ labActiveRuntime: id, labModels: [], labActiveModel: null });
+    if (id) get().fetchLabModels(id);
+  },
+
+  setLabActiveModel(model) {
+    set({ labActiveModel: model });
+  },
+
+  async fetchLabModels(runtimeId) {
+    try {
+      const data = await api.get(`/lab/runtimes/${runtimeId}/models`);
+      const updates = { labModels: data };
+      if (data.length === 1 && !get().labActiveModel) {
+        updates.labActiveModel = data[0].id || data[0].name;
+      }
+      set(updates);
+    } catch { set({ labModels: [] }); }
+  },
+
+  newLabSession() {
+    const id = `lab-${Date.now()}`;
+    const session = { id, messages: [], createdAt: Date.now() };
+    set((s) => ({
+      labSessions: [session, ...s.labSessions],
+      labActiveSession: id,
+      labMetrics: {
+        ttft: null, tokensPerSec: null, tokensPerSecHistory: [], ttftHistory: [],
+        memory: null, peakMemory: null, totalTokens: 0, promptTokens: 0, completionTokens: 0,
+        generationTime: null, generationCount: 0, sessionStartTime: null,
+      },
+    }));
+    return id;
+  },
+
+  loadLabSession(id) {
+    set({ labActiveSession: id });
+  },
+
+  async sendLabMessage(text) {
+    const st = get();
+    if (st.labStreaming) return;
+    let sessionId = st.labActiveSession;
+    if (!sessionId) sessionId = get().newLabSession();
+
+    const userMsg = { role: 'user', content: text, timestamp: Date.now() };
+    set((s) => {
+      const sessions = s.labSessions.map((sess) =>
+        sess.id === sessionId ? { ...sess, messages: [...sess.messages, userMsg] } : sess,
+      );
+      return { labSessions: sessions, labStreaming: true };
+    });
+
+    const assistantMsg = { role: 'assistant', content: '', timestamp: Date.now(), metrics: null };
+    set((s) => {
+      const sessions = s.labSessions.map((sess) =>
+        sess.id === sessionId ? { ...sess, messages: [...sess.messages, assistantMsg] } : sess,
+      );
+      return { labSessions: sessions };
+    });
+
+    const abortController = new AbortController();
+    set({ labAbortController: abortController });
+
+    const startTime = performance.now();
+    let firstTokenTime = null;
+    let tokenCount = 0;
+
+    try {
+      const p = st.labParameters;
+      const parameters = {};
+      if (p.temperature !== undefined) parameters.temperature = p.temperature;
+      if (p.topP !== undefined) parameters.top_p = p.topP;
+      if (p.topK !== undefined) parameters.top_k = p.topK;
+      if (p.minP !== undefined && p.minP > 0) parameters.min_p = p.minP;
+      if (p.repeatPenalty !== undefined) parameters.repeat_penalty = p.repeatPenalty;
+      if (p.maxTokens !== undefined) parameters.max_tokens = p.maxTokens;
+      if (p.frequencyPenalty !== undefined) parameters.frequency_penalty = p.frequencyPenalty;
+      if (p.presencePenalty !== undefined) parameters.presence_penalty = p.presencePenalty;
+      parameters.enable_thinking = !!p.thinking;
+      if (p.seed != null) parameters.seed = p.seed;
+      if (p.stopSequences?.length) parameters.stop = p.stopSequences;
+      if (p.jsonMode) parameters.response_format = { type: 'json_object' };
+
+      const messages = [];
+      if (st.labSystemPrompt) messages.push({ role: 'system', content: st.labSystemPrompt });
+      const sessionMsgs = get().labSessions.find((s) => s.id === sessionId)?.messages || [];
+      for (const m of sessionMsgs) {
+        if (m.role === 'assistant' && !m.content) continue;
+        messages.push({ role: m.role, content: m.content });
+      }
+
+      const body = {
+        runtimeId: st.labActiveRuntime,
+        model: st.labActiveModel,
+        messages,
+        parameters,
+        sessionId,
+      };
+
+      const res = await fetch('/api/lab/inference', {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify(body),
+        signal: abortController.signal,
+      });
+
+      if (!res.ok) {
+        let errMsg;
+        try { errMsg = (await res.json()).error || `HTTP ${res.status}`; } catch { errMsg = `HTTP ${res.status}`; }
+        throw new Error(errMsg);
+      }
+
+      const reader = res.body.getReader();
+      const decoder = new TextDecoder();
+      let buffer = '';
+      let fullContent = '';
+      let fullReasoning = '';
+
+      while (true) {
+        const { done, value } = await reader.read();
+        if (done) break;
+        buffer += decoder.decode(value, { stream: true });
+        const lines = buffer.split('\n');
+        buffer = lines.pop() || '';
+
+        for (const line of lines) {
+          if (!line.startsWith('data: ')) continue;
+          const payload = line.slice(6);
+          if (payload === '[DONE]') continue;
+          try {
+            const parsed = JSON.parse(payload);
+
+            // Support both raw OpenAI format (piped) and legacy wrapper format
+            const delta = parsed.choices?.[0]?.delta;
+            const reasoningText = delta?.reasoning_content || delta?.reasoning || (parsed.type === 'reasoning' ? parsed.content : null);
+            let contentText = delta?.content || (parsed.type === 'token' ? parsed.content : null);
+
+            // Raw fallback: if no known field matched, extract any string from delta
+            if (!reasoningText && !contentText && delta) {
+              for (const v of Object.values(delta)) {
+                if (typeof v === 'string' && v) { contentText = v; break; }
+              }
+            }
+            // Last resort: if the parsed object itself has text but no choices wrapper
+            if (!reasoningText && !contentText && !delta && parsed.response) {
+              contentText = typeof parsed.response === 'string' ? parsed.response : null;
+            }
+            if (!reasoningText && !contentText && !delta && typeof parsed.text === 'string') {
+              contentText = parsed.text;
+            }
+            if (!reasoningText && !contentText && !delta && typeof parsed.output === 'string') {
+              contentText = parsed.output;
+            }
+
+            if (reasoningText) {
+              if (!firstTokenTime) firstTokenTime = performance.now();
+              tokenCount++;
+              fullReasoning += reasoningText;
+              set((s) => {
+                const sessions = s.labSessions.map((sess) => {
+                  if (sess.id !== sessionId) return sess;
+                  const msgs = [...sess.messages];
+                  msgs[msgs.length - 1] = { ...msgs[msgs.length - 1], reasoning: fullReasoning };
+                  return { ...sess, messages: msgs };
+                });
+                return { labSessions: sessions };
+              });
+            }
+            if (contentText) {
+              if (!firstTokenTime) firstTokenTime = performance.now();
+              tokenCount++;
+              fullContent += contentText;
+              set((s) => {
+                const sessions = s.labSessions.map((sess) => {
+                  if (sess.id !== sessionId) return sess;
+                  const msgs = [...sess.messages];
+                  msgs[msgs.length - 1] = { ...msgs[msgs.length - 1], content: fullContent };
+                  return { ...sess, messages: msgs };
+                });
+                return { labSessions: sessions };
+              });
+            }
+
+            // Handle done event (legacy wrapper) or finish_reason (raw OpenAI)
+            if (parsed.type === 'done' && parsed.metrics) {
+              const elapsed = performance.now() - startTime;
+              const ttft = firstTokenTime ? firstTokenTime - startTime : null;
+              const tps = tokenCount > 0 && elapsed > 0 ? (tokenCount / (elapsed / 1000)) : null;
+              const msgMetrics = { ttft, tokensPerSec: tps, tokens: tokenCount, generationTime: elapsed, ...parsed.metrics };
+
+              set((s) => {
+                const tpsHist = [...s.labMetrics.tokensPerSecHistory, tps].slice(-20);
+                const ttftHist = [...s.labMetrics.ttftHistory, ttft].filter((v) => v != null).slice(-20);
+                const mem = parsed.metrics.memoryUsage || s.labMetrics.memory;
+                const sessions = s.labSessions.map((sess) => {
+                  if (sess.id !== sessionId) return sess;
+                  const msgs = [...sess.messages];
+                  msgs[msgs.length - 1] = { ...msgs[msgs.length - 1], metrics: msgMetrics };
+                  return { ...sess, messages: msgs };
+                });
+                return {
+                  labSessions: sessions,
+                  labMetrics: {
+                    ...s.labMetrics,
+                    ttft, tokensPerSec: tps, tokensPerSecHistory: tpsHist, ttftHistory: ttftHist,
+                    memory: mem, peakMemory: Math.max(mem || 0, s.labMetrics.peakMemory || 0) || null,
+                    totalTokens: s.labMetrics.totalTokens + (parsed.metrics.totalTokens || tokenCount),
+                    promptTokens: s.labMetrics.promptTokens + (parsed.metrics.promptTokens || 0),
+                    completionTokens: s.labMetrics.completionTokens + (parsed.metrics.completionTokens || tokenCount),
+                    generationTime: parsed.metrics.generationTime || elapsed,
+                    generationCount: s.labMetrics.generationCount + 1,
+                    sessionStartTime: s.labMetrics.sessionStartTime || Date.now(),
+                  },
+                };
+              });
+            }
+            if (parsed.type === 'error') {
+              throw new Error(parsed.error || 'Inference error');
+            }
+          } catch (e) {
+            if (e.message && e.message !== 'Inference error' && !e.message.startsWith('HTTP ')) continue;
+            throw e;
+          }
+        }
+      }
+
+      // Strip <think> tags from content — some models embed reasoning in content field
+      if (fullContent && fullContent.includes('<think>')) {
+        const stripped = fullContent.replace(/<think>[\s\S]*?<\/think>/g, '').trim();
+        const thinkMatch = fullContent.match(/<think>([\s\S]*?)<\/think>/);
+        if (thinkMatch && !fullReasoning) fullReasoning = thinkMatch[1].trim();
+        // If stripping left nothing, check for unclosed <think> (model still mid-thought)
+        if (stripped) {
+          fullContent = stripped;
+        } else if (fullContent.includes('<think>') && !fullContent.includes('</think>')) {
+          const afterTag = fullContent.replace(/<think>/, '').trim();
+          if (!fullReasoning) fullReasoning = afterTag;
+          fullContent = '';
+        }
+        set((s) => {
+          const sessions = s.labSessions.map((sess) => {
+            if (sess.id !== sessionId) return sess;
+            const msgs = [...sess.messages];
+            msgs[msgs.length - 1] = { ...msgs[msgs.length - 1], content: fullContent || undefined, reasoning: fullReasoning || undefined };
+            return { ...sess, messages: msgs };
+          });
+          return { labSessions: sessions };
+        });
+      }
+
+      // If stream ended with reasoning but no content, promote reasoning as the response
+      if (fullReasoning && !fullContent) {
+        fullContent = fullReasoning;
+        set((s) => {
+          const sessions = s.labSessions.map((sess) => {
+            if (sess.id !== sessionId) return sess;
+            const msgs = [...sess.messages];
+            msgs[msgs.length - 1] = { ...msgs[msgs.length - 1], content: fullContent };
+            return { ...sess, messages: msgs };
+          });
+          return { labSessions: sessions };
+        });
+      }
+
+      // If stream ended with zero content at all, surface a fallback message
+      if (!fullContent && !fullReasoning) {
+        fullContent = '[Model returned an empty response — try a different prompt or check server logs]';
+        set((s) => {
+          const sessions = s.labSessions.map((sess) => {
+            if (sess.id !== sessionId) return sess;
+            const msgs = [...sess.messages];
+            msgs[msgs.length - 1] = { ...msgs[msgs.length - 1], content: fullContent };
+            return { ...sess, messages: msgs };
+          });
+          return { labSessions: sessions };
+        });
+      }
+
+      // Compute final metrics from client-side timing
+      const elapsed = performance.now() - startTime;
+      const ttft = firstTokenTime ? firstTokenTime - startTime : null;
+      const tps = tokenCount > 0 && elapsed > 0 ? (tokenCount / (elapsed / 1000)) : null;
+      if (tokenCount > 0) {
+        set((s) => {
+          const tpsHist = [...s.labMetrics.tokensPerSecHistory, tps].slice(-20);
+          const ttftHist = [...s.labMetrics.ttftHistory, ttft].filter((v) => v != null).slice(-20);
+          const sessions = s.labSessions.map((sess) => {
+            if (sess.id !== sessionId) return sess;
+            const msgs = [...sess.messages];
+            const last = msgs[msgs.length - 1];
+            if (!last?.metrics) {
+              msgs[msgs.length - 1] = { ...last, metrics: { ttft, tokensPerSec: tps, tokens: tokenCount, generationTime: elapsed } };
+            }
+            return { ...sess, messages: msgs };
+          });
+          return {
+            labSessions: sessions,
+            labMetrics: {
+              ...s.labMetrics, ttft, tokensPerSec: tps,
+              tokensPerSecHistory: tpsHist, ttftHistory: ttftHist,
+              totalTokens: s.labMetrics.totalTokens + tokenCount,
+              completionTokens: s.labMetrics.completionTokens + tokenCount,
+              generationTime: elapsed,
+              generationCount: s.labMetrics.generationCount + 1,
+              sessionStartTime: s.labMetrics.sessionStartTime || Date.now(),
+            },
+          };
+        });
+      }
+    } catch (err) {
+      if (err.name === 'AbortError') {
+        // User cancelled — keep whatever content was already streamed
+      } else {
+        set((s) => {
+          const sessions = s.labSessions.map((sess) => {
+            if (sess.id !== sessionId) return sess;
+            const msgs = [...sess.messages];
+            msgs[msgs.length - 1] = { ...msgs[msgs.length - 1], content: `Error: ${err.message}`, error: true };
+            return { ...sess, messages: msgs };
+          });
+          return { labSessions: sessions };
+        });
+      }
+    } finally {
+      set({ labStreaming: false, labAbortController: null });
+    }
+  },
+
+  stopLabInference() {
+    const ctrl = get().labAbortController;
+    if (ctrl) ctrl.abort();
+  },
+
+  saveLabPreset(name) {
+    const st = get();
+    const preset = {
+      id: `preset-${Date.now()}`,
+      name,
+      parameters: { ...st.labParameters },
+      systemPrompt: st.labSystemPrompt,
+      runtimeId: st.labActiveRuntime,
+      model: st.labActiveModel,
+      createdAt: Date.now(),
+    };
+    const presets = [...st.labPresets.filter((p) => p.name !== name), preset];
+    set({ labPresets: presets, labActivePreset: preset.id });
+    persistJSON('groove:labPresets', presets);
+    get().addToast('success', `Preset "${name}" saved`);
+    return preset;
+  },
+
+  loadLabPreset(id) {
+    const preset = get().labPresets.find((p) => p.id === id);
+    if (!preset) return;
+    const defaults = {
+      temperature: 0.7, topP: 0.9, topK: 40, minP: 0, repeatPenalty: 1.1,
+      maxTokens: 2048, frequencyPenalty: 0, presencePenalty: 0,
+      thinking: false, seed: null, stopSequences: [], jsonMode: false,
+    };
+    const merged = { ...defaults, ...preset.parameters };
+    const updates = {
+      labParameters: merged,
+      labSystemPrompt: preset.systemPrompt || '',
+      labActivePreset: id,
+    };
+    if (preset.model) updates.labActiveModel = preset.model;
+    set(updates);
+    persistJSON('groove:labParameters', merged);
+    if (preset.systemPrompt !== undefined) localStorage.setItem('groove:labSystemPrompt', preset.systemPrompt);
+  },
+
+  deleteLabPreset(id) {
+    const presets = get().labPresets.filter((p) => p.id !== id);
+    set({ labPresets: presets, labActivePreset: get().labActivePreset === id ? null : get().labActivePreset });
+    persistJSON('groove:labPresets', presets);
+    get().addToast('success', 'Preset deleted');
+  },
+
+  async launchLabAssistant(backend, model) {
+    const existing = get().labAssistantAgentId;
+    if (existing) {
+      const agent = get().agents.find((a) => a.id === existing);
+      if (agent && agent.status === 'running') {
+        set({ labAssistantMode: true });
+        return;
+      }
+    }
+    try {
+      const body = { backend };
+      if (model) body.model = { id: model.id, filename: model.filename, parameters: model.parameters, quantization: model.quantization };
+      const data = await api.post('/lab/assistant', body);
+      localStorage.setItem('groove:labAssistantAgentId', data.agentId);
+      localStorage.setItem('groove:labAssistantBackend', backend);
+      set({ labAssistantAgentId: data.agentId, labAssistantMode: true, labAssistantBackend: backend });
+      get().addToast('info', `Lab Assistant started for ${backend}`);
+    } catch (err) {
+      get().addToast('error', 'Failed to start assistant', err.message);
+    }
+  },
+
+  dismissLabAssistant() {
+    set({ labAssistantMode: false });
+  },
+
+  clearLabAssistant() {
+    const id = get().labAssistantAgentId;
+    if (id) api.delete(`/agents/${encodeURIComponent(id)}`).catch(() => {});
+    localStorage.removeItem('groove:labAssistantAgentId');
+    localStorage.removeItem('groove:labAssistantBackend');
+    set({ labAssistantAgentId: null, labAssistantMode: false, labAssistantBackend: null });
+  },
+
+  setLabAssistantMode(mode) {
+    set({ labAssistantMode: mode });
+  },
+
+  async onLabAssistantComplete() {
+    const prevIds = new Set(get().labRuntimes.map((r) => r.id));
+    try {
+      const raw = await api.get('/lab/runtimes');
+      const data = raw.map((rt) => ({
+        ...rt,
+        status: rt.online === true ? 'connected' : rt.online === false ? 'error' : rt.status,
+      }));
+      set({ labRuntimes: data });
+      persistJSON('groove:labRuntimes', data);
+      const newRuntime = data.find((r) => !prevIds.has(r.id));
+      if (newRuntime) {
+        set({ labActiveRuntime: newRuntime.id, labModels: [], labActiveModel: null });
+        try {
+          const models = await api.get(`/lab/runtimes/${newRuntime.id}/models`);
+          set({ labModels: models });
+          if (models.length > 0) set({ labActiveModel: models[0].id || models[0].name });
+        } catch { /* models may not be available yet */ }
+      }
+    } catch { /* ignore */ }
+  },
+});