Add managed local model runtime

app.js

@@ -159,6 +159,22 @@ async function testRuntime() {
   }
 }
 
+async function startRuntime() {
+  const model = state.selected;
+  if (!model) return appendLog('RUNTIME', 'Select a model pack first.', 'warning');
+  try {
+    const response = await fetch('http://127.0.0.1:4777/api/models/start', {
+      method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify({ id: model.id })
+    });
+    const result = await response.json();
+    if (!response.ok) throw new Error(result.error || 'runtime start failed');
+    appendLog('RUNTIME', `${model.name} is starting at ${result.endpoint}.`);
+    setTimeout(testRuntime, 1500);
+  } catch (error) {
+    appendLog('RUNTIME', `Local daemon unavailable: ${error.message}. Start daemon/server.mjs first.`, 'warning');
+  }
+}
+
 function sendChat() {
   const input = $('#input');
   const value = input.value.trim();
@@ -237,7 +253,7 @@ async function boot() {
   openFile('agent.ts');
   document.querySelectorAll('[data-file]').forEach(button => button.onclick = () => openFile(button.dataset.file));
   $('#review-button').onclick = runReview;
-  $('#connection-button').onclick = testRuntime;
+  $('#connection-button').onclick = startRuntime;
   $('#send-button').onclick = sendChat;
   $('#node-button').onclick = () => { localNodeId(); renderNode(); appendLog('NODE', 'Local identity created. No network connection or private data was shared.'); };
   $('#case-button').onclick = () => {

daemon/model-manager.mjs

@@ -0,0 +1,59 @@
+import { promises as fs } from 'node:fs';
+import path from 'node:path';
+import { spawn } from 'node:child_process';
+
+export class ModelManager {
+  constructor({ manifestPath, modelDir, binaryPath, spawnProcess = spawn } = {}) {
+    this.manifestPath = manifestPath;
+    this.modelDir = path.resolve(modelDir);
+    this.binaryPath = binaryPath;
+    this.spawnProcess = spawnProcess;
+    this.models = new Map();
+    this.processes = new Map();
+  }
+
+  async load() {
+    const manifest = JSON.parse(await fs.readFile(this.manifestPath, 'utf8'));
+    this.models = new Map(manifest.models.map(model => [model.id, model]));
+    return manifest;
+  }
+
+  status() {
+    return [...this.models.values()].map(model => ({
+      id: model.id,
+      name: model.name,
+      status: this.processes.has(model.id) ? 'running' : model.status,
+      endpoint: model.endpoint
+    }));
+  }
+
+  async start(id) {
+    const model = this.models.get(id);
+    if (!model) throw new Error('model is not allowlisted');
+    if (this.processes.has(id)) return { id, status: 'running', endpoint: model.endpoint };
+    if (!this.binaryPath) throw new Error('llama-server binary is not configured');
+    await fs.access(this.binaryPath).catch(() => { throw new Error('llama-server binary is unavailable'); });
+    const file = path.resolve(this.modelDir, path.basename(model.artifact_uri.replace('local://', '')));
+    if (!file.startsWith(`${this.modelDir}${path.sep}`)) throw new Error('model path escaped model directory');
+    await fs.access(file);
+    await this.stopAll();
+    const endpoint = new URL(model.endpoint);
+    const args = ['-m', file, '--host', '127.0.0.1', '--port', String(endpoint.port || 8080), '--ctx-size', String(model.context_tokens || 2048), '--threads', '4', '--parallel', '1', '--log-disable'];
+    const child = this.spawnProcess(this.binaryPath, args, { stdio: 'ignore' });
+    this.processes.set(id, child);
+    child.once('exit', () => this.processes.delete(id));
+    return { id, status: 'starting', endpoint: model.endpoint };
+  }
+
+  async stop(id) {
+    const child = this.processes.get(id);
+    if (!child) return { id, status: 'stopped' };
+    child.kill('SIGTERM');
+    this.processes.delete(id);
+    return { id, status: 'stopped' };
+  }
+
+  async stopAll() {
+    for (const id of [...this.processes.keys()]) await this.stop(id);
+  }
+}

daemon/server.mjs

@@ -2,10 +2,15 @@ import http from 'node:http';
 import { execFile } from 'node:child_process';
 import { promises as fs } from 'node:fs';
 import path from 'node:path';
+import { ModelManager } from './model-manager.mjs';
 
 const HOST = '127.0.0.1';
 const PORT = Number(process.env.AIDE_DAEMON_PORT || 4777);
 const WORKSPACE = path.resolve(process.env.AIDE_WORKSPACE || process.cwd());
+const MODEL_DIR = path.resolve(process.env.AIDE_MODEL_DIR || path.join(WORKSPACE, 'models'));
+const MANIFEST = path.join(WORKSPACE, 'models', 'manifest.json');
+const modelManager = new ModelManager({ manifestPath: MANIFEST, modelDir: MODEL_DIR, binaryPath: process.env.AIDE_LLAMA_SERVER || '/root/runtime/llama-b10333/llama-server' });
+await modelManager.load().catch(() => {});
 
 function json(response, status, body) {
   const payload = JSON.stringify(body);
@@ -13,12 +18,21 @@ function json(response, status, body) {
     'Content-Type': 'application/json; charset=utf-8',
     'Content-Length': Buffer.byteLength(payload),
     'Access-Control-Allow-Origin': 'http://127.0.0.1:4173',
-    'Access-Control-Allow-Methods': 'GET, OPTIONS',
+    'Access-Control-Allow-Methods': 'GET, POST, OPTIONS',
     'Access-Control-Allow-Headers': 'Content-Type'
   });
   response.end(payload);
 }
 
+async function body(request) {
+  let data = '';
+  for await (const chunk of request) {
+    data += chunk;
+    if (Buffer.byteLength(data) > 16 * 1024) throw new Error('request too large');
+  }
+  return data ? JSON.parse(data) : {};
+}
+
 function runGit(args) {
   return new Promise((resolve, reject) => {
     execFile('git', args, { cwd: WORKSPACE, timeout: 5000, maxBuffer: 256 * 1024 }, (error, stdout, stderr) => {
@@ -52,6 +66,16 @@ const server = http.createServer(async (request, response) => {
         return json(response, 200, { workspace: WORKSPACE, status: '', unavailable: error.message });
       }
     }
+    if (request.method === 'GET' && request.url === '/api/models/status') {
+      return json(response, 200, { models: modelManager.status() });
+    }
+    if (request.method === 'POST' && request.url === '/api/models/start') {
+      return json(response, 200, await modelManager.start((await body(request)).id));
+    }
+    if (request.method === 'POST' && request.url === '/api/models/stop') {
+      const id = (await body(request)).id;
+      return json(response, 200, id ? await modelManager.stop(id) : (await modelManager.stopAll(), { status: 'stopped' }));
+    }
     return json(response, 404, { error: 'not found' });
   } catch (error) {
     return json(response, 500, { error: 'local daemon error' });

daemon/test-model-manager.mjs

@@ -0,0 +1,16 @@
+import assert from 'node:assert/strict';
+import { mkdtemp, writeFile, mkdir } from 'node:fs/promises';
+import { tmpdir } from 'node:os';
+import path from 'node:path';
+import { ModelManager } from './model-manager.mjs';
+
+const root = await mkdtemp(path.join(tmpdir(), 'aide-manager-'));
+await mkdir(path.join(root, 'models'));
+await writeFile(path.join(root, 'models', 'safe.gguf'), 'test');
+await writeFile(path.join(root, 'models', 'manifest.json'), JSON.stringify({ models: [{ id: 'safe', name: 'Safe', status: 'pending', endpoint: 'http://127.0.0.1:9001/v1', artifact_uri: 'local://safe.gguf', context_tokens: 2048 }] }));
+const manager = new ModelManager({ manifestPath: path.join(root, 'models', 'manifest.json'), modelDir: path.join(root, 'models'), binaryPath: '/missing' });
+await manager.load();
+assert.equal(manager.status()[0].status, 'pending');
+await assert.rejects(manager.start('unknown'), /not allowlisted/);
+await assert.rejects(manager.start('safe'), /llama-server binary/);
+console.log('model manager test passed');

index.html

@@ -79,7 +79,7 @@
       <div class="lane-grid" id="lane-grid"></div>
       <div class="runtime-row"><span class="dot green"></span><span>Local adapter</span><b id="runtime-name">OpenAI-compatible HTTP</b></div>
       <div id="collab-log" class="collab-log"><div class="empty-state">Cross-chat is idle.<br>Start a bounded review to create a research, build, and verify run.</div></div>
-      <div class="agent-actions"><button id="review-button" class="primary">START BOUNDED REVIEW</button><button id="connection-button" class="secondary">TEST LOCAL RUNTIME</button></div>
+      <div class="agent-actions"><button id="review-button" class="primary">START BOUNDED REVIEW</button><button id="connection-button" class="secondary">START / TEST RUNTIME</button></div>
       <div class="chat-row"><input id="input" placeholder="Ask the local lanes..." aria-label="Ask the local lanes"><button id="send-button">SEND</button></div>
       <div id="chat" class="chat" aria-live="polite"></div>
     </aside>

package.json

@@ -5,7 +5,7 @@
   "description": "Local-first sovereign development workbench",
   "type": "module",
   "scripts": {
-    "test": "node tests/smoke.mjs && node harness/test-orchestrator.mjs",
+    "test": "node tests/smoke.mjs && node harness/test-orchestrator.mjs && node daemon/test-model-manager.mjs",
     "check": "node --check app.js && node --check daemon/server.mjs && node --check harness/orchestrator.mjs && node --check harness/checks.mjs",
     "veritas": "node harness/run-veritas.mjs"
   },