@gns-foundation/hive-worker 0.5.0 → 0.5.1

package/dist/mobydb_hooks.d.ts

@@ -1,4 +1,4 @@
-import { type MobyDB } from './mobydb';
+import { type MobyDB } from './mobydb.js';
 /**
  * Initialize MobyDB for this worker.
  * Call once at worker startup after identity is loaded.

package/dist/mobydb_hooks.js

@@ -5,7 +5,7 @@
 // Hooks MobyDB into every compute operation.
 // Call initWorkerMobyDB() at worker startup.
 // ============================================================
-import { initMobyDB, getMobyDB, currentEpoch } from './mobydb';
+import { initMobyDB, getMobyDB, currentEpoch } from './mobydb.js';
 let sealInterval = null;
 let syncInterval = null;
 // ─── Central API ────────────────────────────

package/dist/mobydb_hooks.js.map

@@ -1 +1 @@
-{"version":3,"file":"mobydb_hooks.js","sourceRoot":"","sources":["../src/mobydb_hooks.ts"],"names":[],"mappings":"AAAA,+DAA+D;AAC/D,qCAAqC;AACrC,iDAAiD;AACjD,EAAE;AACF,6CAA6C;AAC7C,6CAA6C;AAC7C,+DAA+D;AAE/D,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAe,MAAM,UAAU,CAAC;AAE5E,IAAI,YAAY,GAA0B,IAAI,CAAC;AAC/C,IAAI,YAAY,GAA0B,IAAI,CAAC;AAE/C,+CAA+C;AAC/C,MAAM,WAAW,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,+CAA+C,CAAC;AAEnG,+CAA+C;AAE/C;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAAC,QAAgB,EAAE,MAAc;IAC/D,MAAM,EAAE,GAAG,UAAU,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IAExC,uCAAuC;IACvC,IAAI,CAAC,YAAY,EAAE,CAAC;QAClB,YAAY,GAAG,WAAW,CAAC,GAAG,EAAE;YAC9B,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;YACzB,IAAI,CAAC,IAAI;gBAAE,OAAO;YAClB,MAAM,KAAK,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC;YAC9B,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACrB,OAAO,CAAC,GAAG,CAAC,kBAAkB,KAAK,CAAC,MAAM,WAAW,CAAC,CAAC;YACzD,CAAC;QACH,CAAC,EAAE,MAAM,CAAC,CAAC;IACb,CAAC;IAED,2CAA2C;IAC3C,IAAI,CAAC,YAAY,EAAE,CAAC;QAClB,YAAY,GAAG,WAAW,CAAC,GAAG,EAAE;YAC9B,aAAa,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE;gBAC1B,OAAO,CAAC,KAAK,CAAC,4BAA4B,EAAE,GAAG,CAAC,OAAO,CAAC,CAAC;YAC3D,CAAC,CAAC,CAAC;QACL,CAAC,EAAE,MAAM,CAAC,CAAC;IACb,CAAC;IAED,OAAO,CAAC,GAAG,CAAC,oDAAoD,CAAC,CAAC;IAClE,OAAO,EAAE,CAAC;AACZ,CAAC;AAED,+CAA+C;AAE/C;;;GAGG;AACH,MAAM,UAAU,cAAc,CAAC,IAU9B;IACC,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;IACzB,IAAI,CAAC,IAAI;QAAE,OAAO,IAAI,CAAC;IAEvB,MAAM,MAAM,GAAG,IAAI,CAAC,cAAc,CAAC,IAAI,CAAC,CAAC;IACzC,OAAO,MAAM,CAAC,WAAW,CAAC;AAC5B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,IAU/B;IACC,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;IACzB,IAAI,CAAC,IAAI;QAAE,OAAO,IAAI,CAAC;IAEvB,MAAM,MAAM,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;IACpC,OAAO,MAAM,CAAC,WAAW,CAAC;AAC5B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,IAO/B;IACC,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;IACzB,IAAI,CAAC,IAAI;QAAE,OAAO,IAAI,CAAC;IAEvB,MAAM,MAAM,GAAG,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,CAAC;IAC1C,OAAO,MAAM,CAAC,WAAW,CAAC;AAC5B,CAAC;AAED,+CAA+C;AAE/C;;;GAGG;AACH,KAAK,UAAU,aAAa;IAC1B,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;IACzB,IAAI,CAAC,IAAI;QAAE,OAAO;IAElB,MAAM,UAAU,GAAG,IAAI,CAAC,cAAc,CAAC,EAAE,CAAC,CAAC;IAC3C,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO;IAEpC,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IACnE,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO;IAEjC,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,WAAW,cAAc,EAAE;YACpD,MAAM,EAAE,MAAM;YACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;YAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,OAAO,EAAE,CAAC;SAClC,CAAC,CAAC;QAEH,IAAI,GAAG,CAAC,EAAE,EAAE,CAAC;YACX,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,GAAG,CAAC,aAAa,OAAO,CAAC,MAAM,qBAAqB,CAAC,CAAC;QAChE,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACP,wBAAwB;IAC1B,CAAC;AACH,CAAC;AAED,+CAA+C;AAE/C,MAAM,UAAU,eAAe;IAC7B,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;IACzB,IAAI,CAAC,IAAI;QAAE,OAAO,IAAI,CAAC;IAEvB,OAAO;QACL,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,OAAO,EAAE,IAAI,CAAC,WAAW;QACzB,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE;QACnB,KAAK,EAAE,YAAY,EAAE;KACtB,CAAC;AACJ,CAAC;AAED,+CAA+C;AAE/C,MAAM,UAAU,cAAc;IAC5B,IAAI,YAAY,EAAE,CAAC;QAAC,aAAa,CAAC,YAAY,CAAC,CAAC;QAAC,YAAY,GAAG,IAAI,CAAC;IAAC,CAAC;IACvE,IAAI,YAAY,EAAE,CAAC;QAAC,aAAa,CAAC,YAAY,CAAC,CAAC;QAAC,YAAY,GAAG,IAAI,CAAC;IAAC,CAAC;IACvE,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;IACzB,IAAI,IAAI,EAAE,CAAC;QACT,6BAA6B;QAC7B,IAAI,CAAC,QAAQ,EAAE,CAAC;QAChB,IAAI,CAAC,KAAK,EAAE,CAAC;IACf,CAAC;AACH,CAAC"}
+{"version":3,"file":"mobydb_hooks.js","sourceRoot":"","sources":["../src/mobydb_hooks.ts"],"names":[],"mappings":"AAAA,+DAA+D;AAC/D,qCAAqC;AACrC,iDAAiD;AACjD,EAAE;AACF,6CAA6C;AAC7C,6CAA6C;AAC7C,+DAA+D;AAE/D,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAe,MAAM,aAAa,CAAC;AAE/E,IAAI,YAAY,GAA0B,IAAI,CAAC;AAC/C,IAAI,YAAY,GAA0B,IAAI,CAAC;AAE/C,+CAA+C;AAC/C,MAAM,WAAW,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,+CAA+C,CAAC;AAEnG,+CAA+C;AAE/C;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAAC,QAAgB,EAAE,MAAc;IAC/D,MAAM,EAAE,GAAG,UAAU,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IAExC,uCAAuC;IACvC,IAAI,CAAC,YAAY,EAAE,CAAC;QAClB,YAAY,GAAG,WAAW,CAAC,GAAG,EAAE;YAC9B,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;YACzB,IAAI,CAAC,IAAI;gBAAE,OAAO;YAClB,MAAM,KAAK,GAAG,IAAI,CAAC,QAAQ,EAAE,CAAC;YAC9B,IAAI,KAAK,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBACrB,OAAO,CAAC,GAAG,CAAC,kBAAkB,KAAK,CAAC,MAAM,WAAW,CAAC,CAAC;YACzD,CAAC;QACH,CAAC,EAAE,MAAM,CAAC,CAAC;IACb,CAAC;IAED,2CAA2C;IAC3C,IAAI,CAAC,YAAY,EAAE,CAAC;QAClB,YAAY,GAAG,WAAW,CAAC,GAAG,EAAE;YAC9B,aAAa,EAAE,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE;gBAC1B,OAAO,CAAC,KAAK,CAAC,4BAA4B,EAAE,GAAG,CAAC,OAAO,CAAC,CAAC;YAC3D,CAAC,CAAC,CAAC;QACL,CAAC,EAAE,MAAM,CAAC,CAAC;IACb,CAAC;IAED,OAAO,CAAC,GAAG,CAAC,oDAAoD,CAAC,CAAC;IAClE,OAAO,EAAE,CAAC;AACZ,CAAC;AAED,+CAA+C;AAE/C;;;GAGG;AACH,MAAM,UAAU,cAAc,CAAC,IAU9B;IACC,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;IACzB,IAAI,CAAC,IAAI;QAAE,OAAO,IAAI,CAAC;IAEvB,MAAM,MAAM,GAAG,IAAI,CAAC,cAAc,CAAC,IAAI,CAAC,CAAC;IACzC,OAAO,MAAM,CAAC,WAAW,CAAC;AAC5B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,IAU/B;IACC,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;IACzB,IAAI,CAAC,IAAI;QAAE,OAAO,IAAI,CAAC;IAEvB,MAAM,MAAM,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;IACpC,OAAO,MAAM,CAAC,WAAW,CAAC;AAC5B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,IAO/B;IACC,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;IACzB,IAAI,CAAC,IAAI;QAAE,OAAO,IAAI,CAAC;IAEvB,MAAM,MAAM,GAAG,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,CAAC;IAC1C,OAAO,MAAM,CAAC,WAAW,CAAC;AAC5B,CAAC;AAED,+CAA+C;AAE/C;;;GAGG;AACH,KAAK,UAAU,aAAa;IAC1B,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;IACzB,IAAI,CAAC,IAAI;QAAE,OAAO;IAElB,MAAM,UAAU,GAAG,IAAI,CAAC,cAAc,CAAC,EAAE,CAAC,CAAC;IAC3C,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO;IAEpC,MAAM,OAAO,GAAG,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IACnE,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO;IAEjC,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,WAAW,cAAc,EAAE;YACpD,MAAM,EAAE,MAAM;YACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;YAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,OAAO,EAAE,CAAC;SAClC,CAAC,CAAC;QAEH,IAAI,GAAG,CAAC,EAAE,EAAE,CAAC;YACX,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,CAAC;YAC5B,OAAO,CAAC,GAAG,CAAC,aAAa,OAAO,CAAC,MAAM,qBAAqB,CAAC,CAAC;QAChE,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACP,wBAAwB;IAC1B,CAAC;AACH,CAAC;AAED,+CAA+C;AAE/C,MAAM,UAAU,eAAe;IAC7B,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;IACzB,IAAI,CAAC,IAAI;QAAE,OAAO,IAAI,CAAC;IAEvB,OAAO;QACL,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,OAAO,EAAE,IAAI,CAAC,WAAW;QACzB,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE;QACnB,KAAK,EAAE,YAAY,EAAE;KACtB,CAAC;AACJ,CAAC;AAED,+CAA+C;AAE/C,MAAM,UAAU,cAAc;IAC5B,IAAI,YAAY,EAAE,CAAC;QAAC,aAAa,CAAC,YAAY,CAAC,CAAC;QAAC,YAAY,GAAG,IAAI,CAAC;IAAC,CAAC;IACvE,IAAI,YAAY,EAAE,CAAC;QAAC,aAAa,CAAC,YAAY,CAAC,CAAC;QAAC,YAAY,GAAG,IAAI,CAAC;IAAC,CAAC;IACvE,MAAM,IAAI,GAAG,SAAS,EAAE,CAAC;IACzB,IAAI,IAAI,EAAE,CAAC;QACT,6BAA6B;QAC7B,IAAI,CAAC,QAAQ,EAAE,CAAC;QAChB,IAAI,CAAC,KAAK,EAAE,CAAC;IACf,CAAC;AACH,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gns-foundation/hive-worker",
-  "version": "0.5.0",
+  "version": "0.5.1",
   "description": "Turn your device into a GEIANT Hive compute node. Earn GNS tokens.",
   "bin": {
     "hive-worker": "./dist/cli.js"

package/src/mobydb_hooks.ts

@@ -6,7 +6,7 @@
 // Call initWorkerMobyDB() at worker startup.
 // ============================================================
 
-import { initMobyDB, getMobyDB, currentEpoch, type MobyDB } from './mobydb';
+import { initMobyDB, getMobyDB, currentEpoch, type MobyDB } from './mobydb.js';
 
 let sealInterval: NodeJS.Timeout | null = null;
 let syncInterval: NodeJS.Timeout | null = null;

package/package.json.v0.5.bak.20260515T094956Z

@@ -1,43 +0,0 @@
-{
-  "name": "@gns-foundation/hive-worker",
-  "version": "0.1.11",
-  "description": "Turn your device into a GEIANT Hive compute node. Earn GNS tokens.",
-  "bin": {
-    "hive-worker": "./dist/cli.js"
-  },
-  "scripts": {
-    "build": "tsc",
-    "dev": "ts-node src/cli.ts",
-    "prepublishOnly": "npm run build"
-  },
-  "keywords": [
-    "geiant",
-    "hive",
-    "gns",
-    "ai",
-    "inference",
-    "decentralized"
-  ],
-  "author": "GNS Foundation <cayerbe@ulissy.app>",
-  "license": "BSL-1.1",
-  "dependencies": {
-    "better-sqlite3": "^12.8.0",
-    "chalk": "^5.3.0",
-    "commander": "^12.1.0",
-    "h3-js": "^4.1.0",
-    "node-fetch": "^3.3.2",
-    "ora": "^8.0.1",
-    "tweetnacl": "^1.0.3",
-    "tweetnacl-util": "^0.15.1"
-  },
-  "devDependencies": {
-    "@types/better-sqlite3": "^7.6.13",
-    "@types/node": "^20.12.0",
-    "ts-node": "^10.9.2",
-    "typescript": "^5.4.5"
-  },
-  "engines": {
-    "node": ">=18.0.0"
-  },
-  "type": "module"
-}

package/src/cli.ts.v0.5.bak.20260515T094956Z

@@ -1,569 +0,0 @@
-#!/usr/bin/env node
-// ============================================================
-// HIVE WORKER CLI
-// Usage:
-//   npx @gns-foundation/hive-worker join [--handle <@name>] [--rpc-port <port>]
-//   npx @gns-foundation/hive-worker status
-//   npx @gns-foundation/hive-worker leave
-//   npx @gns-foundation/hive-worker whoami
-//   npx @gns-foundation/hive-worker models list
-//   npx @gns-foundation/hive-worker models fetch <model-id>
-// ============================================================
-
-import { program } from 'commander';
-import { loadOrCreateIdentity, shortPk, identityPath } from './identity.js';
-import { initWorkerMobyDB, shutdownMobyDB } from "./mobydb_hooks.js";
-import { detectHardware, detectGeo } from './hardware.js';
-import { registerNode, heartbeat, deregisterNode, fetchSwarmStats, fetchTokenBalance } from './registry.js';
-import { findRpcBinary, startRpcServer, stopRpcServer, DEFAULT_RPC_PORT, type LlamaRpcHandle } from './llama.js';
-import {
-  printBanner, printSuccess, printInfo, printWarn, printError,
-  renderDashboard, clearDashboard, hideCursor,
-  type DashboardState,
-} from './dashboard.js';
-import { startJobPoller, claimJobPipeline, markComputing, postResult, postFailure, postChunk, type PollController, type HiveJob, type JobResult } from './jobs.js';
-import { executeJob, downloadModel, listCachedModels, resolveModelUrl, findLlamaCli } from './executor.js';
-import { creditWorker, SPLIT } from './settlement.js';
-
-import os from 'os';
-
-// ─── Get local LAN IP ─────────────────────────────────────────
-function getLocalIp(): string | null {
-  const nets = os.networkInterfaces();
-  for (const name of Object.keys(nets)) {
-    for (const net of nets[name] ?? []) {
-      if (net.family === 'IPv4' && !net.internal) return net.address;
-    }
-  }
-  return null;
-}
-
-const HEARTBEAT_INTERVAL_MS     = 30_000;
-const STATS_REFRESH_INTERVAL_MS = 60_000;
-const DASHBOARD_REFRESH_MS      = 1_000;
-const JOB_POLL_INTERVAL_MS      = 5_000;   // poll for new jobs every 5s
-
-// ─── JOIN command ─────────────────────────────────────────────
-
-async function cmdJoin(opts: {
-  handle?: string;
-  rpcPort?: string;
-  skipRpc?: boolean;
-  noJobs?: boolean;
-}): Promise<void> {
-  printBanner();
-
-  // 1. Identity
-  const { identity, isNew } = loadOrCreateIdentity();
-  if (isNew) {
-    printSuccess(`New identity generated: ${shortPk(identity.pk)}`);
-    printInfo(`Stored at ${identityPath()}`);
-  } else {
-    printSuccess(`Identity loaded: ${shortPk(identity.pk)}`);
-  }
-
-  // 2. Handle
-  const handle = opts.handle?.replace(/^@/, '') ?? null;
-  if (handle) printSuccess(`Handle: @${handle}`);
-  else printInfo('No handle — use --handle @yourname to claim one.');
-
-  // 3. Hardware
-  printInfo('Detecting hardware...');
-  const hw = detectHardware();
-  printSuccess(`${hw.cpuCores}-core ${hw.arch} · ${hw.ramGb} GB RAM · ~${hw.estimatedTflops} TFLOPS`);
-  if (hw.gpuModel) printSuccess(`GPU: ${hw.gpuModel}`);
-
-  // 4. Geo
-  printInfo('Detecting location...');
-  const geo = await detectGeo();
-  printSuccess(`Location: ${geo.city}, ${geo.country} · H3 cell ${geo.h3Cell.slice(0, 10)}…`);
-
-  // Phase 4: Embedded MobyDB
-  initWorkerMobyDB(identity.pk, geo.h3Cell);
-
-  // 5. llama-cli check (for job execution)
-  const llamaCliAvailable = !!findLlamaCli();
-  if (!opts.noJobs) {
-    if (llamaCliAvailable) {
-      printSuccess('llama-cli found — job execution enabled');
-    } else {
-      printWarn('llama-cli not found — joining as observer (no inference jobs).');
-      printWarn('Install llama.cpp to earn GNS: https://github.com/ggerganov/llama.cpp');
-    }
-  }
-
-  // logs declared early so the rpc-server callback can reference it immediately
-  const logs: string[] = [`Joined at ${new Date().toISOString()}`];
-
-  // 6. RPC server (for pipeline/shard mode)
-  let rpcHandle: LlamaRpcHandle | null = null;
-  const rpcBinary = findRpcBinary();
-  const rpcPort = opts.skipRpc ? null : parseInt(opts.rpcPort ?? String(DEFAULT_RPC_PORT), 10);
-
-  if (!opts.skipRpc && rpcBinary) {
-    printInfo(`Starting llama.cpp rpc-server on :${rpcPort}...`);
-    rpcHandle = startRpcServer(rpcPort!, (line) => logs.push(line));
-    if (rpcHandle) printSuccess(`RPC server PID ${rpcHandle.pid} on :${rpcPort}`);
-  } else if (!opts.skipRpc) {
-    printWarn('rpc-server not found — pipeline mode disabled.');
-  }
-
-  // 7. Register
-  printInfo('Registering with GEIANT Hive swarm...');
-  try {
-    const cachedModelIds = listCachedModels().map(m => m.modelId);
-    const localIp = rpcHandle ? getLocalIp() : null;
-    await registerNode(identity, hw, geo, handle, rpcHandle?.port ?? null, cachedModelIds, localIp, !!rpcHandle);
-    printSuccess('Registered in swarm registry');
-  } catch (err) {
-    printError(`Registration failed: ${err instanceof Error ? err.message : String(err)}`);
-    printWarn('Continuing in offline mode — will retry on heartbeat');
-  }
-
-  // 8. Initial stats
-  let swarmStats = await fetchSwarmStats();
-  let tokensEarned = await fetchTokenBalance(identity.pk);
-  printSuccess(`Swarm: ${swarmStats.activeNodes} active nodes · ${swarmStats.totalTflops} TFLOPS`);
-
-  // Cached models
-  const cachedModels = listCachedModels();
-  if (cachedModels.length > 0) {
-    printSuccess(`Models cached: ${cachedModels.map(m => m.modelId).join(', ')}`);
-  } else if (!opts.noJobs && llamaCliAvailable) {
-    printWarn('No models cached. Fetch one: hive-worker models fetch phi-3-mini');
-  }
-
-  console.log('');
-  printInfo('Entering live mode. Press Ctrl+C to disconnect.\n');
-
-  // ─── Live state ───────────────────────────────────────────────
-
-  hideCursor();
-
-  const startTime = Date.now();
-  let heartbeatCount = 0;
-  let lastHeartbeat: Date | null = null;
-  let jobsCompleted = 0;
-
-  const addLog = (msg: string) => {
-    logs.push(`[${new Date().toLocaleTimeString()}] ${msg}`);
-    if (logs.length > 50) logs.splice(0, logs.length - 50);
-  };
-
-  const state: DashboardState = {
-    pk:            identity.pk,
-    handle,
-    status:        'idle',
-    hw,
-    geo,
-    rpcPort:       rpcHandle?.port ?? null,
-    rpcAvailable:  !!rpcBinary,
-    tokensEarned,
-    swarmStats,
-    uptimeSeconds: 0,
-    heartbeatCount: 0,
-    lastHeartbeat: null,
-    logs,
-  };
-
-  // ─── Heartbeat timer ─────────────────────────────────────────
-
-  const heartbeatTimer = setInterval(async () => {
-    try {
-      const freshModels = listCachedModels().map(m => m.modelId);
-      const localIp = rpcHandle ? getLocalIp() : null;
-      await heartbeat(identity.pk, state.status, freshModels, localIp);
-      heartbeatCount++;
-      lastHeartbeat = new Date();
-      state.heartbeatCount = heartbeatCount;
-      state.lastHeartbeat = lastHeartbeat;
-      addLog(`Heartbeat #${heartbeatCount}`);
-    } catch (err) {
-      addLog(`Heartbeat error: ${err instanceof Error ? err.message : String(err)}`);
-    }
-  }, HEARTBEAT_INTERVAL_MS);
-
-  // ─── Stats refresh timer ─────────────────────────────────────
-
-  const statsTimer = setInterval(async () => {
-    try {
-      swarmStats    = await fetchSwarmStats();
-      tokensEarned  = await fetchTokenBalance(identity.pk);
-      state.swarmStats   = swarmStats;
-      state.tokensEarned = tokensEarned;
-    } catch { /* non-fatal */ }
-  }, STATS_REFRESH_INTERVAL_MS);
-
-  // ─── Job polling loop ─────────────────────────────────────────
-
-  let jobPoller: PollController | null = null;
-
-  // When RPC server is running, pipeline poller handles everything
-  // Standard poller only runs when no RPC server is available
-  const useStandardPoller = !rpcHandle;
-
-  if (!opts.noJobs && useStandardPoller && (llamaCliAvailable || cachedModels.length > 0)) {
-    jobPoller = startJobPoller({
-      workerPk:   identity.pk,
-      h3Cell:     geo.h3Cell,
-      intervalMs: JOB_POLL_INTERVAL_MS,
-
-      onJobClaimed: (job: HiveJob) => {
-        state.status = 'computing';
-        addLog(`Job claimed: ${job.id.slice(0, 8)} · model=${job.model_id} · ${job.max_tokens} tokens`);
-      },
-
-      onJobCompleted: (job: HiveJob, result: JobResult) => {
-        state.status = 'idle';
-        jobsCompleted++;
-        addLog(
-          `Job ${job.id.slice(0, 8)} done · ${result.tokensPerSecond} tok/s · ` +
-          `earned ${(job.gns_reward * SPLIT.WORKER).toFixed(4)} GNS`,
-        );
-        // Optimistic local credit (server-side Stellar TX is async)
-        state.tokensEarned += job.gns_reward * SPLIT.WORKER;
-        // Kick off background settlement
-        creditWorker(identity.pk, job.id, job.gns_reward)
-          .then(rec => {
-            if (rec.stellarTxHash) {
-              addLog(`Settled · TX ${rec.stellarTxHash.slice(0, 12)}…`);
-            }
-          })
-          .catch(err => addLog(`Settlement error: ${err instanceof Error ? err.message : String(err)}`));
-      },
-
-      onJobFailed: (job: HiveJob, error: string) => {
-        state.status = 'idle';
-        addLog(`Job ${job.id.slice(0, 8)} failed: ${error.slice(0, 60)}`);
-      },
-
-      onLog: addLog,
-
-      executor: async (job: HiveJob, onToken?: (t: string) => void) => {
-        return executeJob(job, {
-          onToken,
-          onLog: addLog,
-          // No RPC peers in standard mode — solo execution
-        });
-      },
-    });
-
-    addLog('Job poller started — polling every 5s');
-
-    // ── Pipeline poll loop (only when RPC server is running) ──
-    // Polls for jobs and uses split inference when peers are available.
-    if (rpcHandle) {
-      const PIPELINE_INTERVAL = 3000; // faster than standard poll — pipeline takes priority
-      const pipelineTimer = setInterval(async () => {
-        if (state.status === 'computing') return; // already busy
-
-        try {
-          const result = await claimJobPipeline(
-            identity.pk,
-            geo.h3Cell,
-          );
-          if (!result) return;
-
-          const { job, peers } = result;
-          if (peers.length === 0) {
-            // No peers available — run solo on this worker
-            addLog(`Solo job ${job.id.slice(0, 8)} (no RPC peers available)`);
-            try {
-              await markComputing(job.id);
-              const execResult = await executeJob(job, { onLog: addLog });
-              await postResult(job.id, execResult);
-              jobsCompleted++;
-              addLog(`Solo job ${job.id.slice(0, 8)} done · ${execResult.tokensPerSecond} tok/s`);
-              state.tokensEarned += job.gns_reward * SPLIT.WORKER;
-              await creditWorker(identity.pk, job.id, job.gns_reward).catch(() => {});
-            } catch (execErr) {
-              const errMsg = execErr instanceof Error ? execErr.message : String(execErr);
-              await postFailure(job.id, errMsg).catch(() => {});
-            } finally {
-              state.status = 'idle';
-            }
-            return;
-          }
-
-          state.status = 'computing';
-          addLog(`Pipeline job ${job.id.slice(0, 8)} · ${peers.length} peer(s) · ${peers.map(p => p.rpc_host).join(', ')}`);
-
-          try {
-            await markComputing(job.id);
-            const execResult = await executeJob(job, {
-              onLog: addLog,
-              rpcPeers: peers,
-            });
-            await postResult(job.id, execResult);
-            jobsCompleted++;
-            addLog(`Pipeline job ${job.id.slice(0, 8)} done · ${execResult.tokensPerSecond} tok/s (distributed)`);
-            state.tokensEarned += job.gns_reward * SPLIT.WORKER;
-          } catch (execErr) {
-            const errMsg = execErr instanceof Error ? execErr.message : String(execErr);
-            await postFailure(job.id, errMsg).catch(() => {});
-            addLog(`Pipeline job ${job.id.slice(0, 8)} failed: ${errMsg.slice(0, 60)}`);
-          } finally {
-            state.status = 'idle';
-          }
-        } catch (err) {
-          // Non-fatal — log and continue
-          addLog(`Pipeline poll error: ${err instanceof Error ? err.message : String(err)}`);
-        }
-      }, PIPELINE_INTERVAL);
-
-      // Store timer for cleanup
-      const origStop = jobPoller!.stop.bind(jobPoller);
-      jobPoller = {
-        ...jobPoller!,
-        stop: () => { origStop(); clearInterval(pipelineTimer); },
-      };
-      addLog('Pipeline poller started — RPC server ready for distributed inference');
-    }
-  } else if (opts.noJobs) {
-    addLog('Job execution disabled (--no-jobs flag)');
-  } else if (!rpcHandle) {
-    addLog('Observer mode — install llama.cpp + fetch a model to execute jobs');
-  }
-
-  // ── Pipeline poller — runs when RPC server is available ──────
-  // This is the primary job handler when rpc-server is running.
-  // It claims jobs and runs them distributed (or solo if no peers).
-  if (!opts.noJobs && rpcHandle && (llamaCliAvailable || cachedModels.length > 0)) {
-    const PIPELINE_INTERVAL = 3000;
-    const pipelineTimer = setInterval(async () => {
-      if (state.status === 'computing') return;
-
-      try {
-        const result = await claimJobPipeline(identity.pk, geo.h3Cell);
-        if (!result) return;
-
-        const { job, peers } = result;
-        state.status = 'computing';
-
-        if (peers.length === 0) {
-          addLog(`Solo job ${job.id.slice(0, 8)} · model=${job.model_id}`);
-        } else {
-          addLog(`Pipeline job ${job.id.slice(0, 8)} · ${peers.length} peer(s) · ${peers.map(p => p.rpc_host).join(', ')}`);
-        }
-
-        try {
-          await markComputing(job.id);
-          const execResult = await executeJob(job, {
-            onLog: addLog,
-            rpcPeers: peers.length > 0 ? peers : undefined,
-          });
-          await postResult(job.id, execResult);
-          jobsCompleted++;
-          const mode = peers.length > 0 ? 'distributed' : 'solo';
-          addLog(`Job ${job.id.slice(0, 8)} done · ${execResult.tokensPerSecond} tok/s (${mode})`);
-          state.tokensEarned += job.gns_reward * SPLIT.WORKER;
-          await creditWorker(identity.pk, job.id, job.gns_reward).catch(() => {});
-        } catch (execErr) {
-          const errMsg = execErr instanceof Error ? execErr.message : String(execErr);
-          await postFailure(job.id, errMsg).catch(() => {});
-          addLog(`Job ${job.id.slice(0, 8)} failed: ${errMsg.slice(0, 60)}`);
-        } finally {
-          state.status = 'idle';
-        }
-      } catch (err) {
-        // Non-fatal poll errors
-      }
-    }, PIPELINE_INTERVAL);
-
-    // Wrap timer in jobPoller for cleanup
-    const prevStop = jobPoller?.stop.bind(jobPoller);
-    jobPoller = {
-      stop: () => { prevStop?.(); clearInterval(pipelineTimer); },
-    } as any;
-
-    addLog('Pipeline poller started — RPC server ready for distributed inference');
-  }
-
-  // ─── Dashboard refresh ────────────────────────────────────────
-
-  const dashTimer = setInterval(() => {
-    state.uptimeSeconds = Math.floor((Date.now() - startTime) / 1000);
-    renderDashboard(state);
-  }, DASHBOARD_REFRESH_MS);
-
-  renderDashboard(state);
-
-  // ─── Graceful shutdown ────────────────────────────────────────
-
-  const shutdown = async (signal: string) => {
-    clearInterval(heartbeatTimer);
-    clearInterval(statsTimer);
-    clearInterval(dashTimer);
-    jobPoller?.stop();
-    shutdownMobyDB();
-    clearDashboard();
-
-    console.log('\n');
-    printInfo(`Received ${signal}. Disconnecting gracefully...`);
-
-    try {
-      await deregisterNode(identity.pk);
-      printSuccess('Marked offline in swarm registry');
-    } catch { /* best-effort */ }
-
-    if (rpcHandle) {
-      stopRpcServer(rpcHandle);
-      printSuccess('RPC server stopped');
-    }
-
-    const earned = await fetchTokenBalance(identity.pk).catch(() => state.tokensEarned);
-    printSuccess(`Session complete · ${jobsCompleted} jobs · ${earned.toFixed(4)} GNS earned`);
-    console.log('');
-    process.exit(0);
-  };
-
-  process.on('SIGINT',  () => shutdown('SIGINT'));
-  process.on('SIGTERM', () => shutdown('SIGTERM'));
-}
-
-// ─── STATUS command ───────────────────────────────────────────
-
-async function cmdStatus(): Promise<void> {
-  printBanner();
-  const { identity } = loadOrCreateIdentity();
-  printInfo(`Identity: ${shortPk(identity.pk)}`);
-
-  const stats  = await fetchSwarmStats();
-  const tokens = await fetchTokenBalance(identity.pk);
-  printInfo(`Swarm: ${stats.activeNodes} active nodes · ${stats.totalTflops} TFLOPS`);
-  printInfo(`GNS earned: ${tokens.toFixed(4)}`);
-
-  const models = listCachedModels();
-  if (models.length > 0) {
-    printInfo(`Cached models: ${models.map(m => `${m.modelId} (${m.sizeMb} MB)`).join(', ')}`);
-  } else {
-    printWarn('No models cached. Run: hive-worker models fetch phi-3-mini');
-  }
-
-  const llamaCli = findLlamaCli();
-  const rpcBin   = findRpcBinary();
-  printInfo(`llama-cli: ${llamaCli ?? 'not found'}`);
-  printInfo(`rpc-server: ${rpcBin ?? 'not found'}`);
-}
-
-// ─── LEAVE command ────────────────────────────────────────────
-
-async function cmdLeave(): Promise<void> {
-  printBanner();
-  const { identity } = loadOrCreateIdentity();
-  printInfo(`Deregistering ${shortPk(identity.pk)} from swarm...`);
-  await deregisterNode(identity.pk);
-  printSuccess('Marked offline. Identity preserved at ' + identityPath());
-  printInfo('Run `hive-worker join` to rejoin.');
-}
-
-// ─── WHOAMI command ───────────────────────────────────────────
-
-async function cmdWhoami(): Promise<void> {
-  const { identity, isNew } = loadOrCreateIdentity();
-  if (isNew) printWarn('No identity found — a new one was just generated.');
-  console.log('');
-  console.log(`  pk:      ${identity.pk}`);
-  console.log(`  short:   ${shortPk(identity.pk)}`);
-  console.log(`  created: ${identity.createdAt}`);
-  console.log(`  file:    ${identityPath()}`);
-  console.log('');
-}
-
-// ─── MODELS LIST command ──────────────────────────────────────
-
-async function cmdModelsList(): Promise<void> {
-  const cached = listCachedModels();
-  console.log('');
-  if (cached.length === 0) {
-    printWarn('No models cached yet.');
-    printInfo('Fetch one: hive-worker models fetch phi-3-mini');
-  } else {
-    printSuccess(`${cached.length} model(s) cached:`);
-    for (const m of cached) {
-      console.log(`    ${m.modelId.padEnd(20)} ${m.sizeMb} MB`);
-    }
-  }
-  console.log('');
-  printInfo('Available models: phi-3-mini, gemma-2-2b, tinyllama');
-  console.log('');
-}
-
-// ─── MODELS FETCH command ─────────────────────────────────────
-
-async function cmdModelsFetch(modelId: string): Promise<void> {
-  printBanner();
-  const job = {
-    model_id:  modelId,
-    model_url: null,
-  } as Pick<HiveJob, 'model_id' | 'model_url'>;
-
-  const url = resolveModelUrl(job as HiveJob);
-  if (!url) {
-    printError(`Unknown model "${modelId}". Available: phi-3-mini, gemma-2-2b, tinyllama`);
-    process.exit(1);
-  }
-
-  printInfo(`Downloading ${modelId}...`);
-  printInfo(`Source: ${url}`);
-  console.log('');
-
-  let lastPct = -1;
-  const dest = await downloadModel(modelId, url, (pct, mbDone, mbTotal) => {
-    if (pct !== lastPct) {
-      process.stdout.write(`\r  Downloading... ${pct}% (${mbDone} / ${mbTotal} MB)   `);
-      lastPct = pct;
-    }
-  });
-
-  console.log('');
-  printSuccess(`Saved to ${dest}`);
-  printInfo('You can now run: hive-worker join');
-}
-
-// ─── CLI definition ───────────────────────────────────────────
-
-program
-  .name('hive-worker')
-  .description('GEIANT Hive compute node — earn GNS tokens for idle compute')
-  .version('0.1.0');
-
-program
-  .command('join')
-  .description('Join the GEIANT Hive swarm and start earning GNS')
-  .option('--handle <n>',      'Your GNS handle (e.g. @alice)')
-  .option('--rpc-port <port>', 'Port for llama.cpp rpc-server', String(DEFAULT_RPC_PORT))
-  .option('--skip-rpc',         'Skip starting rpc-server')
-  .option('--no-jobs',         'Observer only — do not execute inference jobs')
-  .action(async (opts) => { await cmdJoin(opts); });
-
-program
-  .command('status')
-  .description('Show swarm status and your token balance')
-  .action(async () => { await cmdStatus(); });
-
-program
-  .command('leave')
-  .description('Gracefully disconnect from the swarm')
-  .action(async () => { await cmdLeave(); });
-
-program
-  .command('whoami')
-  .description('Show your Hive identity')
-  .action(async () => { await cmdWhoami(); });
-
-const models = program
-  .command('models')
-  .description('Manage locally cached inference models');
-
-models
-  .command('list')
-  .description('List cached models')
-  .action(async () => { await cmdModelsList(); });
-
-models
-  .command('fetch <model-id>')
-  .description('Download a model (phi-3-mini, gemma-2-2b, tinyllama)')
-  .action(async (modelId: string) => { await cmdModelsFetch(modelId); });
-
-program.parse();

package/src/executor.ts.v0.5.bak.20260515T094956Z

@@ -1,279 +0,0 @@
-// ============================================================
-// HIVE WORKER — EXECUTOR
-// Runs llama-cli for an assigned job.
-//
-// Two modes:
-//   pipeline  — rpc-server already running, orchestrator calls us
-//               (worker just tracks status, no local llama-cli)
-//   solo      — worker claims and runs the whole model locally
-//               (for small models: phi-3-mini, gemma-2-2b, etc.)
-//
-// Solo mode is the active path for hive-worker v0.1 / v0.2.
-// Pipeline mode is the full swarm path (Phase 3 whitepaper).
-// ============================================================
-
-import { spawn } from 'child_process';
-import { execSync } from 'child_process';
-import os from 'os';
-import path from 'path';
-import type { HiveJob, JobResult, RpcPeer } from './jobs.js';
-
-// ─── Binary detection ─────────────────────────────────────────
-
-const LLAMA_CLI_CANDIDATES = [
-  // llama-completion: pure completion mode, clean stdout, no banner
-  `${os.homedir()}/llama.cpp/build/bin/llama-completion`,
-  `${os.homedir()}/llama.cpp/build/llama-completion`,
-  'llama-completion',
-  // Fallback to llama-cli
-  `${os.homedir()}/llama.cpp/build/bin/llama-cli`,
-  `${os.homedir()}/llama.cpp/build/llama-cli`,
-  'llama-cli',
-];
-
-export function findLlamaCli(): string | null {
-  for (const candidate of LLAMA_CLI_CANDIDATES) {
-    try {
-      execSync(`test -f "${candidate}" || which "${candidate}" 2>/dev/null`, { timeout: 2000 });
-      return candidate;
-    } catch { /* not found */ }
-  }
-  return null;
-}
-
-// ─── Model cache ──────────────────────────────────────────────
-
-const MODEL_CACHE_DIR = path.join(os.homedir(), '.hive', 'models');
-
-import fs from 'fs';
-
-export function ensureModelCacheDir(): void {
-  fs.mkdirSync(MODEL_CACHE_DIR, { recursive: true });
-}
-
-export function modelCachePath(modelId: string): string {
-  // Normalize: "phi-3-mini" → "phi-3-mini.gguf"
-  const filename = modelId.endsWith('.gguf') ? modelId : `${modelId}.gguf`;
-  return path.join(MODEL_CACHE_DIR, filename);
-}
-
-export function isModelCached(modelId: string): boolean {
-  return fs.existsSync(modelCachePath(modelId));
-}
-
-// Known small models with public GGUF URLs (Q4_K_M quantizations)
-const KNOWN_MODELS: Record<string, string> = {
-  'phi-3-mini':
-    'https://huggingface.co/microsoft/Phi-3-mini-4k-instruct-gguf/resolve/main/Phi-3-mini-4k-instruct-q4.gguf',
-  'gemma-2-2b':
-    'https://huggingface.co/bartowski/gemma-2-2b-it-GGUF/resolve/main/gemma-2-2b-it-Q4_K_M.gguf',
-  'tinyllama':
-    'https://huggingface.co/TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF/resolve/main/tinyllama-1.1b-chat-v1.0.Q4_K_M.gguf',
-};
-
-export function resolveModelUrl(job: HiveJob): string | null {
-  if (job.model_url) return job.model_url;
-  return KNOWN_MODELS[job.model_id] ?? null;
-}
-
-// ─── Download model (streaming, with progress) ───────────────
-
-export async function downloadModel(
-  modelId: string,
-  url: string,
-  onProgress: (pct: number, mbDone: number, mbTotal: number) => void,
-): Promise<string> {
-  ensureModelCacheDir();
-  const dest = modelCachePath(modelId);
-  const tmpDest = dest + '.download';
-
-  const resp = await fetch(url, { signal: AbortSignal.timeout(300_000) });
-  if (!resp.ok) throw new Error(`Model download failed: ${resp.status}`);
-
-  const totalBytes = parseInt(resp.headers.get('content-length') ?? '0', 10);
-  const totalMb = totalBytes / 1024 / 1024;
-
-  const writer = fs.createWriteStream(tmpDest);
-  let downloaded = 0;
-
-  if (!resp.body) throw new Error('No response body');
-
-  const reader = resp.body.getReader();
-  while (true) {
-    const { done, value } = await reader.read();
-    if (done) break;
-    writer.write(value);
-    downloaded += value.length;
-    if (totalBytes > 0) {
-      onProgress(
-        Math.round((downloaded / totalBytes) * 100),
-        Math.round(downloaded / 1024 / 1024),
-        Math.round(totalMb),
-      );
-    }
-  }
-
-  await new Promise<void>((res, rej) => {
-    writer.end(() => res());
-    writer.on('error', rej);
-  });
-
-  fs.renameSync(tmpDest, dest);
-  return dest;
-}
-
-// ─── Execute inference ────────────────────────────────────────
-
-export interface ExecutorOptions {
-  onToken?: (token: string) => void;
-  onLog?:   (line: string) => void;
-  rpcPeers?: RpcPeer[];  // peer RPC servers for pipeline parallelism
-}
-
-export async function executeJob(
-  job: HiveJob,
-  onTokenOrOpts: ((t: string) => void) | ExecutorOptions = {},
-): Promise<JobResult> {
-  const opts: ExecutorOptions = typeof onTokenOrOpts === 'function'
-    ? { onToken: onTokenOrOpts }
-    : onTokenOrOpts;
-  const llamaCli = findLlamaCli();
-  if (!llamaCli) {
-    throw new Error(
-      'llama-cli not found. Install llama.cpp: https://github.com/ggerganov/llama.cpp',
-    );
-  }
-
-  const modelPath = modelCachePath(job.model_id);
-  if (!fs.existsSync(modelPath)) {
-    throw new Error(
-      `Model "${job.model_id}" not cached at ${modelPath}. ` +
-      `Run: hive-worker models fetch ${job.model_id}`,
-    );
-  }
-
-  return runLlamaCli(llamaCli, modelPath, job, opts);
-}
-
-function runLlamaCli(
-  binary: string,
-  modelPath: string,
-  job: HiveJob,
-  opts: ExecutorOptions,
-): Promise<JobResult> {
-  return new Promise((resolve, reject) => {
-    const startMs = Date.now();
-    const threads = Math.max(1, Math.floor(os.cpus().length / 2));
-
-    // llama-completion: pure completion mode, clean stdout, no banner
-    const args = [
-      '--model',     modelPath,
-      '--prompt',    job.prompt,
-      '--n-predict', String(job.max_tokens),
-      '--temp',      String(job.temperature),
-      '--ctx-size',  '4096',
-      '--threads',   String(threads),
-      '--single-turn',  // exit after one completion
-      '-ngl', '0',      // CPU-only (Metal tensor API disabled on pre-M5)
-    ];
-
-    // Pipeline parallelism: add --rpc for each peer worker
-    const peers = opts.rpcPeers ?? [];
-    for (const peer of peers) {
-      args.push('--rpc', `${peer.rpc_host}:${peer.rpc_port}`);
-      opts.onLog?.(`Pipeline peer: ${peer.rpc_host}:${peer.rpc_port} (${peer.tflops} TFLOPS)`);
-    }
-    if (peers.length > 0) {
-      opts.onLog?.(`Pipeline mode: ${peers.length + 1} nodes, splitting ${job.max_tokens} tokens`);
-    }
-
-    opts.onLog?.(`Executing ${job.model_id} (${job.max_tokens} tokens)`);
-
-    const proc = spawn(binary, args, {
-      stdio: ['ignore', 'pipe', 'pipe'],
-    });
-
-    let stdout = '';
-    let stderr = '';
-    let streamBuffer = '';  // buffer for <|assistant|> detection before streaming
-    let streamingStarted = false;
-
-    proc.stdout.on('data', (chunk: Buffer) => {
-      const text = chunk.toString();
-      stdout += text;
-
-      // Stream tokens to caller once we've passed the <|assistant|> marker
-      if (opts.onToken) {
-        if (!streamingStarted) {
-          streamBuffer += text;
-          const markerIdx = streamBuffer.lastIndexOf('<|assistant|>');
-          if (markerIdx !== -1) {
-            // Found marker — emit everything after it
-            streamingStarted = true;
-            const afterMarker = streamBuffer.slice(markerIdx + '<|assistant|>'.length);
-            if (afterMarker.length > 0) opts.onToken(afterMarker);
-          }
-        } else {
-          // Already past marker — emit each chunk immediately
-          opts.onToken(text);
-        }
-      }
-    });
-    proc.stderr.on('data', (chunk: Buffer) => {
-      stderr += chunk.toString();
-      opts.onLog?.(chunk.toString().slice(0, 60).trim());
-    });
-
-    proc.on('error', (err) => reject(new Error(`spawn error: ${err.message}`)));
-
-    proc.on('close', (code) => {
-      const wallMs = Date.now() - startMs;
-      opts.onLog?.(`stdout=${stdout.length}b stderr=${stderr.length}b code=${code}`);
-
-      if (code !== 0 && stdout.trim().length === 0) {
-        reject(new Error(`llama-cli exited ${code}: ${stderr.slice(-200)}`));
-        return;
-      }
-
-      // llama-completion echoes the prompt with chat template then the response.
-      // Strip everything up to and including <|assistant|> token.
-      let resultText = stdout.trim();
-      const assistantMarker = '<|assistant|>';
-      const assistantIdx = resultText.lastIndexOf(assistantMarker);
-      if (assistantIdx !== -1) {
-        resultText = resultText.slice(assistantIdx + assistantMarker.length).trim();
-      }
-
-      // Parse tok/s from stderr timing line
-      let tokensPerSecond = 0;
-      const tokensGenerated = resultText.split(/\s+/).filter(Boolean).length;
-      const tpsMatch = stderr.match(/eval time\s*=\s*([\d.]+)\s*ms\s*\/\s*(\d+)\s*tokens/);
-      if (tpsMatch) {
-        const evalMs = parseFloat(tpsMatch[1]);
-        tokensPerSecond = Math.round((parseInt(tpsMatch[2], 10) / evalMs) * 1000 * 10) / 10;
-      } else if (wallMs > 0) {
-        tokensPerSecond = Math.round((tokensGenerated / wallMs) * 1000 * 10) / 10;
-      }
-
-      resolve({ resultText, tokensGenerated, tokensPerSecond });
-    });
-
-    const deadline = new Date(job.timeout_at).getTime();
-    const remaining = deadline - Date.now() - 10_000;
-    if (remaining > 0) {
-      setTimeout(() => { proc.kill(); reject(new Error('Job timed out')); }, remaining);
-    }
-  });
-}
-
-
-// ─── List cached models ───────────────────────────────────────
-
-export function listCachedModels(): Array<{ modelId: string; sizeMb: number }> {
-  ensureModelCacheDir();
-  const files = fs.readdirSync(MODEL_CACHE_DIR).filter(f => f.endsWith('.gguf'));
-  return files.map(f => ({
-    modelId: f.replace('.gguf', ''),
-    sizeMb:  Math.round(fs.statSync(path.join(MODEL_CACHE_DIR, f)).size / 1024 / 1024),
-  }));
-}