offgrid-ai 0.9.5 → 0.9.6

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "offgrid-ai",
-  "version": "0.9.5",
+  "version": "0.9.6",
   "description": "Privacy-first CLI for running local LLMs — discover, configure, run, benchmark",
   "author": "Eeshan Srivastava (https://eeshans.com)",
   "type": "module",

package/src/benchmark/finalize.mjs

@@ -94,7 +94,7 @@ async function responseErrorDetail(response) {
   }
 }
 
-export async function finalizeBenchmarkRun(runDirectory, runResult, speedMetrics) {
+export async function finalizeBenchmarkRun(runDirectory, runResult, speedMetrics, speedMetricsError = null) {
   const metadataPath = join(runDirectory, "metadata.json");
   const metadata = JSON.parse(await readFile(metadataPath, "utf8"));
   const now = new Date();
@@ -149,6 +149,7 @@ export async function finalizeBenchmarkRun(runDirectory, runResult, speedMetrics
 
   metadata.runner.speedMetrics = speedMetrics;
   metadata.runner.metricSource = speedMetrics?.metricSource ?? null;
+  metadata.runner.speedMetricsError = speedMetricsError ?? null;
 
   metadata.results = {
     wallClockMs: runResult.wallClockMs,
@@ -231,6 +232,10 @@ export function renderBenchmarkSummary(metadata) {
       console.log(pc.dim("\n" + tip));
     }
   }
+
+  if (status === "completed" && !runner?.speedMetrics && runner?.speedMetricsError) {
+    console.log(pc.dim(`\nSpeed metrics unavailable: ${runner.speedMetricsError}`));
+  }
 }
 
 function wrapText(text, width = 64) {

package/src/benchmark/flow.mjs

@@ -88,15 +88,18 @@ export async function runPreparedBenchmark(profile, runDirectory, options = {})
     const runResult = await runBenchmarkInPi(profile, runDirectory, { signal: controller.signal });
 
     let speedMetrics = null;
+    let speedMetricsError = null;
     if (!runResult.error) {
       try {
         speedMetrics = await queryServerMetrics(profile);
       } catch (err) {
-        runResult.error = { message: `Speed metrics query failed: ${err.message}` };
+        // Non-fatal: speed metrics are a supplementary measurement, not the
+        // benchmark itself. Don't poison the run result; surface it as a note.
+        speedMetricsError = err.message;
       }
     }
 
-    metadata = await finalizeBenchmarkRun(runDirectory, runResult, speedMetrics);
+    metadata = await finalizeBenchmarkRun(runDirectory, runResult, speedMetrics, speedMetricsError);
     renderBenchmarkSummary(metadata);
   } catch (err) {
     const failedResult = {

package/src/benchmark/metrics.mjs

@@ -4,6 +4,8 @@ import { backendFor } from "../backends.mjs";
 import { apiRootUrl } from "../process.mjs";
 
 const BENCH_SPEED_PROMPT = "Write a one-sentence summary of machine learning.";
+const SPEED_QUERY_TIMEOUT_MS = 120_000;
+const SPEED_QUERY_MAX_TOKENS = 64;
 
 export async function queryServerMetrics(profile) {
   const backend = backendFor(profile.backend);
@@ -26,13 +28,14 @@ async function queryLlamaCppMetrics(profile) {
     model: profile.modelAlias,
     messages: [{ role: "user", content: BENCH_SPEED_PROMPT }],
     stream: false,
+    max_tokens: SPEED_QUERY_MAX_TOKENS,
   };
 
   const response = await fetch(profile.baseUrl.replace(/\/$/u, "") + "/chat/completions", {
     method: "POST",
     headers: { "Content-Type": "application/json" },
     body: JSON.stringify(body),
-    signal: AbortSignal.timeout(60000),
+    signal: AbortSignal.timeout(SPEED_QUERY_TIMEOUT_MS),
   });
 
   if (!response.ok) {
@@ -66,13 +69,14 @@ async function queryOmlxMetrics(profile) {
     messages: [{ role: "user", content: BENCH_SPEED_PROMPT }],
     stream: true,
     stream_options: { include_usage: true },
+    max_tokens: SPEED_QUERY_MAX_TOKENS,
   };
 
   const response = await fetch(profile.baseUrl.replace(/\/$/u, "") + "/chat/completions", {
     method: "POST",
     headers: { "Content-Type": "application/json" },
     body: JSON.stringify(body),
-    signal: AbortSignal.timeout(60000),
+    signal: AbortSignal.timeout(SPEED_QUERY_TIMEOUT_MS),
   });
 
   if (!response.ok) {
@@ -117,6 +121,7 @@ async function queryOllamaMetrics(profile) {
     model: profile.modelAlias,
     prompt: BENCH_SPEED_PROMPT,
     stream: false,
+    options: { num_predict: SPEED_QUERY_MAX_TOKENS },
   };
 
   const apiBaseUrl = apiRootUrl(profile.baseUrl || backendFor(profile.backend).apiBaseUrl || "");
@@ -125,7 +130,7 @@ async function queryOllamaMetrics(profile) {
     method: "POST",
     headers: { "Content-Type": "application/json" },
     body: JSON.stringify(body),
-    signal: AbortSignal.timeout(60000),
+    signal: AbortSignal.timeout(SPEED_QUERY_TIMEOUT_MS),
   });
 
   if (!response.ok) {

package/src/benchmark/pi-runner.mjs

@@ -5,7 +5,7 @@ import { join } from "node:path";
 import { spawn } from "node:child_process";
 import {
   BENCH_COLORS, renderStreamEvent,
-  formatToolCall, printFinalLine,
+  formatToolCall, printFinalLine, stopExecTimer,
 } from "./stream-renderer.mjs";
 import { piModelString } from "./shared.mjs";
 
@@ -58,7 +58,8 @@ export async function runBenchmarkInPi(profile, runDirectory, { signal } = {}) {
     turnHadToolError: false,
     modelPrinted: false,
     activeTool: null,
-    status: { mode: "idle", toolName: null, bytes: 0, tokens: 0 },
+    execTimer: null,
+    status: { mode: "idle", toolName: null, bytes: 0, tokens: 0, execStartedAt: null },
   };
 
   function appendResponse(text) {
@@ -193,6 +194,7 @@ export async function runBenchmarkInPi(profile, runDirectory, { signal } = {}) {
   return new Promise((resolve) => {
     child.on("exit", async (code) => {
       if (signal) signal.removeEventListener("abort", abortListener);
+      stopExecTimer(renderState);
       if (streamBuffer.trim()) {
         processLine(streamBuffer);
       }
@@ -225,6 +227,7 @@ export async function runBenchmarkInPi(profile, runDirectory, { signal } = {}) {
 
     child.on("error", async (err) => {
       if (signal) signal.removeEventListener("abort", abortListener);
+      stopExecTimer(renderState);
       await streamHandle.close();
       await stderrHandle.close();
       runResult.error = { message: err.message };

package/src/benchmark/stream-renderer.mjs

@@ -107,6 +107,7 @@ export function renderStreamEvent(parsed, state, opts = {}) {
       };
       resetStatus(state, "exec", parsed.toolName);
       printFinalLine(BENCH_COLORS.tool(formatToolStart(parsed.toolName, parsed.args ?? {}, state)));
+      startExecTimer(state);
       break;
     }
     case "tool_execution_update": {
@@ -114,11 +115,13 @@ export function renderStreamEvent(parsed, state, opts = {}) {
       if (text) {
         if (verbose) process.stdout.write(BENCH_COLORS.toolOutput(text));
         if (state.activeTool) state.activeTool.outputText = text;
-        updateStatusFromDelta(state, text, "exec");
+        state.status.bytes += Buffer.byteLength(text, "utf8");
+        printExecStatus(state);
       }
       break;
     }
     case "tool_execution_end": {
+      stopExecTimer(state);
       const lines = formatToolEnd(parsed, state);
       if (parsed.isError) state.turnHadToolError = true;
       for (const line of lines) printFinalLine(line);
@@ -133,6 +136,7 @@ export function renderStreamEvent(parsed, state, opts = {}) {
       break;
     }
     case "turn_end": {
+      stopExecTimer(state);
       const usage = parsed.message?.usage;
       const tokenPart = usage ? ` · ${formatTokens(usage.output ?? usage.totalTokens ?? 0)} tokens` : "";
       const marker = state.turnHadToolError ? BENCH_COLORS.warning("⚠") : BENCH_COLORS.success("✓");
@@ -141,7 +145,7 @@ export function renderStreamEvent(parsed, state, opts = {}) {
       break;
     }
     case "agent_end":
-      clearStatusLine();
+      stopExecTimer(state);
       break;
     default:
       break;
@@ -172,6 +176,31 @@ export function updateStatusFromDelta(state, delta, mode = state.status.mode) {
   printStatusLine(BENCH_COLORS.dim(`Turn ${state.turn} ${modeLabel}${label} · ${bytes} (~${tokens} tokens)`));
 }
 
+export function startExecTimer(state) {
+  stopExecTimer(state);
+  state.status.execStartedAt = Date.now();
+  state.status.bytes = 0;
+  if (!process.stdout.isTTY) return;
+  printExecStatus(state);
+  state.execTimer = setInterval(() => printExecStatus(state), 1000);
+}
+
+export function stopExecTimer(state) {
+  if (state.execTimer) {
+    clearInterval(state.execTimer);
+    state.execTimer = null;
+  }
+  clearStatusLine();
+}
+
+export function printExecStatus(state) {
+  if (!process.stdout.isTTY) return;
+  const elapsed = state.status.execStartedAt ? Math.floor((Date.now() - state.status.execStartedAt) / 1000) : 0;
+  const tool = state.status.toolName ?? "tool";
+  const bytes = formatBytes(state.status.bytes);
+  printStatusLine(BENCH_COLORS.dim(`Turn ${state.turn} running ${tool}… ${elapsed}s · ${bytes}`));
+}
+
 export function formatToolStart(toolName, args, state) {
   if (toolName === "read") return `→ read ${displayPath(args.path, state)}`;
   if (toolName === "write") {

package/src/commands/models.mjs

@@ -1,7 +1,7 @@
 import { ensureDirs } from "../config.mjs";
 import { backendFor, BACKENDS } from "../backends.mjs";
 import { createProfileFromModel, readProfile, saveProfile, deleteProfile, profileJsonPath } from "../profiles.mjs";
-import { isProfileRunning, isProfileServerUp, stopProfile } from "../process.mjs";
+import { isProfileRunning, isProfileServerUp, modelAvailableOnServer, stopProfile } from "../process.mjs";
 import { syncPiConfig, removeFromPiConfig } from "../harness-pi.mjs";
 import { configureLocalProfile } from "../profile-setup.mjs";
 import { pc, startInteractive, createPrompt } from "../ui.mjs";
@@ -41,14 +41,18 @@ export async function modelCommandCenter(initialCatalog) {
 
   const runningProfilesNow = [];
   const serverUpIds = new Set();
+  const modelMissingIds = new Set();
   for (const profile of normalized.profiles) {
     if (await isProfileRunning(profile)) {
       runningProfilesNow.push(profile);
       continue;
     }
-    if (backendFor(profile.backend).type === "managed-server" && await isProfileServerUp(profile)) serverUpIds.add(profile.id);
+    if (backendFor(profile.backend).type === "managed-server" && await isProfileServerUp(profile)) {
+      if (await modelAvailableOnServer(profile)) serverUpIds.add(profile.id);
+      else modelMissingIds.add(profile.id);
+    }
   }
-  printWorkspaceHeader(normalized, runningProfilesNow, serverUpIds);
+  printWorkspaceHeader(normalized, runningProfilesNow, serverUpIds, modelMissingIds);
   await printBenchmarkLine();
 
   const nameWidth = modelNameWidth(allItems);
@@ -57,6 +61,7 @@ export async function modelCommandCenter(initialCatalog) {
     if (item.type === "profile") {
       if (item.fileMissing) return "missing";
       if (runningProfilesNow.some((profile) => profile.id === item.profile.id)) return "running";
+      if (modelMissingIds.has(item.profile.id)) return "missing";
       if (serverUpIds.has(item.profile.id)) return "serverup";
       return "ready";
     }
@@ -72,8 +77,8 @@ export async function modelCommandCenter(initialCatalog) {
     const bucket = grouped.get(group);
     if (!bucket || bucket.length === 0) continue;
     for (const item of bucket) {
-      const opt = modelSelectOption(item, { runningProfilesNow, serverUpIds, nameWidth });
-      choices.push({ value: opt.value, label: opt.label });
+      const opt = modelSelectOption(item, { runningProfilesNow, serverUpIds, modelMissingIds, nameWidth });
+      choices.push({ value: opt.value, label: opt.label, hint: opt.hint });
     }
   }
 

package/src/model-presenters.mjs

@@ -80,12 +80,17 @@ function optionLabel({ status, source, name, ctx, size, nameWidth }) {
   return [status, source, pc.bold(optionPad(name, null, nameWidth)), ctx, pc.dim(size)].join(OPTION_SEPARATOR);
 }
 
-export function modelSelectOption(item, { runningProfilesNow, serverUpIds, nameWidth }) {
+export function modelSelectOption(item, { runningProfilesNow, serverUpIds, modelMissingIds, nameWidth }) {
   if (item.type === "profile") {
     const backend = backendFor(item.profile.backend);
     const running = runningProfilesNow.some((profile) => profile.id === item.profile.id);
     const serverUp = !running && !item.fileMissing && serverUpIds?.has(item.profile.id);
-    const status = item.fileMissing ? "missing" : running ? "running" : serverUp ? "serverup" : "ready";
+    const modelMissing = !item.fileMissing && modelMissingIds?.has(item.profile.id);
+    const status = item.fileMissing || modelMissing ? "missing" : running ? "running" : serverUp ? "serverup" : "ready";
+    const drafterMissing = Boolean(item.profile.drafterPath) && !existsSync(item.profile.drafterPath);
+    const hint = drafterMissing ? "MTP drafter missing — reconfigure"
+      : modelMissing ? `${backend.label} model no longer available`
+      : undefined;
     return {
       value: itemKey(item),
       label: optionLabel({
@@ -96,6 +101,7 @@ export function modelSelectOption(item, { runningProfilesNow, serverUpIds, nameW
         ctx: optionCtxLabel(item),
         size: optionSizeLabel(item),
       }),
+      ...(hint ? { hint: pc.red(hint) } : {}),
     };
   }
   if (item.type === "new") {
@@ -125,10 +131,10 @@ export function modelSelectOption(item, { runningProfilesNow, serverUpIds, nameW
   };
 }
 
-export function printWorkspaceHeader(normalized, runningProfilesNow, serverUpIds = new Set()) {
+export function printWorkspaceHeader(normalized, runningProfilesNow, serverUpIds = new Set(), modelMissingIds = new Set()) {
   const profiles = normalized.profiles;
   const isRunning = (p) => runningProfilesNow.some((r) => r.id === p.id);
-  const isMissing = (p) => isProfileFileMissing(p);
+  const isMissing = (p) => isProfileFileMissing(p) || modelMissingIds.has(p.id);
   const readyCount = profiles.filter((p) => !isMissing(p) && !isRunning(p) && !serverUpIds.has(p.id)).length;
   const runningCount = runningProfilesNow.length;
   const serverUpCount = profiles.filter((p) => !isMissing(p) && serverUpIds.has(p.id) && !isRunning(p)).length;

package/src/profile-setup.mjs

@@ -1,3 +1,4 @@
+import { existsSync } from "node:fs";
 import { execFile } from "node:child_process";
 import { promisify } from "node:util";
 import { estimateMemory } from "./estimate.mjs";
@@ -36,6 +37,10 @@ export async function configureLocalProfile(prompt, profile) {
   // so that re-setup can pick up MTP availability, vision changes, etc.
   const freshCaps = detectCapabilities(profile.modelPath, profile.mmprojPath);
   let drafterPath = profile.drafterPath ?? null;
+  if (drafterPath && !existsSync(drafterPath)) {
+    // Stored drafter is no longer on disk — drop it and re-scan for a fresh one.
+    drafterPath = null;
+  }
   if (!drafterPath) {
     const { drafters } = await scanGgufModels();
     const drafter = matchDrafter(profile.modelPath, drafters);
@@ -47,6 +52,12 @@ export async function configureLocalProfile(prompt, profile) {
   if (hasMtp && configured.backend !== "llama-cpp-mtp") {
     configured = { ...configured, backend: "llama-cpp-mtp", providerId: "llama-cpp-mtp", drafterPath, capabilities: { ...configured.capabilities, mtp: true } };
   }
+  // If the profile was MTP but the drafter is now gone (and the model isn't
+  // natively MTP), switch back to plain llama.cpp so the server can start.
+  if (!hasMtp && configured.backend === "llama-cpp-mtp") {
+    console.log(pc.yellow("MTP drafter no longer found — switching to llama.cpp without speculative decoding."));
+    configured = removeMtpDefaults(configured);
+  }
   if (drafterPath && !configured.drafterPath) {
     configured = { ...configured, drafterPath };
   }