omnius 1.0.108 → 1.0.110

package/dist/index.js

@@ -563485,6 +563485,537 @@ var init_task_complete_box = __esm({
   }
 });
 
+// packages/cli/src/tui/model-picker.ts
+import { totalmem as totalmem3 } from "node:os";
+function isImageGenModel(name10, family) {
+  return IMAGE_GEN_PATTERNS.some((p2) => p2.test(name10) || family && p2.test(family));
+}
+function parseShowNumCtx(show) {
+  const sources = [show.parameters, show.modelfile];
+  for (const source of sources) {
+    if (!source) continue;
+    const match = source.match(/\b(?:PARAMETER\s+)?num_ctx\s+(\d+)/i);
+    if (match) return parseInt(match[1], 10);
+  }
+  return null;
+}
+async function fetchOllamaModels(baseUrl) {
+  const url = `${normalizeBaseUrl(baseUrl)}/api/tags`;
+  const resp = await fetch(url, {
+    signal: AbortSignal.timeout(1e4)
+  });
+  if (!resp.ok) {
+    throw new Error(`Failed to fetch models: HTTP ${resp.status}`);
+  }
+  const data = await resp.json();
+  const models = data.models ?? [];
+  const result = models.map((m2) => {
+    const family = m2.details?.family;
+    return {
+      name: m2.name,
+      size: formatBytes3(m2.size),
+      sizeBytes: m2.size,
+      modified: formatRelativeTime(m2.modified_at),
+      parameterSize: m2.details?.parameter_size,
+      contextLength: void 0,
+      caps: void 0,
+      isImageGen: isImageGenModel(m2.name, family),
+      family
+    };
+  }).sort((a2, b) => b.sizeBytes - a2.sizeBytes);
+  const normalized = normalizeBaseUrl(baseUrl);
+  const showResults = await Promise.allSettled(
+    result.map(
+      (m2) => fetch(`${normalized}/api/show`, {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({ name: m2.name }),
+        signal: AbortSignal.timeout(5e3)
+      }).then((r2) => r2.ok ? r2.json() : null)
+    )
+  );
+  for (let i2 = 0; i2 < result.length; i2++) {
+    const sr = showResults[i2];
+    if (sr?.status !== "fulfilled" || !sr.value) continue;
+    const show = sr.value;
+    const explicitNumCtx = parseShowNumCtx(show);
+    if (explicitNumCtx) {
+      result[i2].contextLength = explicitNumCtx;
+      continue;
+    }
+    if (show.model_info) {
+      const info = show.model_info;
+      const arch3 = info["general.architecture"];
+      const paramCount = info["general.parameter_count"];
+      const fileSizeGB = result[i2].sizeBytes > 0 ? result[i2].sizeBytes / 1024 ** 3 : paramCount ? paramCount * 0.6 / 1024 ** 3 : 4;
+      if (arch3) {
+        const archMax = info[`${arch3}.context_length`];
+        const nLayers = info[`${arch3}.block_count`];
+        const nKVHeads = info[`${arch3}.attention.head_count_kv`] ?? info[`${arch3}.attention.head_count`];
+        const keyDim = info[`${arch3}.attention.key_length`];
+        const valDim = info[`${arch3}.attention.value_length`] ?? keyDim;
+        if (archMax && nLayers && nKVHeads && keyDim && valDim) {
+          const kvBytesPerToken = nLayers * nKVHeads * (keyDim + valDim) * 2;
+          result[i2].contextLength = estimateRealisticContext(kvBytesPerToken, archMax, fileSizeGB);
+        } else if (archMax) {
+          const kvEstimate = fileSizeGB <= 5 ? 524288 : fileSizeGB <= 20 ? 1048576 : 1572864;
+          result[i2].contextLength = estimateRealisticContext(kvEstimate, archMax, fileSizeGB);
+        }
+      }
+    }
+    const modelCaps = { vision: false, toolUse: false, thinking: false };
+    const nameLower = result[i2].name.toLowerCase();
+    if (Array.isArray(show.capabilities)) {
+      if (show.capabilities.includes("vision")) modelCaps.vision = true;
+      if (show.capabilities.includes("tools")) modelCaps.toolUse = true;
+      if (show.capabilities.includes("thinking")) modelCaps.thinking = true;
+    }
+    if (show.model_info) {
+      for (const key of Object.keys(show.model_info)) {
+        const k = key.toLowerCase();
+        if (k.includes("vision.block_count") || k.includes("clip.") || k.includes("image_token_id") || k.includes("projector")) {
+          const val = show.model_info[key];
+          if (val !== null && val !== void 0 && val !== 0 && val !== "") {
+            modelCaps.vision = true;
+          }
+        }
+      }
+    }
+    if (/qwen3|qwen2\.5|llama3\.[13]|mistral|mixtral|command-r|gemma3|devstral|deepseek/.test(nameLower)) {
+      modelCaps.toolUse = true;
+    }
+    if (show.template && (show.template.includes("<think>") || show.template.includes("thinking"))) {
+      modelCaps.thinking = true;
+    }
+    result[i2].caps = modelCaps;
+  }
+  return result;
+}
+async function fetchOpenAIModels(baseUrl, apiKey) {
+  const normalized = normalizeBaseUrl(baseUrl);
+  const url = `${normalized}/v1/models`;
+  const isAnthropic = /api\.anthropic\.com/i.test(baseUrl);
+  const headers = {};
+  if (apiKey) {
+    if (isAnthropic) {
+      headers["x-api-key"] = apiKey;
+      headers["anthropic-version"] = "2023-06-01";
+    } else {
+      headers["Authorization"] = `Bearer ${apiKey}`;
+    }
+  }
+  const resp = await fetch(url, {
+    headers,
+    signal: AbortSignal.timeout(15e3)
+  });
+  if (!resp.ok) {
+    throw new Error(`Failed to fetch models: HTTP ${resp.status}`);
+  }
+  const data = await resp.json();
+  const models = data.data ?? [];
+  return models.map((m2) => ({
+    name: m2.id,
+    size: "",
+    sizeBytes: 0,
+    modified: m2.created ? formatRelativeTime(new Date(m2.created * 1e3).toISOString()) : "",
+    parameterSize: m2.owned_by ?? void 0,
+    contextLength: m2.context_length ?? m2.max_model_len ?? void 0
+  })).sort((a2, b) => a2.name.localeCompare(b.name));
+}
+async function fetchPeerModels(peerId, authKey) {
+  try {
+    const { NexusTool: NexusTool2 } = await Promise.resolve().then(() => (init_dist5(), dist_exports));
+    const { existsSync: existsSync131, readFileSync: readFileSync107 } = await import("node:fs");
+    const { join: join148 } = await import("node:path");
+    const cwd4 = process.cwd();
+    const nexusTool = new NexusTool2(cwd4);
+    const nexusDir = nexusTool.getNexusDir();
+    let isLocalPeer = false;
+    try {
+      const statusPath = join148(nexusDir, "status.json");
+      if (existsSync131(statusPath)) {
+        const status = JSON.parse(readFileSync107(statusPath, "utf8"));
+        if (status.peerId === peerId) isLocalPeer = true;
+      }
+    } catch {
+    }
+    if (isLocalPeer) {
+      const pricingPath = join148(nexusDir, "pricing.json");
+      if (existsSync131(pricingPath)) {
+        try {
+          const pricing = JSON.parse(readFileSync107(pricingPath, "utf8"));
+          const localModels = (pricing.models || []).map((m2) => ({
+            name: m2.model || "unknown",
+            size: m2.parameterSize || "",
+            modified: "",
+            sizeBytes: 0,
+            parameterSize: m2.parameterSize || "remote"
+          }));
+          if (localModels.length > 0) return localModels;
+        } catch {
+        }
+      }
+    }
+    const cachePath = join148(nexusDir, "peer-models-cache.json");
+    if (existsSync131(cachePath)) {
+      try {
+        const cache8 = JSON.parse(readFileSync107(cachePath, "utf8"));
+        if (cache8.peerId === peerId && cache8.models?.length > 0) {
+          const age = Date.now() - new Date(cache8.cachedAt).getTime();
+          if (age < 5 * 60 * 1e3) {
+            return cache8.models.map((m2) => ({
+              name: m2.name || "unknown",
+              size: m2.size || m2.parameterSize || "",
+              modified: "",
+              sizeBytes: 0,
+              parameterSize: m2.parameterSize || "remote"
+            }));
+          }
+        }
+      } catch {
+      }
+    }
+    try {
+      const capsResult = await nexusTool.execute({
+        action: "query_peer_caps",
+        peer_id: peerId,
+        ...authKey ? { auth_key: authKey } : {}
+      });
+      if (capsResult.success && capsResult.output) {
+        let capsData = null;
+        try {
+          capsData = JSON.parse(capsResult.output);
+        } catch {
+        }
+        if (capsData?.models && capsData.models.length > 0) {
+          return capsData.models.map((m2) => ({
+            name: m2.name || "unknown",
+            size: m2.parameterSize || "",
+            modified: "",
+            sizeBytes: 0,
+            parameterSize: m2.parameterSize || "remote"
+          }));
+        }
+        if (capsData?.capabilities && capsData.capabilities.length > 0) {
+          const models = [];
+          for (const cap of capsData.capabilities) {
+            if (typeof cap === "string" && cap.startsWith("inference:")) {
+              const capName = cap.slice(10);
+              const modelName = capName.replace(/_(\d+[bBmMkK])$/, ":$1").replace(/_latest$/, ":latest");
+              models.push({
+                name: modelName,
+                size: "",
+                modified: "",
+                sizeBytes: 0,
+                parameterSize: "remote"
+              });
+            }
+          }
+          if (models.length > 0) return models;
+        }
+      }
+    } catch {
+    }
+    try {
+      const natsResult = await nexusTool.execute({
+        action: "discover_peer_caps",
+        peer_id: peerId
+      });
+      if (natsResult.success && natsResult.output) {
+        let natsPeer = null;
+        try {
+          natsPeer = JSON.parse(natsResult.output);
+        } catch {
+        }
+        if (natsPeer?.capabilities && natsPeer.capabilities.length > 0) {
+          const models = [];
+          for (const cap of natsPeer.capabilities) {
+            if (typeof cap === "string" && cap.startsWith("inference:")) {
+              const capName = cap.slice(10);
+              const modelName = capName.replace(/_(\d+[bBmMkK])$/, ":$1").replace(/_latest$/, ":latest");
+              models.push({
+                name: modelName,
+                size: "",
+                modified: "",
+                sizeBytes: 0,
+                parameterSize: "remote"
+              });
+            }
+          }
+          if (models.length > 0) return models;
+        }
+      }
+    } catch {
+    }
+    try {
+      const result = await nexusTool.execute({
+        action: "find_agent",
+        peer_id: peerId
+      });
+      if (result.success && result.output) {
+        const models = [];
+        const capMatches = result.output.matchAll(/inference:([^\s,\]]+)/g);
+        for (const m2 of capMatches) {
+          const capName = m2[1];
+          const modelName = capName.replace(/_(\d+[bBmMkK])$/, ":$1").replace(/_latest$/, ":latest");
+          models.push({
+            name: modelName,
+            size: "",
+            modified: "",
+            sizeBytes: 0,
+            parameterSize: "remote"
+          });
+        }
+        if (models.length > 0) return models;
+      }
+    } catch {
+    }
+    if (isLocalPeer) {
+      const pricingPath = join148(nexusDir, "pricing.json");
+      if (existsSync131(pricingPath)) {
+        try {
+          const pricing = JSON.parse(readFileSync107(pricingPath, "utf8"));
+          return (pricing.models || []).map((m2) => ({
+            name: m2.model || "unknown",
+            size: m2.parameterSize || "",
+            modified: "",
+            sizeBytes: 0,
+            parameterSize: m2.parameterSize || "remote"
+          }));
+        } catch {
+        }
+      }
+    }
+    return [];
+  } catch {
+    return [];
+  }
+}
+async function fetchModels(baseUrl, apiKey) {
+  if (baseUrl.startsWith("peer://")) {
+    return fetchPeerModels(baseUrl.slice(7), apiKey);
+  }
+  const provider = detectProvider(baseUrl);
+  if (provider.id === "ollama") {
+    let ollamaErr;
+    try {
+      return await fetchOllamaModels(baseUrl);
+    } catch (err) {
+      ollamaErr = err instanceof Error ? err : new Error(String(err));
+      try {
+        return await fetchOpenAIModels(baseUrl, apiKey);
+      } catch {
+        throw new Error(`Cannot reach Ollama at ${baseUrl}: ${ollamaErr.message}`);
+      }
+    }
+  }
+  let lastErr;
+  for (let attempt = 0; attempt < 2; attempt++) {
+    try {
+      return await fetchOpenAIModels(baseUrl, apiKey);
+    } catch (err) {
+      lastErr = err instanceof Error ? err : new Error(String(err));
+      if (attempt === 0) await new Promise((r2) => setTimeout(r2, 1e3));
+    }
+  }
+  try {
+    return await fetchOllamaModels(baseUrl);
+  } catch {
+    throw new Error(`Cannot fetch models from ${provider.label} at ${baseUrl}: ${lastErr?.message ?? "unknown error"}`);
+  }
+}
+function stripLatest(modelName) {
+  return modelName.replace(/:latest$/i, "");
+}
+function findModel(models, query) {
+  const exact = models.find((m2) => m2.name === query);
+  if (exact) return exact;
+  const partial = models.find((m2) => m2.name.startsWith(query));
+  if (partial) return partial;
+  const fuzzy = models.find((m2) => m2.name.includes(query));
+  return fuzzy;
+}
+async function queryModelContextSize(baseUrl, modelName) {
+  try {
+    const normalized = normalizeBaseUrl(baseUrl);
+    const res = await fetch(`${normalized}/api/show`, {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ name: modelName }),
+      signal: AbortSignal.timeout(1e4)
+    });
+    if (!res.ok) return null;
+    const data = await res.json();
+    const explicitNumCtx = parseShowNumCtx(data);
+    if (explicitNumCtx) return explicitNumCtx;
+    if (data.model_info) {
+      const info = data.model_info;
+      const arch3 = info["general.architecture"];
+      const paramCount = info["general.parameter_count"];
+      const modelSizeGB2 = paramCount ? paramCount * 0.6 / 1024 ** 3 : 4;
+      if (arch3) {
+        const archMax = info[`${arch3}.context_length`];
+        const nLayers = info[`${arch3}.block_count`];
+        const nKVHeads = info[`${arch3}.attention.head_count_kv`] ?? info[`${arch3}.attention.head_count`];
+        const keyDim = info[`${arch3}.attention.key_length`];
+        const valDim = info[`${arch3}.attention.value_length`] ?? keyDim;
+        if (archMax && nLayers && nKVHeads && keyDim && valDim) {
+          const kvBytesPerToken = nLayers * nKVHeads * (keyDim + valDim) * 2;
+          return estimateRealisticContext(kvBytesPerToken, archMax, modelSizeGB2);
+        }
+        if (archMax) {
+          const kvEstimate = modelSizeGB2 <= 5 ? 524288 : modelSizeGB2 <= 20 ? 1048576 : 1572864;
+          return estimateRealisticContext(kvEstimate, archMax, modelSizeGB2);
+        }
+      }
+    }
+    return null;
+  } catch {
+    return null;
+  }
+}
+function estimateRealisticContext(kvBytesPerToken, archMax, modelSizeGB2) {
+  const totalMemGB = totalmem3() / 1024 ** 3;
+  const usableBytes = totalMemGB * 0.7 * 1024 ** 3;
+  const maxTokens = Math.floor(usableBytes / kvBytesPerToken);
+  let numCtx = Math.max(2048, Math.floor(maxTokens / 1024) * 1024);
+  numCtx = Math.min(numCtx, 131072, archMax);
+  if (modelSizeGB2 && modelSizeGB2 > 0) {
+    const maxKVBytes = modelSizeGB2 * 4 * 1024 ** 3;
+    const budgetCap = Math.max(2048, Math.floor(maxKVBytes / kvBytesPerToken / 1024) * 1024);
+    numCtx = Math.min(numCtx, budgetCap);
+  }
+  return numCtx;
+}
+async function queryOpenAIContextSize(baseUrl, modelName, apiKey) {
+  try {
+    const models = await fetchOpenAIModels(baseUrl, apiKey);
+    const model = models.find((m2) => m2.name === modelName);
+    if (model?.contextLength) return model.contextLength;
+    if (model?.size) {
+      const match = model.size.match(/(\d+)K ctx/);
+      if (match) return parseInt(match[1], 10) * 1024;
+    }
+    return null;
+  } catch {
+    return null;
+  }
+}
+async function queryContextSize(baseUrl, modelName, apiKey) {
+  if (baseUrl.startsWith("peer://")) return 32768;
+  const ollamaSize = await queryModelContextSize(baseUrl, modelName);
+  if (ollamaSize) return ollamaSize;
+  return queryOpenAIContextSize(baseUrl, modelName, apiKey);
+}
+async function queryModelCapabilities(baseUrl, modelName) {
+  const caps = { vision: false, toolUse: false, thinking: false };
+  if (baseUrl.startsWith("peer://")) {
+    const nameLower = modelName.toLowerCase();
+    if (/qwen3|qwen2\.5|llama3\.[13]|mistral|mixtral|command-r|gemma3|devstral|deepseek/.test(nameLower)) {
+      caps.toolUse = true;
+    }
+    if (/qwen3|deepseek-r1/.test(nameLower)) {
+      caps.thinking = true;
+    }
+    return caps;
+  }
+  try {
+    const normalized = normalizeBaseUrl(baseUrl);
+    const res = await fetch(`${normalized}/api/show`, {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ name: modelName }),
+      signal: AbortSignal.timeout(1e4)
+    });
+    if (!res.ok) return caps;
+    const data = await res.json();
+    if (Array.isArray(data.capabilities)) {
+      if (data.capabilities.includes("vision")) caps.vision = true;
+      if (data.capabilities.includes("tools")) caps.toolUse = true;
+      if (data.capabilities.includes("thinking")) caps.thinking = true;
+    }
+    if (data.model_info) {
+      for (const key of Object.keys(data.model_info)) {
+        const k = key.toLowerCase();
+        if (k.includes("vision.block_count") || k.includes("clip.") || k.includes("image_token_id") || k.includes("projector") || k.includes("vision.embedding_length")) {
+          const val = data.model_info[key];
+          if (val !== null && val !== void 0 && val !== 0 && val !== "") {
+            caps.vision = true;
+          }
+        }
+      }
+    }
+    const nameLower = modelName.toLowerCase();
+    if (/qwen3|qwen2\.5|llama3\.[13]|mistral|mixtral|command-r|gemma3|devstral|deepseek/.test(nameLower)) {
+      caps.toolUse = true;
+    }
+    if (data.template) {
+      if (data.template.includes("<think>") || data.template.includes("thinking")) {
+        caps.thinking = true;
+      }
+    }
+    return caps;
+  } catch {
+    return caps;
+  }
+}
+function formatBytes3(bytes) {
+  if (bytes < 1024) return `${bytes} B`;
+  const units = ["KB", "MB", "GB", "TB"];
+  let size = bytes;
+  let i2 = -1;
+  while (size >= 1024 && i2 < units.length - 1) {
+    size /= 1024;
+    i2++;
+  }
+  return `${size.toFixed(1)} ${units[i2] ?? "B"}`;
+}
+function formatContextLength(tokens) {
+  if (tokens >= 1e6) return `${(tokens / 1e6).toFixed(1)}M ctx`;
+  if (tokens >= 1024) return `${Math.round(tokens / 1024)}K ctx`;
+  return `${tokens} ctx`;
+}
+function formatCaps(caps) {
+  const tags = [];
+  if (caps.vision) tags.push("vision");
+  if (caps.toolUse) tags.push("tools");
+  if (caps.thinking) tags.push("think");
+  return tags.join("+");
+}
+function formatRelativeTime(iso2) {
+  const now = Date.now();
+  const then = new Date(iso2).getTime();
+  const diffMs = now - then;
+  const minutes = Math.floor(diffMs / 6e4);
+  if (minutes < 1) return "just now";
+  if (minutes < 60) return `${minutes}m ago`;
+  const hours = Math.floor(minutes / 60);
+  if (hours < 24) return `${hours}h ago`;
+  const days = Math.floor(hours / 24);
+  if (days < 7) return `${days}d ago`;
+  const weeks = Math.floor(days / 7);
+  if (weeks < 5) return `${weeks}w ago`;
+  const months = Math.floor(days / 30);
+  return `${months}mo ago`;
+}
+var IMAGE_GEN_PATTERNS;
+var init_model_picker = __esm({
+  "packages/cli/src/tui/model-picker.ts"() {
+    "use strict";
+    init_dist();
+    IMAGE_GEN_PATTERNS = [
+      /flux/i,
+      /z-image/i,
+      /stable-diffusion/i,
+      /sdxl/i,
+      /dall/i,
+      /kandinsky/i,
+      /midjourney/i,
+      /imagen/i
+    ];
+  }
+});
+
 // packages/cli/src/tui/render.ts
 var render_exports = {};
 __export(render_exports, {
@@ -564353,8 +564884,9 @@ function renderModelList(models, current) {
   ${c3.bold("Available models:")}
 
 `);
+  const currentKey = stripLatest(current);
   for (const m2 of models) {
-    const isCurrent = m2.name === current;
+    const isCurrent = stripLatest(m2.name) === currentKey;
     const marker = isCurrent ? c3.green("●") : c3.dim("○");
     const name10 = isCurrent ? c3.bold(c3.green(m2.name)) : m2.name;
     const size = c3.dim(m2.size);
@@ -564413,6 +564945,7 @@ var init_render = __esm({
     init_config();
     init_text_selection();
     init_task_complete_box();
+    init_model_picker();
     isTTY2 = process.stdout.isTTY ?? false;
     c3 = {
       bold: (t2) => ansi2("1", t2),
@@ -566137,7 +566670,7 @@ import { spawn as spawn24, exec as exec2 } from "node:child_process";
 import { EventEmitter as EventEmitter7 } from "node:events";
 import { randomBytes as randomBytes18, timingSafeEqual } from "node:crypto";
 import { URL as URL2 } from "node:url";
-import { loadavg, cpus as cpus2, totalmem as totalmem3, freemem as freemem3 } from "node:os";
+import { loadavg, cpus as cpus2, totalmem as totalmem4, freemem as freemem3 } from "node:os";
 import { existsSync as existsSync84, readFileSync as readFileSync66, writeFileSync as writeFileSync42, unlinkSync as unlinkSync13, mkdirSync as mkdirSync47, readdirSync as readdirSync28, statSync as statSync31, statfsSync as statfsSync4 } from "node:fs";
 import { join as join99 } from "node:path";
 function cleanForwardHeaders(raw, targetHost) {
@@ -566243,7 +566776,7 @@ function parseRateLimitHeaders(headers) {
 async function collectSystemMetricsAsync() {
   const [l1, l5, l15] = loadavg();
   const cores = cpus2().length;
-  const totalMem = totalmem3();
+  const totalMem = totalmem4();
   const freeMem = freemem3();
   const usedMem = totalMem - freeMem;
   let disk = {
@@ -568835,716 +569368,188 @@ var init_call_agent = __esm({
             feed.push({
               ts: Date.now(),
               source: "call",
-              sourceId: this.clientId,
-              summary: content,
-              toolName,
-              success
-            });
-          }
-          if (event.type === "model_response" && event.content) {
-            this.emit("response", event.content);
-          }
-        });
-      }
-      /** Process a voice transcript — queues if already processing */
-      handleTranscript(text) {
-        if (this.disposed) return;
-        this.conversationHistory.push({ role: "user", text });
-        if (this.processing) {
-          this.pendingTranscripts.push(text);
-          return;
-        }
-        this.processTranscript(text).catch((err) => {
-          this.emit("error", err instanceof Error ? err : new Error(String(err)));
-        });
-      }
-      /** Dispose and clean up */
-      dispose() {
-        this.disposed = true;
-        this.pendingTranscripts.length = 0;
-        this.runner = null;
-      }
-      // ── Private ──────────────────────────────────────────────────────────
-      async processTranscript(text) {
-        if (!this.runner || this.disposed) return;
-        this.processing = true;
-        try {
-          const historyContext = this.conversationHistory.slice(-10).map((h) => `${h.role === "user" ? "User" : "You"}: ${h.text}`).join("\n");
-          const feed = getActivityFeed();
-          const activitySummary = feed.getSummary(
-            this.tier === "admin" ? 20 : 10,
-            this.tier === "admin"
-          );
-          const wantsAction = /\b(read|open|show|run|execute|check|look at|find|search|grep|edit|write|fix|test|build|deploy|install|create|delete|remove|update|change|modify|commit|push|pull)\b/i.test(text) && !/\b(how are you|what's up|hello|hi|hey|can you hear|stop|quit|bye|thanks|thank you|ok|okay|sure|yeah|yes|no)\b/i.test(text);
-          if (!wantsAction) {
-            try {
-              const chatMessages = [
-                { role: "system", content: this.buildSystemPrompt() },
-                ...this.conversationHistory.slice(-6).map((h) => ({
-                  role: h.role === "user" ? "user" : "assistant",
-                  content: h.text
-                })),
-                { role: "user", content: text }
-              ];
-              const chatResult = await this.backend.chatCompletion({
-                messages: chatMessages,
-                tools: [],
-                temperature: 0.4,
-                maxTokens: 256,
-                timeoutMs: 15e3
-              });
-              const reply = (chatResult.choices[0]?.message?.content ?? "").trim();
-              if (!reply) return;
-              this.conversationHistory.push({ role: "assistant", text: reply });
-              this.emit("response", reply);
-            } catch {
-              this.emit("response", "Sorry, I couldn't process that.");
-            }
-          } else {
-            const taskPrompt = [
-              `User said: "${text}"`,
-              "",
-              historyContext ? `Conversation so far:
-${historyContext}
-` : "",
-              `Background activity:
-${activitySummary}
-`,
-              "The user is requesting an action. Use tools as needed, then call task_complete with a brief spoken summary of what you did (1-2 sentences)."
-            ].join("\n");
-            const result = await this.runner.run(taskPrompt, `Working directory: ${this.repoRoot}`);
-            if (result.summary) {
-              this.conversationHistory.push({ role: "assistant", text: result.summary });
-            }
-          }
-        } catch (err) {
-          this.emit("error", err instanceof Error ? err : new Error(String(err)));
-        } finally {
-          this.processing = false;
-          this.emit("done");
-          if (this.pendingTranscripts.length > 0) {
-            const next = this.pendingTranscripts.shift();
-            this.processTranscript(next).catch((err) => {
-              this.emit("error", err instanceof Error ? err : new Error(String(err)));
-            });
-          }
-        }
-      }
-      buildSystemPrompt() {
-        const base3 = [
-          "You are a voice assistant on a LIVE AUDIO CALL. This is a real-time conversation.",
-          "",
-          "CRITICAL RULES FOR VOICE CALLS:",
-          "1. ALWAYS respond IMMEDIATELY with speech. Do NOT use tools before responding.",
-          "2. Your response goes through text-to-speech — keep it SHORT (1-3 sentences).",
-          "3. NEVER use code blocks, markdown, or long technical text.",
-          "4. Be conversational and natural, like talking to a colleague.",
-          "5. Call task_complete with your spoken response as the summary.",
-          "6. Only use tools (file_read, grep, shell, etc.) if the user EXPLICITLY asks you to look something up, run a command, or make a change. For normal conversation, NEVER call tools.",
-          "7. If the user asks what's happening, summarize from the activity context below — do NOT run tools to find out."
-        ];
-        if (this.opts.emotionContext) {
-          base3.push("", "Mood:", this.opts.emotionContext);
-        }
-        if (this.tier === "admin") {
-          base3.push(
-            "",
-            "ADMIN call — you CAN use tools IF the user explicitly requests an action (e.g. 'read that file', 'run the tests').",
-            "But for general chat, status questions, or greetings — respond immediately WITHOUT tools."
-          );
-        } else {
-          base3.push(
-            "",
-            "PUBLIC call — read-only access. Answer questions about the project conversationally."
-          );
-        }
-        return base3.join("\n");
-      }
-      buildTools() {
-        if (this.tier === "admin") {
-          return this.buildAdminTools();
-        }
-        return this.buildPublicTools();
-      }
-      buildAdminTools() {
-        const debateAdapter = async (prompt) => {
-          const r2 = await this.backend.chatCompletion({
-            messages: [{ role: "user", content: prompt }],
-            tools: [],
-            temperature: 0.7,
-            maxTokens: 800,
-            timeoutMs: 12e4
-          });
-          return r2.choices[0]?.message?.content ?? "";
-        };
-        const replayAdapter = async (prompt) => {
-          const r2 = await this.backend.chatCompletion({
-            messages: [{ role: "user", content: prompt }],
-            tools: [],
-            temperature: 0,
-            maxTokens: 1500,
-            timeoutMs: 12e4
-          });
-          return r2.choices[0]?.message?.content ?? "";
-        };
-        const tools = [
-          new FileReadTool(this.repoRoot),
-          new FileWriteTool(this.repoRoot),
-          new FileEditTool(this.repoRoot),
-          new ShellTool(this.repoRoot),
-          new GrepSearchTool(this.repoRoot),
-          new GlobFindTool(this.repoRoot),
-          new ListDirectoryTool(this.repoRoot),
-          new WebSearchTool(),
-          new WebFetchTool(),
-          new MemoryReadTool(this.repoRoot),
-          new MemoryWriteTool(this.repoRoot),
-          new MemorySearchTool(this.repoRoot),
-          new DebateTool(debateAdapter),
-          new ReplayWithInterventionTool({ workingDir: this.repoRoot, callable: replayAdapter })
-        ];
-        return tools.map(adaptTool);
-      }
-      buildPublicTools() {
-        const tools = [
-          new FileReadTool(this.repoRoot),
-          new GrepSearchTool(this.repoRoot),
-          new GlobFindTool(this.repoRoot),
-          new ListDirectoryTool(this.repoRoot),
-          new MemoryReadTool(this.repoRoot),
-          new MemorySearchTool(this.repoRoot)
-        ];
-        return tools.map(adaptTool);
-      }
-    };
-  }
-});
-
-// packages/cli/src/tui/model-picker.ts
-import { totalmem as totalmem4 } from "node:os";
-function isImageGenModel(name10, family) {
-  return IMAGE_GEN_PATTERNS.some((p2) => p2.test(name10) || family && p2.test(family));
-}
-function parseShowNumCtx(show) {
-  const sources = [show.parameters, show.modelfile];
-  for (const source of sources) {
-    if (!source) continue;
-    const match = source.match(/\b(?:PARAMETER\s+)?num_ctx\s+(\d+)/i);
-    if (match) return parseInt(match[1], 10);
-  }
-  return null;
-}
-async function fetchOllamaModels(baseUrl) {
-  const url = `${normalizeBaseUrl(baseUrl)}/api/tags`;
-  const resp = await fetch(url, {
-    signal: AbortSignal.timeout(1e4)
-  });
-  if (!resp.ok) {
-    throw new Error(`Failed to fetch models: HTTP ${resp.status}`);
-  }
-  const data = await resp.json();
-  const models = data.models ?? [];
-  const result = models.map((m2) => {
-    const family = m2.details?.family;
-    return {
-      name: m2.name,
-      size: formatBytes3(m2.size),
-      sizeBytes: m2.size,
-      modified: formatRelativeTime(m2.modified_at),
-      parameterSize: m2.details?.parameter_size,
-      contextLength: void 0,
-      caps: void 0,
-      isImageGen: isImageGenModel(m2.name, family),
-      family
-    };
-  }).sort((a2, b) => b.sizeBytes - a2.sizeBytes);
-  const normalized = normalizeBaseUrl(baseUrl);
-  const showResults = await Promise.allSettled(
-    result.map(
-      (m2) => fetch(`${normalized}/api/show`, {
-        method: "POST",
-        headers: { "Content-Type": "application/json" },
-        body: JSON.stringify({ name: m2.name }),
-        signal: AbortSignal.timeout(5e3)
-      }).then((r2) => r2.ok ? r2.json() : null)
-    )
-  );
-  for (let i2 = 0; i2 < result.length; i2++) {
-    const sr = showResults[i2];
-    if (sr?.status !== "fulfilled" || !sr.value) continue;
-    const show = sr.value;
-    const explicitNumCtx = parseShowNumCtx(show);
-    if (explicitNumCtx) {
-      result[i2].contextLength = explicitNumCtx;
-      continue;
-    }
-    if (show.model_info) {
-      const info = show.model_info;
-      const arch3 = info["general.architecture"];
-      const paramCount = info["general.parameter_count"];
-      const fileSizeGB = result[i2].sizeBytes > 0 ? result[i2].sizeBytes / 1024 ** 3 : paramCount ? paramCount * 0.6 / 1024 ** 3 : 4;
-      if (arch3) {
-        const archMax = info[`${arch3}.context_length`];
-        const nLayers = info[`${arch3}.block_count`];
-        const nKVHeads = info[`${arch3}.attention.head_count_kv`] ?? info[`${arch3}.attention.head_count`];
-        const keyDim = info[`${arch3}.attention.key_length`];
-        const valDim = info[`${arch3}.attention.value_length`] ?? keyDim;
-        if (archMax && nLayers && nKVHeads && keyDim && valDim) {
-          const kvBytesPerToken = nLayers * nKVHeads * (keyDim + valDim) * 2;
-          result[i2].contextLength = estimateRealisticContext(kvBytesPerToken, archMax, fileSizeGB);
-        } else if (archMax) {
-          const kvEstimate = fileSizeGB <= 5 ? 524288 : fileSizeGB <= 20 ? 1048576 : 1572864;
-          result[i2].contextLength = estimateRealisticContext(kvEstimate, archMax, fileSizeGB);
-        }
-      }
-    }
-    const modelCaps = { vision: false, toolUse: false, thinking: false };
-    const nameLower = result[i2].name.toLowerCase();
-    if (Array.isArray(show.capabilities)) {
-      if (show.capabilities.includes("vision")) modelCaps.vision = true;
-      if (show.capabilities.includes("tools")) modelCaps.toolUse = true;
-      if (show.capabilities.includes("thinking")) modelCaps.thinking = true;
-    }
-    if (show.model_info) {
-      for (const key of Object.keys(show.model_info)) {
-        const k = key.toLowerCase();
-        if (k.includes("vision.block_count") || k.includes("clip.") || k.includes("image_token_id") || k.includes("projector")) {
-          const val = show.model_info[key];
-          if (val !== null && val !== void 0 && val !== 0 && val !== "") {
-            modelCaps.vision = true;
-          }
-        }
-      }
-    }
-    if (/qwen3|qwen2\.5|llama3\.[13]|mistral|mixtral|command-r|gemma3|devstral|deepseek/.test(nameLower)) {
-      modelCaps.toolUse = true;
-    }
-    if (show.template && (show.template.includes("<think>") || show.template.includes("thinking"))) {
-      modelCaps.thinking = true;
-    }
-    result[i2].caps = modelCaps;
-  }
-  return result;
-}
-async function fetchOpenAIModels(baseUrl, apiKey) {
-  const normalized = normalizeBaseUrl(baseUrl);
-  const url = `${normalized}/v1/models`;
-  const isAnthropic = /api\.anthropic\.com/i.test(baseUrl);
-  const headers = {};
-  if (apiKey) {
-    if (isAnthropic) {
-      headers["x-api-key"] = apiKey;
-      headers["anthropic-version"] = "2023-06-01";
-    } else {
-      headers["Authorization"] = `Bearer ${apiKey}`;
-    }
-  }
-  const resp = await fetch(url, {
-    headers,
-    signal: AbortSignal.timeout(15e3)
-  });
-  if (!resp.ok) {
-    throw new Error(`Failed to fetch models: HTTP ${resp.status}`);
-  }
-  const data = await resp.json();
-  const models = data.data ?? [];
-  return models.map((m2) => ({
-    name: m2.id,
-    size: "",
-    sizeBytes: 0,
-    modified: m2.created ? formatRelativeTime(new Date(m2.created * 1e3).toISOString()) : "",
-    parameterSize: m2.owned_by ?? void 0,
-    contextLength: m2.context_length ?? m2.max_model_len ?? void 0
-  })).sort((a2, b) => a2.name.localeCompare(b.name));
-}
-async function fetchPeerModels(peerId, authKey) {
-  try {
-    const { NexusTool: NexusTool2 } = await Promise.resolve().then(() => (init_dist5(), dist_exports));
-    const { existsSync: existsSync131, readFileSync: readFileSync107 } = await import("node:fs");
-    const { join: join148 } = await import("node:path");
-    const cwd4 = process.cwd();
-    const nexusTool = new NexusTool2(cwd4);
-    const nexusDir = nexusTool.getNexusDir();
-    let isLocalPeer = false;
-    try {
-      const statusPath = join148(nexusDir, "status.json");
-      if (existsSync131(statusPath)) {
-        const status = JSON.parse(readFileSync107(statusPath, "utf8"));
-        if (status.peerId === peerId) isLocalPeer = true;
-      }
-    } catch {
-    }
-    if (isLocalPeer) {
-      const pricingPath = join148(nexusDir, "pricing.json");
-      if (existsSync131(pricingPath)) {
-        try {
-          const pricing = JSON.parse(readFileSync107(pricingPath, "utf8"));
-          const localModels = (pricing.models || []).map((m2) => ({
-            name: m2.model || "unknown",
-            size: m2.parameterSize || "",
-            modified: "",
-            sizeBytes: 0,
-            parameterSize: m2.parameterSize || "remote"
-          }));
-          if (localModels.length > 0) return localModels;
-        } catch {
-        }
-      }
-    }
-    const cachePath = join148(nexusDir, "peer-models-cache.json");
-    if (existsSync131(cachePath)) {
-      try {
-        const cache8 = JSON.parse(readFileSync107(cachePath, "utf8"));
-        if (cache8.peerId === peerId && cache8.models?.length > 0) {
-          const age = Date.now() - new Date(cache8.cachedAt).getTime();
-          if (age < 5 * 60 * 1e3) {
-            return cache8.models.map((m2) => ({
-              name: m2.name || "unknown",
-              size: m2.size || m2.parameterSize || "",
-              modified: "",
-              sizeBytes: 0,
-              parameterSize: m2.parameterSize || "remote"
-            }));
+              sourceId: this.clientId,
+              summary: content,
+              toolName,
+              success
+            });
           }
-        }
-      } catch {
-      }
-    }
-    try {
-      const capsResult = await nexusTool.execute({
-        action: "query_peer_caps",
-        peer_id: peerId,
-        ...authKey ? { auth_key: authKey } : {}
-      });
-      if (capsResult.success && capsResult.output) {
-        let capsData = null;
-        try {
-          capsData = JSON.parse(capsResult.output);
-        } catch {
-        }
-        if (capsData?.models && capsData.models.length > 0) {
-          return capsData.models.map((m2) => ({
-            name: m2.name || "unknown",
-            size: m2.parameterSize || "",
-            modified: "",
-            sizeBytes: 0,
-            parameterSize: m2.parameterSize || "remote"
-          }));
-        }
-        if (capsData?.capabilities && capsData.capabilities.length > 0) {
-          const models = [];
-          for (const cap of capsData.capabilities) {
-            if (typeof cap === "string" && cap.startsWith("inference:")) {
-              const capName = cap.slice(10);
-              const modelName = capName.replace(/_(\d+[bBmMkK])$/, ":$1").replace(/_latest$/, ":latest");
-              models.push({
-                name: modelName,
-                size: "",
-                modified: "",
-                sizeBytes: 0,
-                parameterSize: "remote"
-              });
-            }
+          if (event.type === "model_response" && event.content) {
+            this.emit("response", event.content);
           }
-          if (models.length > 0) return models;
+        });
+      }
+      /** Process a voice transcript — queues if already processing */
+      handleTranscript(text) {
+        if (this.disposed) return;
+        this.conversationHistory.push({ role: "user", text });
+        if (this.processing) {
+          this.pendingTranscripts.push(text);
+          return;
         }
+        this.processTranscript(text).catch((err) => {
+          this.emit("error", err instanceof Error ? err : new Error(String(err)));
+        });
       }
-    } catch {
-    }
-    try {
-      const natsResult = await nexusTool.execute({
-        action: "discover_peer_caps",
-        peer_id: peerId
-      });
-      if (natsResult.success && natsResult.output) {
-        let natsPeer = null;
+      /** Dispose and clean up */
+      dispose() {
+        this.disposed = true;
+        this.pendingTranscripts.length = 0;
+        this.runner = null;
+      }
+      // ── Private ──────────────────────────────────────────────────────────
+      async processTranscript(text) {
+        if (!this.runner || this.disposed) return;
+        this.processing = true;
         try {
-          natsPeer = JSON.parse(natsResult.output);
-        } catch {
-        }
-        if (natsPeer?.capabilities && natsPeer.capabilities.length > 0) {
-          const models = [];
-          for (const cap of natsPeer.capabilities) {
-            if (typeof cap === "string" && cap.startsWith("inference:")) {
-              const capName = cap.slice(10);
-              const modelName = capName.replace(/_(\d+[bBmMkK])$/, ":$1").replace(/_latest$/, ":latest");
-              models.push({
-                name: modelName,
-                size: "",
-                modified: "",
-                sizeBytes: 0,
-                parameterSize: "remote"
+          const historyContext = this.conversationHistory.slice(-10).map((h) => `${h.role === "user" ? "User" : "You"}: ${h.text}`).join("\n");
+          const feed = getActivityFeed();
+          const activitySummary = feed.getSummary(
+            this.tier === "admin" ? 20 : 10,
+            this.tier === "admin"
+          );
+          const wantsAction = /\b(read|open|show|run|execute|check|look at|find|search|grep|edit|write|fix|test|build|deploy|install|create|delete|remove|update|change|modify|commit|push|pull)\b/i.test(text) && !/\b(how are you|what's up|hello|hi|hey|can you hear|stop|quit|bye|thanks|thank you|ok|okay|sure|yeah|yes|no)\b/i.test(text);
+          if (!wantsAction) {
+            try {
+              const chatMessages = [
+                { role: "system", content: this.buildSystemPrompt() },
+                ...this.conversationHistory.slice(-6).map((h) => ({
+                  role: h.role === "user" ? "user" : "assistant",
+                  content: h.text
+                })),
+                { role: "user", content: text }
+              ];
+              const chatResult = await this.backend.chatCompletion({
+                messages: chatMessages,
+                tools: [],
+                temperature: 0.4,
+                maxTokens: 256,
+                timeoutMs: 15e3
               });
+              const reply = (chatResult.choices[0]?.message?.content ?? "").trim();
+              if (!reply) return;
+              this.conversationHistory.push({ role: "assistant", text: reply });
+              this.emit("response", reply);
+            } catch {
+              this.emit("response", "Sorry, I couldn't process that.");
+            }
+          } else {
+            const taskPrompt = [
+              `User said: "${text}"`,
+              "",
+              historyContext ? `Conversation so far:
+${historyContext}
+` : "",
+              `Background activity:
+${activitySummary}
+`,
+              "The user is requesting an action. Use tools as needed, then call task_complete with a brief spoken summary of what you did (1-2 sentences)."
+            ].join("\n");
+            const result = await this.runner.run(taskPrompt, `Working directory: ${this.repoRoot}`);
+            if (result.summary) {
+              this.conversationHistory.push({ role: "assistant", text: result.summary });
             }
           }
-          if (models.length > 0) return models;
+        } catch (err) {
+          this.emit("error", err instanceof Error ? err : new Error(String(err)));
+        } finally {
+          this.processing = false;
+          this.emit("done");
+          if (this.pendingTranscripts.length > 0) {
+            const next = this.pendingTranscripts.shift();
+            this.processTranscript(next).catch((err) => {
+              this.emit("error", err instanceof Error ? err : new Error(String(err)));
+            });
+          }
         }
       }
-    } catch {
-    }
-    try {
-      const result = await nexusTool.execute({
-        action: "find_agent",
-        peer_id: peerId
-      });
-      if (result.success && result.output) {
-        const models = [];
-        const capMatches = result.output.matchAll(/inference:([^\s,\]]+)/g);
-        for (const m2 of capMatches) {
-          const capName = m2[1];
-          const modelName = capName.replace(/_(\d+[bBmMkK])$/, ":$1").replace(/_latest$/, ":latest");
-          models.push({
-            name: modelName,
-            size: "",
-            modified: "",
-            sizeBytes: 0,
-            parameterSize: "remote"
-          });
+      buildSystemPrompt() {
+        const base3 = [
+          "You are a voice assistant on a LIVE AUDIO CALL. This is a real-time conversation.",
+          "",
+          "CRITICAL RULES FOR VOICE CALLS:",
+          "1. ALWAYS respond IMMEDIATELY with speech. Do NOT use tools before responding.",
+          "2. Your response goes through text-to-speech — keep it SHORT (1-3 sentences).",
+          "3. NEVER use code blocks, markdown, or long technical text.",
+          "4. Be conversational and natural, like talking to a colleague.",
+          "5. Call task_complete with your spoken response as the summary.",
+          "6. Only use tools (file_read, grep, shell, etc.) if the user EXPLICITLY asks you to look something up, run a command, or make a change. For normal conversation, NEVER call tools.",
+          "7. If the user asks what's happening, summarize from the activity context below — do NOT run tools to find out."
+        ];
+        if (this.opts.emotionContext) {
+          base3.push("", "Mood:", this.opts.emotionContext);
         }
-        if (models.length > 0) return models;
-      }
-    } catch {
-    }
-    if (isLocalPeer) {
-      const pricingPath = join148(nexusDir, "pricing.json");
-      if (existsSync131(pricingPath)) {
-        try {
-          const pricing = JSON.parse(readFileSync107(pricingPath, "utf8"));
-          return (pricing.models || []).map((m2) => ({
-            name: m2.model || "unknown",
-            size: m2.parameterSize || "",
-            modified: "",
-            sizeBytes: 0,
-            parameterSize: m2.parameterSize || "remote"
-          }));
-        } catch {
+        if (this.tier === "admin") {
+          base3.push(
+            "",
+            "ADMIN call — you CAN use tools IF the user explicitly requests an action (e.g. 'read that file', 'run the tests').",
+            "But for general chat, status questions, or greetings — respond immediately WITHOUT tools."
+          );
+        } else {
+          base3.push(
+            "",
+            "PUBLIC call — read-only access. Answer questions about the project conversationally."
+          );
         }
+        return base3.join("\n");
       }
-    }
-    return [];
-  } catch {
-    return [];
-  }
-}
-async function fetchModels(baseUrl, apiKey) {
-  if (baseUrl.startsWith("peer://")) {
-    return fetchPeerModels(baseUrl.slice(7), apiKey);
-  }
-  const provider = detectProvider(baseUrl);
-  if (provider.id === "ollama") {
-    let ollamaErr;
-    try {
-      return await fetchOllamaModels(baseUrl);
-    } catch (err) {
-      ollamaErr = err instanceof Error ? err : new Error(String(err));
-      try {
-        return await fetchOpenAIModels(baseUrl, apiKey);
-      } catch {
-        throw new Error(`Cannot reach Ollama at ${baseUrl}: ${ollamaErr.message}`);
-      }
-    }
-  }
-  let lastErr;
-  for (let attempt = 0; attempt < 2; attempt++) {
-    try {
-      return await fetchOpenAIModels(baseUrl, apiKey);
-    } catch (err) {
-      lastErr = err instanceof Error ? err : new Error(String(err));
-      if (attempt === 0) await new Promise((r2) => setTimeout(r2, 1e3));
-    }
-  }
-  try {
-    return await fetchOllamaModels(baseUrl);
-  } catch {
-    throw new Error(`Cannot fetch models from ${provider.label} at ${baseUrl}: ${lastErr?.message ?? "unknown error"}`);
-  }
-}
-function findModel(models, query) {
-  const exact = models.find((m2) => m2.name === query);
-  if (exact) return exact;
-  const partial = models.find((m2) => m2.name.startsWith(query));
-  if (partial) return partial;
-  const fuzzy = models.find((m2) => m2.name.includes(query));
-  return fuzzy;
-}
-async function queryModelContextSize(baseUrl, modelName) {
-  try {
-    const normalized = normalizeBaseUrl(baseUrl);
-    const res = await fetch(`${normalized}/api/show`, {
-      method: "POST",
-      headers: { "Content-Type": "application/json" },
-      body: JSON.stringify({ name: modelName }),
-      signal: AbortSignal.timeout(1e4)
-    });
-    if (!res.ok) return null;
-    const data = await res.json();
-    const explicitNumCtx = parseShowNumCtx(data);
-    if (explicitNumCtx) return explicitNumCtx;
-    if (data.model_info) {
-      const info = data.model_info;
-      const arch3 = info["general.architecture"];
-      const paramCount = info["general.parameter_count"];
-      const modelSizeGB2 = paramCount ? paramCount * 0.6 / 1024 ** 3 : 4;
-      if (arch3) {
-        const archMax = info[`${arch3}.context_length`];
-        const nLayers = info[`${arch3}.block_count`];
-        const nKVHeads = info[`${arch3}.attention.head_count_kv`] ?? info[`${arch3}.attention.head_count`];
-        const keyDim = info[`${arch3}.attention.key_length`];
-        const valDim = info[`${arch3}.attention.value_length`] ?? keyDim;
-        if (archMax && nLayers && nKVHeads && keyDim && valDim) {
-          const kvBytesPerToken = nLayers * nKVHeads * (keyDim + valDim) * 2;
-          return estimateRealisticContext(kvBytesPerToken, archMax, modelSizeGB2);
-        }
-        if (archMax) {
-          const kvEstimate = modelSizeGB2 <= 5 ? 524288 : modelSizeGB2 <= 20 ? 1048576 : 1572864;
-          return estimateRealisticContext(kvEstimate, archMax, modelSizeGB2);
+      buildTools() {
+        if (this.tier === "admin") {
+          return this.buildAdminTools();
         }
+        return this.buildPublicTools();
       }
-    }
-    return null;
-  } catch {
-    return null;
-  }
-}
-function estimateRealisticContext(kvBytesPerToken, archMax, modelSizeGB2) {
-  const totalMemGB = totalmem4() / 1024 ** 3;
-  const usableBytes = totalMemGB * 0.7 * 1024 ** 3;
-  const maxTokens = Math.floor(usableBytes / kvBytesPerToken);
-  let numCtx = Math.max(2048, Math.floor(maxTokens / 1024) * 1024);
-  numCtx = Math.min(numCtx, 131072, archMax);
-  if (modelSizeGB2 && modelSizeGB2 > 0) {
-    const maxKVBytes = modelSizeGB2 * 4 * 1024 ** 3;
-    const budgetCap = Math.max(2048, Math.floor(maxKVBytes / kvBytesPerToken / 1024) * 1024);
-    numCtx = Math.min(numCtx, budgetCap);
-  }
-  return numCtx;
-}
-async function queryOpenAIContextSize(baseUrl, modelName, apiKey) {
-  try {
-    const models = await fetchOpenAIModels(baseUrl, apiKey);
-    const model = models.find((m2) => m2.name === modelName);
-    if (model?.contextLength) return model.contextLength;
-    if (model?.size) {
-      const match = model.size.match(/(\d+)K ctx/);
-      if (match) return parseInt(match[1], 10) * 1024;
-    }
-    return null;
-  } catch {
-    return null;
-  }
-}
-async function queryContextSize(baseUrl, modelName, apiKey) {
-  if (baseUrl.startsWith("peer://")) return 32768;
-  const ollamaSize = await queryModelContextSize(baseUrl, modelName);
-  if (ollamaSize) return ollamaSize;
-  return queryOpenAIContextSize(baseUrl, modelName, apiKey);
-}
-async function queryModelCapabilities(baseUrl, modelName) {
-  const caps = { vision: false, toolUse: false, thinking: false };
-  if (baseUrl.startsWith("peer://")) {
-    const nameLower = modelName.toLowerCase();
-    if (/qwen3|qwen2\.5|llama3\.[13]|mistral|mixtral|command-r|gemma3|devstral|deepseek/.test(nameLower)) {
-      caps.toolUse = true;
-    }
-    if (/qwen3|deepseek-r1/.test(nameLower)) {
-      caps.thinking = true;
-    }
-    return caps;
-  }
-  try {
-    const normalized = normalizeBaseUrl(baseUrl);
-    const res = await fetch(`${normalized}/api/show`, {
-      method: "POST",
-      headers: { "Content-Type": "application/json" },
-      body: JSON.stringify({ name: modelName }),
-      signal: AbortSignal.timeout(1e4)
-    });
-    if (!res.ok) return caps;
-    const data = await res.json();
-    if (Array.isArray(data.capabilities)) {
-      if (data.capabilities.includes("vision")) caps.vision = true;
-      if (data.capabilities.includes("tools")) caps.toolUse = true;
-      if (data.capabilities.includes("thinking")) caps.thinking = true;
-    }
-    if (data.model_info) {
-      for (const key of Object.keys(data.model_info)) {
-        const k = key.toLowerCase();
-        if (k.includes("vision.block_count") || k.includes("clip.") || k.includes("image_token_id") || k.includes("projector") || k.includes("vision.embedding_length")) {
-          const val = data.model_info[key];
-          if (val !== null && val !== void 0 && val !== 0 && val !== "") {
-            caps.vision = true;
-          }
-        }
+      buildAdminTools() {
+        const debateAdapter = async (prompt) => {
+          const r2 = await this.backend.chatCompletion({
+            messages: [{ role: "user", content: prompt }],
+            tools: [],
+            temperature: 0.7,
+            maxTokens: 800,
+            timeoutMs: 12e4
+          });
+          return r2.choices[0]?.message?.content ?? "";
+        };
+        const replayAdapter = async (prompt) => {
+          const r2 = await this.backend.chatCompletion({
+            messages: [{ role: "user", content: prompt }],
+            tools: [],
+            temperature: 0,
+            maxTokens: 1500,
+            timeoutMs: 12e4
+          });
+          return r2.choices[0]?.message?.content ?? "";
+        };
+        const tools = [
+          new FileReadTool(this.repoRoot),
+          new FileWriteTool(this.repoRoot),
+          new FileEditTool(this.repoRoot),
+          new ShellTool(this.repoRoot),
+          new GrepSearchTool(this.repoRoot),
+          new GlobFindTool(this.repoRoot),
+          new ListDirectoryTool(this.repoRoot),
+          new WebSearchTool(),
+          new WebFetchTool(),
+          new MemoryReadTool(this.repoRoot),
+          new MemoryWriteTool(this.repoRoot),
+          new MemorySearchTool(this.repoRoot),
+          new DebateTool(debateAdapter),
+          new ReplayWithInterventionTool({ workingDir: this.repoRoot, callable: replayAdapter })
+        ];
+        return tools.map(adaptTool);
       }
-    }
-    const nameLower = modelName.toLowerCase();
-    if (/qwen3|qwen2\.5|llama3\.[13]|mistral|mixtral|command-r|gemma3|devstral|deepseek/.test(nameLower)) {
-      caps.toolUse = true;
-    }
-    if (data.template) {
-      if (data.template.includes("<think>") || data.template.includes("thinking")) {
-        caps.thinking = true;
+      buildPublicTools() {
+        const tools = [
+          new FileReadTool(this.repoRoot),
+          new GrepSearchTool(this.repoRoot),
+          new GlobFindTool(this.repoRoot),
+          new ListDirectoryTool(this.repoRoot),
+          new MemoryReadTool(this.repoRoot),
+          new MemorySearchTool(this.repoRoot)
+        ];
+        return tools.map(adaptTool);
       }
-    }
-    return caps;
-  } catch {
-    return caps;
-  }
-}
-function formatBytes3(bytes) {
-  if (bytes < 1024) return `${bytes} B`;
-  const units = ["KB", "MB", "GB", "TB"];
-  let size = bytes;
-  let i2 = -1;
-  while (size >= 1024 && i2 < units.length - 1) {
-    size /= 1024;
-    i2++;
-  }
-  return `${size.toFixed(1)} ${units[i2] ?? "B"}`;
-}
-function formatContextLength(tokens) {
-  if (tokens >= 1e6) return `${(tokens / 1e6).toFixed(1)}M ctx`;
-  if (tokens >= 1024) return `${Math.round(tokens / 1024)}K ctx`;
-  return `${tokens} ctx`;
-}
-function formatCaps(caps) {
-  const tags = [];
-  if (caps.vision) tags.push("vision");
-  if (caps.toolUse) tags.push("tools");
-  if (caps.thinking) tags.push("think");
-  return tags.join("+");
-}
-function formatRelativeTime(iso2) {
-  const now = Date.now();
-  const then = new Date(iso2).getTime();
-  const diffMs = now - then;
-  const minutes = Math.floor(diffMs / 6e4);
-  if (minutes < 1) return "just now";
-  if (minutes < 60) return `${minutes}m ago`;
-  const hours = Math.floor(minutes / 60);
-  if (hours < 24) return `${hours}h ago`;
-  const days = Math.floor(hours / 24);
-  if (days < 7) return `${days}d ago`;
-  const weeks = Math.floor(days / 7);
-  if (weeks < 5) return `${weeks}w ago`;
-  const months = Math.floor(days / 30);
-  return `${months}mo ago`;
-}
-var IMAGE_GEN_PATTERNS;
-var init_model_picker = __esm({
-  "packages/cli/src/tui/model-picker.ts"() {
-    "use strict";
-    init_dist();
-    IMAGE_GEN_PATTERNS = [
-      /flux/i,
-      /z-image/i,
-      /stable-diffusion/i,
-      /sdxl/i,
-      /dall/i,
-      /kandinsky/i,
-      /midjourney/i,
-      /imagen/i
-    ];
+    };
   }
 });
 
@@ -577955,6 +577960,7 @@ __export(setup_exports, {
   ensurePythonVenv: () => ensurePythonVenv,
   ensureVisionDeps: () => ensureVisionDeps,
   expandedModelName: () => expandedModelName,
+  formatExpandedContextDiagnostic: () => formatExpandedContextDiagnostic,
   getLatestOllamaVersion: () => getLatestOllamaVersion,
   getOllamaVersion: () => getOllamaVersion,
   hasCmd: () => hasCmd,
@@ -578020,7 +578026,7 @@ async function needsTextToolMode(modelName, backendUrl2) {
   const hasTools = await checkToolSupport(modelName, backendUrl2);
   return !hasTools;
 }
-function detectUnifiedMemory() {
+function detectUnifiedMemory(hasDiscreteGpu = false) {
   if (process.platform === "darwin" && process.arch === "arm64") return true;
   if (process.platform === "linux") {
     try {
@@ -578034,6 +578040,16 @@ function detectUnifiedMemory() {
       }
     } catch {
     }
+    try {
+      if (existsSync91("/proc/device-tree/model")) {
+        const model = readFileSync74("/proc/device-tree/model", "utf8").replace(/\0+$/, "").toLowerCase();
+        if (/jetson|tegra|orin|xavier|nano|raspberry|rockchip|rk\d{4}|mt\d{4}/.test(model)) {
+          return true;
+        }
+      }
+    } catch {
+    }
+    if (process.arch === "arm64" && !hasDiscreteGpu) return true;
   }
   return false;
 }
@@ -578127,11 +578143,15 @@ function detectSystemSpecs() {
     } catch {
     }
   }
-  const unifiedMemory = detectUnifiedMemory();
+  const unifiedMemory = detectUnifiedMemory(gpuVramGB > 0);
   if (unifiedMemory && totalRamGB > 0) {
     const floorGB = totalRamGB * 0.8;
     if (availableRamGB < floorGB) availableRamGB = floorGB;
   }
+  if (!unifiedMemory && totalRamGB > 0) {
+    const floorGB = totalRamGB * 0.75;
+    if (availableRamGB < floorGB) availableRamGB = floorGB;
+  }
   return {
     totalRamGB: Math.round(totalRamGB * 10) / 10,
     availableRamGB: Math.round(availableRamGB * 10) / 10,
@@ -578191,11 +578211,15 @@ async function detectSystemSpecsAsync() {
     } catch {
     }
   }
-  const unifiedMemory = detectUnifiedMemory();
+  const unifiedMemory = detectUnifiedMemory(gpuVramGB > 0);
   if (unifiedMemory && totalRamGB > 0) {
     const floorGB = totalRamGB * 0.8;
     if (availableRamGB < floorGB) availableRamGB = floorGB;
   }
+  if (!unifiedMemory && totalRamGB > 0) {
+    const floorGB = totalRamGB * 0.75;
+    if (availableRamGB < floorGB) availableRamGB = floorGB;
+  }
   return {
     totalRamGB: Math.round(totalRamGB * 10) / 10,
     availableRamGB: Math.round(availableRamGB * 10) / 10,
@@ -578259,17 +578283,41 @@ function formatContextLabel(numCtx) {
   return numCtx >= 1024 ? `${Math.floor(numCtx / 1024)}K` : String(numCtx);
 }
 function calculateExpandedVariantContextWindow(specs, modelSizeGB2, kvBytesPerToken, archMax) {
-  const memoryFit = calculateMemoryBoundedNumCtx(
-    specs,
-    modelSizeGB2,
-    kvBytesPerToken,
-    archMax
-  );
-  const archCtx = archMax && archMax > 0 ? Math.max(2048, Math.floor(archMax / 1024) * 1024) : Number.POSITIVE_INFINITY;
+  const ramBudget = specs.availableRamGB > 0 ? specs.availableRamGB : specs.totalRamGB;
+  const vramBudget = specs.availableVramGB > 0 ? specs.availableVramGB : specs.gpuVramGB;
+  const totalAvail = Math.max(vramBudget, ramBudget);
+  const remaining = Math.max(0, totalAvail - modelSizeGB2);
+  const usableGB = remaining * 0.85;
+  const memoryFit = calculateMemoryBoundedNumCtx(specs, modelSizeGB2, kvBytesPerToken, archMax);
+  const archCtxValue = archMax && archMax > 0 ? Math.max(2048, Math.floor(archMax / 1024) * 1024) : null;
+  const archCtx = archCtxValue ?? Number.POSITIVE_INFINITY;
   const floor = Math.min(EXPANDED_VARIANT_MIN_NUM_CTX, archCtx);
   const fits = Math.min(memoryFit, archCtx);
   const numCtx = Math.max(floor, fits);
-  return { numCtx, label: formatContextLabel(numCtx) };
+  let limitedBy;
+  if (numCtx === floor && fits < floor) limitedBy = "floor";
+  else if (archCtxValue !== null && numCtx === archCtxValue) limitedBy = "arch";
+  else limitedBy = "memory";
+  const effectiveKvBpt = kvBytesPerToken && kvBytesPerToken > 0 ? kvBytesPerToken : (modelSizeGB2 <= 5 ? 64 : modelSizeGB2 <= 12 ? 160 : modelSizeGB2 <= 25 ? 256 : 384) * 1024;
+  return {
+    numCtx,
+    label: formatContextLabel(numCtx),
+    math: {
+      numCtx,
+      label: formatContextLabel(numCtx),
+      modelSizeGB: modelSizeGB2,
+      kvBytesPerToken: effectiveKvBpt,
+      kvSource: kvBytesPerToken && kvBytesPerToken > 0 ? "model_info" : "fallback",
+      archMax: archMax && archMax > 0 ? archMax : null,
+      ramBudgetGB: ramBudget,
+      vramBudgetGB: vramBudget,
+      usableGB,
+      memoryFit,
+      archCtx: archCtxValue,
+      floor,
+      limitedBy
+    }
+  };
 }
 function ask(rl, question) {
   return new Promise((resolve52) => {
@@ -580064,7 +580112,7 @@ async function queryModelKVInfo(backendUrl2, modelName) {
     const arch3 = info["general.architecture"];
     if (!arch3) return null;
     const nLayersRaw = info[`${arch3}.block_count`];
-    const nKVHeadsRaw = info[`${arch3}.attention.head_count_kv`] ?? info[`${arch3}.attention.head_count`];
+    const nKVHeadsRaw = info[`${arch3}.attention.head_count_kv`];
     const keyDimRaw = info[`${arch3}.attention.key_length`];
     const valDimRaw = info[`${arch3}.attention.value_length`] ?? keyDimRaw;
     const archMax = info[`${arch3}.context_length`];
@@ -580072,7 +580120,7 @@ async function queryModelKVInfo(backendUrl2, modelName) {
     const keyDim = keyDimRaw ?? 128;
     const valDim = valDimRaw ?? 128;
     const nLayers = nLayersRaw ?? defaultLayersForArch(arch3);
-    const nKVHeads = nKVHeadsRaw ?? 32;
+    const nKVHeads = nKVHeadsRaw ?? 8;
     if (!nLayers) return { archMax };
     const kvBytesPerToken = nLayers * nKVHeads * (keyDim + valDim) * 2;
     return { kvBytesPerToken, archMax };
@@ -580204,6 +580252,21 @@ async function createExpandedVariantAsync(baseModel, specs, sizeGB, kvBytesPerTo
     archMax
   );
 }
+function formatExpandedContextDiagnostic(specs, math) {
+  const fmtGB = (n2) => `${n2.toFixed(1)}GB`;
+  const fmtKB = (n2) => `${Math.round(n2 / 1024)}KB`;
+  const fmtK = (n2) => n2 >= 1024 ? `${Math.floor(n2 / 1024)}K` : String(n2);
+  const memBits = [];
+  if (specs.gpuVramGB > 0) {
+    memBits.push(`VRAM ${fmtGB(specs.availableVramGB || specs.gpuVramGB)}/${fmtGB(specs.gpuVramGB)}`);
+  }
+  memBits.push(`RAM ${fmtGB(specs.availableRamGB)}/${fmtGB(specs.totalRamGB)}${specs.unifiedMemory ? " unified" : ""}`);
+  const mem = memBits.join(", ");
+  const kv = `KV ${fmtKB(math.kvBytesPerToken)}/tok (${math.kvSource})`;
+  const fit2 = `fit ${fmtK(math.memoryFit)}, arch ${math.archCtx !== null ? fmtK(math.archCtx) : "n/a"}, floor ${fmtK(math.floor)}`;
+  const limit = `→ ${fmtK(math.numCtx)} (${math.limitedBy === "floor" ? "min floor" : math.limitedBy === "arch" ? "arch-capped" : "memory-fit"})`;
+  return `[${mem} | model ${fmtGB(math.modelSizeGB)} | ${kv} | ${fit2} ${limit}]`;
+}
 async function ensureExpandedContext(modelName, backendUrl2) {
   if (modelName.includes("cloud") || modelName.includes(":cloud")) {
     return { model: modelName, created: false, contextLabel: "remote", numCtx: 0 };
@@ -580230,11 +580293,11 @@ async function ensureExpandedContext(modelName, backendUrl2) {
       kvInfo?.kvBytesPerToken,
       kvInfo?.archMax
     ).catch(() => ({ repaired: false, currentNumCtx: 0, baseModel: null, resolvedModel: modelName }));
-    return { model: repair.resolvedModel, created: false, contextLabel: ctx3.label, numCtx: ctx3.numCtx };
+    return { model: repair.resolvedModel, created: false, contextLabel: ctx3.label, numCtx: ctx3.numCtx, specs, math: ctx3.math };
   }
   const existing = await checkExpandedVariant(modelName, backendUrl2);
   if (existing === null) {
-    return { model: modelName, created: false, contextLabel: "", numCtx: 0 };
+    return { model: modelName, created: false, contextLabel: "", numCtx: 0, specs, math: ctx3.math };
   }
   if (typeof existing === "string") {
     const lostTools = await wrapperLacksToolsCapability(backendUrl2, existing).catch(() => false);
@@ -580249,7 +580312,7 @@ async function ensureExpandedContext(modelName, backendUrl2) {
           kvInfo?.archMax
         );
         if (rebuilt) {
-          return { model: rebuilt, created: true, contextLabel: ctx3.label, numCtx: ctx3.numCtx };
+          return { model: rebuilt, created: true, contextLabel: ctx3.label, numCtx: ctx3.numCtx, specs, math: ctx3.math };
         }
       } catch {
       }
@@ -580264,13 +580327,13 @@ async function ensureExpandedContext(modelName, backendUrl2) {
       kvInfo?.kvBytesPerToken,
       kvInfo?.archMax
     ).catch(() => ({ repaired: false, currentNumCtx: 0, baseModel: null, resolvedModel: existing }));
-    return { model: repair.resolvedModel, created: false, contextLabel: ctx3.label, numCtx: ctx3.numCtx };
+    return { model: repair.resolvedModel, created: false, contextLabel: ctx3.label, numCtx: ctx3.numCtx, specs, math: ctx3.math };
   }
   const created = await createExpandedVariantAsync(modelName, specs, sizeGB, kvInfo?.kvBytesPerToken, kvInfo?.archMax);
   if (created) {
-    return { model: created, created: true, contextLabel: ctx3.label, numCtx: ctx3.numCtx };
+    return { model: created, created: true, contextLabel: ctx3.label, numCtx: ctx3.numCtx, specs, math: ctx3.math };
   }
-  return { model: modelName, created: false, contextLabel: ctx3.label, numCtx: ctx3.numCtx };
+  return { model: modelName, created: false, contextLabel: ctx3.label, numCtx: ctx3.numCtx, specs, math: ctx3.math };
 }
 function guessBaseFromVariant(variantName, models) {
   const stripped = stripVariantTag(variantName);
@@ -598640,8 +598703,8 @@ async function showModelPicker(ctx3, local = false) {
     }
     const items = [];
     const history = loadUsageHistory("model", ctx3.repoRoot);
-    const liveModelNames = new Set(models.map((m2) => m2.name));
-    const modelMap = new Map(models.map((m2) => [m2.name, m2]));
+    const liveModelNames = new Set(models.map((m2) => stripLatest(m2.name)));
+    const modelMap = new Map(models.map((m2) => [stripLatest(m2.name), m2]));
     if (history.length > 0) {
       items.push({
         key: "__header_recent__",
@@ -598650,8 +598713,9 @@ async function showModelPicker(ctx3, local = false) {
       });
       for (const h of history.slice(0, 8)) {
         const uses = h.localUses > 0 ? `${h.useCount} uses (${h.localUses} local)` : `${h.useCount} uses`;
-        const available = liveModelNames.has(h.value) ? "" : c3.yellow(" [offline]");
-        const meta = modelMap.get(h.value);
+        const hKey = stripLatest(h.value);
+        const available = liveModelNames.has(hKey) ? "" : c3.yellow(" [offline]");
+        const meta = modelMap.get(hKey);
         const ctx4 = meta?.contextLength ? `  ${formatContextLength(meta.contextLength)}` : "";
         const capStr = meta?.caps ? `  ${formatCaps(meta.caps)}` : "";
         items.push({
@@ -598666,9 +598730,9 @@ async function showModelPicker(ctx3, local = false) {
         detail: ""
       });
     }
-    const historyKeys = new Set(history.map((h) => h.value));
+    const historyKeys = new Set(history.map((h) => stripLatest(h.value)));
     for (const m2 of models) {
-      if (history.length > 0 && historyKeys.has(m2.name)) continue;
+      if (history.length > 0 && historyKeys.has(stripLatest(m2.name))) continue;
       const ctx4 = m2.contextLength ? formatContextLength(m2.contextLength) : "";
       const capStr = m2.caps ? formatCaps(m2.caps) : "";
       items.push({
@@ -598679,7 +598743,10 @@ async function showModelPicker(ctx3, local = false) {
     }
     const result = await tuiSelect({
       items,
-      activeKey: ctx3.config.model,
+      // `activeKey` is the keyed currently-selected row. The picker stores
+      // history/recent entries with tag-less keys, so normalize the active
+      // model from config to match.
+      activeKey: stripLatest(ctx3.config.model),
       title: "Select Model",
       rl: ctx3.rl,
       // Skip header rows
@@ -598690,7 +598757,7 @@ async function showModelPicker(ctx3, local = false) {
       renderInfo("Model selection cancelled.");
       return;
     }
-    await switchModel(result.key, ctx3, local);
+    await switchModel(stripLatest(result.key), ctx3, local);
   } catch (err) {
     renderError(
       `Failed to fetch models: ${err instanceof Error ? err.message : String(err)}`
@@ -602241,14 +602308,15 @@ async function switchModel(query, ctx3, local = false) {
         match.name,
         ctx3.config.backendUrl
       );
+      const diag = result.specs && result.math ? "\n  " + c3.dim(formatExpandedContextDiagnostic(result.specs, result.math)) : "";
       if (result.created) {
         renderInfo(
-          `Created expanded context variant: ${c3.bold(result.model)} (${result.contextLabel}, ${result.numCtx} tokens)`
+          `Created expanded context variant: ${c3.bold(result.model)} (${result.contextLabel}, ${result.numCtx} tokens)${diag}`
         );
         finalModel = result.model;
       } else if (result.model !== match.name) {
         renderInfo(
-          `Using expanded context variant: ${c3.bold(result.model)} (${result.contextLabel})`
+          `Using expanded context variant: ${c3.bold(result.model)} (${result.contextLabel})${diag}`
         );
         finalModel = result.model;
       }
@@ -653796,13 +653864,14 @@ This is an independent background session started from /background.`
           currentConfig.model,
           currentConfig.backendUrl
         );
+        const diag = expandResult.specs && expandResult.math ? "\n  " + c3.dim(formatExpandedContextDiagnostic(expandResult.specs, expandResult.math)) : "";
         if (expandResult.created) {
           config = { ...config, model: expandResult.model };
           currentConfig = { ...currentConfig, model: expandResult.model };
           statusBar.setModelName(expandResult.model);
           writeContent(
             () => renderInfo(
-              `Created expanded context model: ${expandResult.model} (${expandResult.contextLabel}, ${expandResult.numCtx} tokens)`
+              `Created expanded context model: ${expandResult.model} (${expandResult.contextLabel}, ${expandResult.numCtx} tokens)${diag}`
             )
           );
         } else if (expandResult.model !== currentConfig.model) {
@@ -653811,7 +653880,7 @@ This is an independent background session started from /background.`
           statusBar.setModelName(expandResult.model);
           writeContent(
             () => renderInfo(
-              `Using expanded context model: ${expandResult.model} (${expandResult.contextLabel})`
+              `Using expanded context model: ${expandResult.model} (${expandResult.contextLabel})${diag}`
             )
           );
         }