@geminilight/mindos 0.6.61 → 0.6.65

package/app/components/shared/ProviderSelect.tsx

@@ -1,55 +1,84 @@
 'use client';
 
 import { useState } from 'react';
-import { CheckCircle2, ChevronDown, SkipForward } from 'lucide-react';
-import { type ProviderId, PROVIDER_PRESETS, groupedProviders } from '@/lib/agent/providers';
+import { CheckCircle2, ChevronDown, SkipForward, Plus } from 'lucide-react';
+import { type ProviderId, PROVIDER_PRESETS, groupedProviders, ALL_PROVIDER_IDS } from '@/lib/agent/providers';
+import { type Provider } from '@/lib/custom-endpoints';
 import { useLocale } from '@/lib/stores/locale-store';
 
 interface ProviderSelectProps {
-  value: ProviderId | 'skip';
-  onChange: (id: ProviderId | 'skip') => void;
+  value: string | 'skip';
+  onChange: (id: string | 'skip') => void;
   showSkip?: boolean;
   compact?: boolean;
+  /** @deprecated Use customProviders (unified Provider[]) instead */
   configuredProviders?: Set<ProviderId>;
+  customProviders?: Provider[];
+  onAdd?: () => void;
 }
 
 export default function ProviderSelect({
   value, onChange, showSkip = false, compact = false, configuredProviders,
+  customProviders, onAdd,
 }: ProviderSelectProps) {
   const { locale } = useLocale();
   const [showMore, setShowMore] = useState(false);
   const groups = groupedProviders();
 
-  const renderItem = (id: ProviderId) => {
+  const hasConfigured = configuredProviders && configuredProviders.size > 0;
+  const hasCustom = customProviders && customProviders.length > 0;
+
+  // In compact settings mode: show provider list + Add button
+  // New model: customProviders IS the full list (unified Provider[])
+  // Legacy model: configuredProviders set + separate customProviders array
+  const useConfiguredMode = compact && (hasConfigured || hasCustom) && !showSkip;
+
+  // Legacy: Sorted configured provider IDs (for backward compat with old callers)
+  const configuredIds = hasConfigured
+    ? ALL_PROVIDER_IDS.filter(id => configuredProviders!.has(id))
+    : [];
+
+  // Add panel shows ALL providers as protocol templates (can add multiple of the same type)
+  const { primary: primaryItems, more: moreItems } = groups;
+
+  /* ── Compact tab button (for legacy builtin-only mode) ── */
+  const renderCompactTab = (id: ProviderId) => {
     const preset = PROVIDER_PRESETS[id];
     const displayName = locale === 'zh' ? preset.nameZh : preset.name;
     const isSelected = value === id;
     const isConfigured = configuredProviders?.has(id);
 
-    if (compact) {
-      return (
-        <button
-          key={id}
-          type="button"
-          onClick={() => onChange(id)}
-          className={`flex items-center gap-2 px-3 py-2 rounded-lg border text-left transition-all text-sm ${
-            isSelected
-              ? 'border-[var(--amber)] bg-[var(--amber-subtle)] shadow-sm'
-              : 'border-border/50 hover:border-border hover:bg-muted/30'
-          }`}
-        >
-          <span className={`font-medium ${isSelected ? 'text-foreground' : 'text-muted-foreground'}`}>
-            {displayName}
-          </span>
-          {isConfigured && !isSelected && (
-            <CheckCircle2 size={12} className="text-success ml-auto shrink-0" />
-          )}
-          {isSelected && (
-            <CheckCircle2 size={14} className="ml-auto shrink-0" style={{ color: 'var(--amber)' }} />
-          )}
-        </button>
-      );
-    }
+    return (
+      <button
+        key={id}
+        type="button"
+        onClick={() => onChange(id)}
+        className={`flex items-center gap-2 px-3 py-2 rounded-lg border text-left transition-all text-sm ${
+          isSelected
+            ? 'border-[var(--amber)] bg-[var(--amber-subtle)] shadow-sm'
+            : 'border-border/50 hover:border-border hover:bg-muted/30'
+        }`}
+      >
+        <span className={`font-medium ${isSelected ? 'text-foreground' : 'text-muted-foreground'}`}>
+          {displayName}
+        </span>
+        {isConfigured && !isSelected && (
+          <CheckCircle2 size={12} className="text-success ml-auto shrink-0" />
+        )}
+        {isSelected && (
+          <CheckCircle2 size={14} className="ml-auto shrink-0" style={{ color: 'var(--amber)' }} />
+        )}
+      </button>
+    );
+  };
+
+  /* ── Full card button (used in setup wizard / non-compact) ── */
+  const renderCard = (id: ProviderId) => {
+    const preset = PROVIDER_PRESETS[id];
+    const displayName = locale === 'zh' ? preset.nameZh : preset.name;
+    const description = locale === 'zh' ? preset.descriptionZh : preset.description;
+    const isSelected = value === id;
+    const isConfigured = configuredProviders?.has(id);
 
     return (
       <button
@@ -64,7 +93,10 @@ export default function ProviderSelect({
       >
         <div className="flex-1 min-w-0">
           <p className="text-sm font-medium" style={{ color: 'var(--foreground)' }}>{displayName}</p>
-          <p className="text-xs mt-0.5" style={{ color: 'var(--muted-foreground)' }}>
+          {description && (
+            <p className="text-xs mt-0.5" style={{ color: 'var(--muted-foreground)' }}>{description}</p>
+          )}
+          <p className={`text-xs ${description ? 'mt-1' : 'mt-0.5'}`} style={{ color: 'var(--muted-foreground)' }}>
             {preset.defaultModel}
           </p>
         </div>
@@ -78,16 +110,71 @@ export default function ProviderSelect({
     );
   };
 
-  const { primary: primaryItems, more: moreItems } = groups;
+  /* ════════════════════════════════════════════
+   *  MODE 1: Provider list + Add button
+   *  (compact settings, has providers)
+   * ════════════════════════════════════════════ */
+  if (useConfiguredMode) {
+    return (
+      <div className="space-y-2">
+        {/* Providers row */}
+        <div className="flex flex-wrap gap-2">
+          {/* Legacy: built-in configured providers */}
+          {configuredIds.map(id => renderCompactTab(id))}
+
+          {/* Unified provider list (or legacy custom providers) */}
+          {customProviders?.map(cp => {
+            const isSelected = value === cp.id;
+            return (
+              <button
+                key={cp.id}
+                type="button"
+                onClick={() => onChange(cp.id)}
+                className={`flex items-center gap-2 px-3 py-2 rounded-lg border text-left transition-all text-sm ${
+                  isSelected
+                    ? 'border-[var(--amber)] bg-[var(--amber-subtle)] shadow-sm'
+                    : 'border-border/50 hover:border-border hover:bg-muted/30'
+                }`}
+              >
+                <span className={`font-medium ${isSelected ? 'text-foreground' : 'text-muted-foreground'}`}>
+                  {cp.name}
+                </span>
+                {isSelected && (
+                  <CheckCircle2 size={14} className="ml-auto shrink-0" style={{ color: 'var(--amber)' }} />
+                )}
+              </button>
+            );
+          })}
+
+          {/* Add button — opens form directly */}
+          {onAdd && (
+            <button
+              type="button"
+              onClick={onAdd}
+              className="flex items-center gap-1.5 px-3 py-2 rounded-lg border border-dashed border-border/50 text-sm text-muted-foreground hover:border-border hover:text-foreground transition-all"
+            >
+              <Plus size={14} />
+              <span>{locale === 'zh' ? '添加' : 'Add'}</span>
+            </button>
+          )}
+        </div>
+      </div>
+    );
+  }
+
+  /* ════════════════════════════════════════════
+   *  MODE 2: Full list (setup wizard / no configured providers)
+   *  Original behavior preserved
+   * ════════════════════════════════════════════ */
 
   return (
     <div className="space-y-2">
-      {/* Primary providers — always visible */}
+      {/* Primary providers */}
       <div className={compact ? 'flex flex-wrap gap-2' : 'grid grid-cols-1 gap-2'}>
-        {primaryItems.map(renderItem)}
+        {primaryItems.map(id => compact ? renderCompactTab(id) : renderCard(id))}
       </div>
 
-      {/* Show more toggle */}
+      {/* More toggle */}
       {moreItems.length > 0 && (
         <>
           <button
@@ -98,12 +185,14 @@ export default function ProviderSelect({
             <ChevronDown size={12} className={`transition-transform ${showMore ? 'rotate-180' : ''}`} />
             {showMore
               ? (locale === 'zh' ? '收起' : 'Show less')
-              : (locale === 'zh' ? `更多 (${moreItems.length})` : `More providers (${moreItems.length})`)}
+              : (locale === 'zh'
+                  ? `更多 (${moreItems.length})`
+                  : `More (${moreItems.length})`)}
           </button>
 
           {showMore && (
             <div className={compact ? 'flex flex-wrap gap-2' : 'grid grid-cols-1 gap-2'}>
-              {moreItems.map(renderItem)}
+              {moreItems.map(id => compact ? renderCompactTab(id) : renderCard(id))}
             </div>
           )}
         </>
@@ -129,6 +218,7 @@ export default function ProviderSelect({
           )}
         </button>
       )}
+
     </div>
   );
 }

package/app/hooks/useAskChat.ts

@@ -28,21 +28,25 @@ export interface AskChatRefs {
 interface UseAskChatOpts {
   currentFile?: string;
   chatMode: AskMode;
-  providerOverride: ProviderId | null;
+  providerOverride: ProviderId | `p_${string}` | null;
+  modelOverride: string | null;
   onFirstMessage?: () => void;
   refs: AskChatRefs;
   errorLabels: { noResponse: string; stopped: string };
   resetInputState: () => void;
+  onRestoreInput?: (userMessage: Message) => void;
 }
 
 export function useAskChat({
   currentFile,
   chatMode,
   providerOverride,
+  modelOverride,
   onFirstMessage,
   refs,
   errorLabels,
   resetInputState,
+  onRestoreInput,
 }: UseAskChatOpts) {
   const [isLoading, setIsLoading] = useState(false);
   const [loadingPhase, setLoadingPhase] = useState<LoadingPhase>('connecting');
@@ -51,10 +55,68 @@ export function useAskChat({
   const abortRef = useRef<AbortController | null>(null);
   const firstMessageFired = useRef(false);
 
-  const stop = useCallback(() => { abortRef.current?.abort(); }, []);
+  // Cooldown guard: after stop+retract, briefly block re-submission so that
+  // the mouseup on the stop-button position doesn't accidentally trigger the
+  // send button that React swaps in at the same DOM position.
+  const submitCooldownRef = useRef(false);
+
+  // Track the pending user message so we can retract it on stop.
+  // `userMessageIndex` is the index of the *user* message inside the messages
+  // array (the assistant placeholder sits at userMessageIndex + 1).
+  const pendingMessageRef = useRef<{
+    userMessageIndex: number;
+    userMessage: Message;
+  } | null>(null);
+
+  // When true the AbortError handler in submit() skips its own setMessages
+  // because stop() already cleaned up the messages array.
+  const retractedRef = useRef(false);
+
+  const stop = useCallback(() => {
+    const pending = pendingMessageRef.current;
+
+    // Abort the fetch first.
+    abortRef.current?.abort();
+
+    if (pending) {
+      retractedRef.current = true;
+
+      // Always remove the user message + assistant response (empty or partial)
+      // from the messages array. The user clicked stop — they don't want this
+      // exchange in the history at all.
+      refs.sessionRef.current?.setMessages(prev => {
+        const updated = [...prev];
+        const idx = pending.userMessageIndex;
+
+        // Remove assistant message first (at idx + 1) — may be empty placeholder
+        // or a partial streamed response.
+        if (idx + 1 < updated.length && updated[idx + 1]?.role === 'assistant') {
+          updated.splice(idx + 1, 1);
+        }
+
+        // Remove the user message.
+        if (idx < updated.length && updated[idx]?.role === 'user') {
+          updated.splice(idx, 1);
+        }
+
+        return updated;
+      });
+
+      // Restore text (+ attachments) back into the input box.
+      onRestoreInput?.(pending.userMessage);
+
+      pendingMessageRef.current = null;
+
+      // Block re-submission for a short window so the browser's mouseup
+      // doesn't hit the send button that replaces the stop button.
+      submitCooldownRef.current = true;
+      setTimeout(() => { submitCooldownRef.current = false; }, 300);
+    }
+  }, [refs, onRestoreInput]);
 
   const submit = useCallback(async (e: React.FormEvent) => {
     e.preventDefault();
+    if (submitCooldownRef.current) return; // ignore accidental re-submit after stop
     const m = refs.mentionRef.current;
     const s = refs.slashRef.current;
     const img = refs.imageUploadRef.current;
@@ -68,15 +130,33 @@ export function useAskChat({
     const skill = refs.selectedSkillRef.current;
     const acpAgent = refs.selectedAcpAgentRef.current;
     const pendingImages = img.images.length > 0 ? [...img.images] : undefined;
+    // Only store explicitly user-chosen files (filter out auto-included currentFile)
+    const explicitAttached = refs.attachedFilesRef.current.filter(f => f !== currentFile);
+    const pendingAttachedFiles = explicitAttached.length > 0 ? explicitAttached : undefined;
+    const pendingUploadedNames = upl.localAttachments
+      .filter(f => f.status !== 'loading')
+      .map(f => f.name);
     const userMsg: Message = {
       role: 'user',
       content: text,
       timestamp: Date.now(),
       ...(skill && { skillName: skill.name }),
       ...(pendingImages && { images: pendingImages }),
+      ...(pendingAttachedFiles && { attachedFiles: pendingAttachedFiles }),
+      ...(pendingUploadedNames.length > 0 && { uploadedFileNames: pendingUploadedNames }),
     };
     img.clearImages();
     const requestMessages = [...sess.messages, userMsg];
+
+    // Track the user message index for potential retraction on stop.
+    // The user message is at requestMessages.length - 1; the assistant
+    // placeholder we're about to insert will be at requestMessages.length.
+    pendingMessageRef.current = {
+      userMessageIndex: requestMessages.length - 1,
+      userMessage: userMsg,
+    };
+    retractedRef.current = false;
+
     sess.setMessages([...requestMessages, { role: 'assistant', content: '', timestamp: Date.now() }]);
 
     resetInputState();
@@ -114,6 +194,7 @@ export function useAskChat({
       selectedAcpAgent: acpAgent,
       mode: chatMode,
       providerOverride: providerOverride ?? undefined,
+      modelOverride: modelOverride ?? undefined,
     });
 
     const doFetch = async (): Promise<{ finalMessage: Message }> => {
@@ -187,6 +268,8 @@ export function useAskChat({
               return updated;
             });
           }
+          // Successfully received response — no longer retractable.
+          pendingMessageRef.current = null;
           return;
         } catch (err) {
           lastError = err instanceof Error ? err : new Error(String(err));
@@ -198,18 +281,21 @@ export function useAskChat({
       if (lastError) throw lastError;
     } catch (err) {
       if ((err as Error).name === 'AbortError') {
-        refs.sessionRef.current?.setMessages(prev => {
-          const updated = [...prev];
-          const lastIdx = updated.length - 1;
-          if (lastIdx >= 0 && updated[lastIdx].role === 'assistant') {
-            const last = updated[lastIdx];
-            const hasContent = last.content.trim() || (last.parts && last.parts.length > 0);
-            if (!hasContent) {
-              updated[lastIdx] = { role: 'assistant', content: `__error__${errorLabels.stopped}` };
+        // If stop() already retracted the messages, skip writing __error__stopped.
+        if (!retractedRef.current) {
+          refs.sessionRef.current?.setMessages(prev => {
+            const updated = [...prev];
+            const lastIdx = updated.length - 1;
+            if (lastIdx >= 0 && updated[lastIdx].role === 'assistant') {
+              const last = updated[lastIdx];
+              const hasContent = last.content.trim() || (last.parts && last.parts.length > 0);
+              if (!hasContent) {
+                updated[lastIdx] = { role: 'assistant', content: `__error__${errorLabels.stopped}` };
+              }
             }
-          }
-          return updated;
-        });
+            return updated;
+          });
+        }
       } else {
         const errMsg = err instanceof Error ? err.message : 'Something went wrong';
         refs.sessionRef.current?.setMessages(prev => {
@@ -230,6 +316,7 @@ export function useAskChat({
       setIsLoading(false);
       setReconnectAttempt(0);
       abortRef.current = null;
+      pendingMessageRef.current = null;
     }
   }, [currentFile, chatMode, providerOverride, errorLabels.noResponse, errorLabels.stopped, onFirstMessage, refs, resetInputState]);
 

package/app/hooks/useAskPanel.ts

@@ -33,6 +33,8 @@ export function useAskPanel(): AskPanelState {
   const [desktopAskPopupOpen, setDesktopAskPopupOpen] = useState(false);
   const [askInitialMessage, setAskInitialMessage] = useState('');
   const [askMaximized, setAskMaximized] = useState(false);
+  const askMaximizedRef = useRef(false);
+  askMaximizedRef.current = askMaximized;
   const [askOpenSource, setAskOpenSource] = useState<'user' | 'guide' | 'guide-next'>('user');
   const [askAcpAgent, setAskAcpAgent] = useState<AcpAgentSelection | null>(null);
   const prevWidthRef = useRef(RIGHT_ASK_DEFAULT_WIDTH);
@@ -60,7 +62,21 @@ export function useAskPanel(): AskPanelState {
       }
     };
     window.addEventListener('storage', onStorage);
-    return () => window.removeEventListener('storage', onStorage);
+
+    // Listen for "dock to panel" from home page fullscreen
+    const onOpenPanel = () => {
+      setAskPanelOpen(true);
+      if (askMaximizedRef.current) {
+        setAskMaximized(false);
+        setAskPanelWidth(prevWidthRef.current);
+      }
+    };
+    window.addEventListener('mindos:open-ask-panel', onOpenPanel);
+
+    return () => {
+      window.removeEventListener('storage', onStorage);
+      window.removeEventListener('mindos:open-ask-panel', onOpenPanel);
+    };
   }, []);
 
   // Bridge useAskModal store → right Ask panel or popup

package/app/lib/acp/registry.ts

@@ -2,10 +2,15 @@
  * ACP Registry Client — Fetch and cache the ACP agent registry.
  * The registry lists available ACP agents (Gemini CLI, Claude, Copilot, etc.)
  * with their transport type, command, and metadata.
+ *
+ * Strategy: Built-in registry from AGENT_DESCRIPTORS is always available as a
+ * baseline. CDN registry is fetched in the background and merged on top —
+ * CDN entries update existing ones, new CDN-only entries are appended.
+ * If CDN is unreachable (e.g. in China), the built-in list still works.
  */
 
 import type { AcpRegistry, AcpRegistryEntry } from './types';
-import { getDescriptorDisplayName, getDescriptorDescription } from './agent-descriptors';
+import { AGENT_DESCRIPTORS, getDescriptorDisplayName, getDescriptorDescription } from './agent-descriptors';
 
 /* ── Constants ─────────────────────────────────────────────────────────── */
 
@@ -13,6 +18,58 @@ const REGISTRY_URL = 'https://cdn.agentclientprotocol.com/registry/v1/latest/reg
 const CACHE_TTL_MS = 60 * 60 * 1000; // 1 hour
 const FETCH_TIMEOUT_MS = 10_000;
 
+/* ── Built-in Registry (from AGENT_DESCRIPTORS) ────────────────────────── */
+
+/**
+ * Generate a baseline registry from the local AGENT_DESCRIPTORS.
+ * This ensures agents like Gemini CLI, CodeBuddy, Claude etc. are always
+ * detectable even when CDN is unreachable.
+ *
+ * We deduplicate by binary name — e.g. 'gemini' and 'gemini-cli' both map
+ * to binary 'gemini', so we only keep the canonical entry (shorter ID or
+ * the one matching the CDN convention).
+ */
+function buildBuiltinRegistry(): AcpRegistryEntry[] {
+  // Preferred IDs per binary — matches what CDN uses
+  const CANONICAL_IDS: Record<string, string> = {
+    'gemini': 'gemini',
+    'claude': 'claude-acp',
+    'codebuddy': 'codebuddy-code',
+    'codex': 'codex-acp',
+    'pi': 'pi-acp',
+  };
+
+  const seen = new Set<string>();
+  const entries: AcpRegistryEntry[] = [];
+
+  for (const [id, desc] of Object.entries(AGENT_DESCRIPTORS)) {
+    // Skip alias entries — only keep the canonical ID for each binary
+    const canonical = CANONICAL_IDS[desc.binary];
+    if (canonical && canonical !== id) continue;
+    if (seen.has(desc.binary)) continue;
+    seen.add(desc.binary);
+
+    entries.push({
+      id,
+      name: desc.displayName ?? id,
+      description: desc.description ?? '',
+      transport: desc.cmd === 'npx' ? 'npx' : 'stdio',
+      command: desc.cmd,
+      args: desc.args,
+      packageName: desc.installCmd?.match(/npm install -g (.+)/)?.[1],
+    });
+  }
+
+  return entries;
+}
+
+let builtinAgents: AcpRegistryEntry[] | null = null;
+
+function getBuiltinAgents(): AcpRegistryEntry[] {
+  if (!builtinAgents) builtinAgents = buildBuiltinRegistry();
+  return builtinAgents;
+}
+
 /* ── Cache ─────────────────────────────────────────────────────────────── */
 
 let cachedRegistry: AcpRegistry | null = null;
@@ -20,47 +77,72 @@ let cachedRegistry: AcpRegistry | null = null;
 /* ── Public API ────────────────────────────────────────────────────────── */
 
 /**
- * Fetch the ACP registry from the CDN. Caches for 1 hour.
- * Returns null if the fetch fails.
+ * Fetch the ACP registry from the CDN and merge with built-in entries.
+ * Caches for 1 hour. Falls back to built-in registry if CDN is unreachable.
  */
-export async function fetchAcpRegistry(): Promise<AcpRegistry | null> {
+export async function fetchAcpRegistry(): Promise<AcpRegistry> {
   // Return cached if still valid
   if (cachedRegistry && Date.now() - new Date(cachedRegistry.fetchedAt).getTime() < CACHE_TTL_MS) {
     return cachedRegistry;
   }
 
+  const builtin = getBuiltinAgents();
+
   try {
     const res = await fetch(REGISTRY_URL, {
       headers: { 'Accept': 'application/json' },
       signal: AbortSignal.timeout(FETCH_TIMEOUT_MS),
     });
 
-    if (!res.ok) return cachedRegistry ?? null;
+    if (!res.ok) {
+      return cachedRegistry ?? makeRegistry(builtin, 'builtin');
+    }
 
     const data = await res.json();
 
     // The registry JSON may have varying shapes; normalize it
-    const agents: AcpRegistryEntry[] = parseRegistryEntries(data);
+    const cdnAgents: AcpRegistryEntry[] = parseRegistryEntries(data);
+
+    // Merge: CDN entries take precedence, built-in entries fill gaps
+    const merged = mergeRegistries(builtin, cdnAgents);
 
     cachedRegistry = {
       version: data.version ?? '1',
-      agents,
+      agents: merged,
       fetchedAt: new Date().toISOString(),
     };
 
     return cachedRegistry;
   } catch {
-    // Return stale cache if available, otherwise null
-    return cachedRegistry ?? null;
+    // CDN unreachable — use built-in as baseline
+    return cachedRegistry ?? makeRegistry(builtin, 'builtin');
   }
 }
 
+/** Merge built-in and CDN registries. CDN entries win on conflict; built-in fills gaps. */
+function mergeRegistries(builtin: AcpRegistryEntry[], cdn: AcpRegistryEntry[]): AcpRegistryEntry[] {
+  const byId = new Map<string, AcpRegistryEntry>();
+
+  // Start with built-in
+  for (const entry of builtin) byId.set(entry.id, entry);
+
+  // CDN overwrites / adds
+  for (const entry of cdn) byId.set(entry.id, entry);
+
+  return Array.from(byId.values());
+}
+
+function makeRegistry(agents: AcpRegistryEntry[], version: string): AcpRegistry {
+  return { version, agents, fetchedAt: new Date().toISOString() };
+}
+
 /**
  * Get all available ACP agents from the registry.
+ * Always returns at least the built-in agents, even if CDN is unreachable.
  */
 export async function getAcpAgents(): Promise<AcpRegistryEntry[]> {
   const registry = await fetchAcpRegistry();
-  return registry?.agents ?? [];
+  return registry.agents;
 }
 
 /**

package/app/lib/agent/context.ts

@@ -62,6 +62,71 @@ export function estimateTokens(messages: AgentMessage[]): number {
 // Context limits by model family
 // ---------------------------------------------------------------------------
 
+/**
+ * Cache for Ollama model context window sizes.
+ * Key: "baseUrl::modelName", Value: num_ctx (token count).
+ * Avoids querying /api/show on every request.
+ */
+const ollamaContextCache = new Map<string, number>();
+
+/**
+ * Query Ollama's /api/show endpoint to get the actual context window (num_ctx)
+ * for a specific model. Returns undefined if the query fails or the model
+ * doesn't report its context size.
+ *
+ * Results are cached in-memory per baseUrl+model for the process lifetime.
+ */
+export async function getOllamaContextWindow(
+  baseUrl: string,
+  modelName: string,
+): Promise<number | undefined> {
+  const cacheKey = `${baseUrl}::${modelName}`;
+  if (ollamaContextCache.has(cacheKey)) return ollamaContextCache.get(cacheKey);
+
+  // Ollama's /api/show endpoint is at the root, not under /v1
+  const ollamaBase = baseUrl.replace(/\/v1\/?$/, '');
+  try {
+    const resp = await fetch(`${ollamaBase}/api/show`, {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({ name: modelName }),
+      signal: AbortSignal.timeout(3000),
+    });
+    if (!resp.ok) return undefined;
+
+    const data = await resp.json() as {
+      model_info?: Record<string, unknown>;
+      parameters?: string;
+    };
+
+    // Method 1: model_info (structured, most reliable)
+    let numCtx: number | undefined;
+    if (data.model_info) {
+      // Different Ollama versions use different key names
+      const ctxVal = data.model_info['context_length']
+        ?? data.model_info['num_ctx']
+        ?? data.model_info['general.context_length'];
+      if (typeof ctxVal === 'number' && ctxVal > 0) {
+        numCtx = ctxVal;
+      }
+    }
+
+    // Method 2: parse from parameters string (fallback)
+    if (!numCtx && typeof data.parameters === 'string') {
+      const match = data.parameters.match(/num_ctx\s+(\d+)/);
+      if (match) numCtx = parseInt(match[1], 10);
+    }
+
+    if (numCtx && numCtx > 0) {
+      ollamaContextCache.set(cacheKey, numCtx);
+      return numCtx;
+    }
+  } catch {
+    // Network error, timeout, or Ollama not running — fail silently
+  }
+  return undefined;
+}
+
 const MODEL_LIMITS: Record<string, number> = {
   'claude': 200_000,
   'gpt-4o': 128_000,