fix: honor Ollama Modelfile num_ctx discovery

extensions/ollama/src/provider-models.test.ts

@@ -3,6 +3,7 @@ import { jsonResponse, requestBodyText, requestUrl } from "../../../src/test-hel
 import {
   buildOllamaModelDefinition,
   enrichOllamaModelsWithContext,
+  parseOllamaNumCtxParameter,
   resetOllamaModelShowInfoCacheForTest,
   resolveOllamaApiBase,
   type OllamaTagModel,
@@ -42,6 +43,58 @@ describe("ollama provider models", () => {
     ]);
   });
 
+  it("uses Modelfile num_ctx when it expands the discovered context window", async () => {
+    const models: OllamaTagModel[] = [{ name: "llama3-32k:latest" }];
+    const fetchMock = vi.fn(async () =>
+      jsonResponse({
+        model_info: { "llama.context_length": 8192 },
+        parameters: 'stop "<|eot_id|>"\nnum_ctx 32768\nnum_keep 5',
+        capabilities: ["completion"],
+      }),
+    );
+    vi.stubGlobal("fetch", fetchMock);
+
+    const enriched = await enrichOllamaModelsWithContext("http://127.0.0.1:11434", models);
+
+    expect(enriched).toEqual([
+      {
+        name: "llama3-32k:latest",
+        contextWindow: 32768,
+        capabilities: ["completion"],
+      },
+    ]);
+  });
+
+  it("keeps the larger native context window when Modelfile num_ctx is smaller", async () => {
+    const models: OllamaTagModel[] = [{ name: "llama3.2:latest" }];
+    const fetchMock = vi.fn(async () =>
+      jsonResponse({
+        model_info: { "llama.context_length": 131072 },
+        parameters: "num_ctx 4096",
+      }),
+    );
+    vi.stubGlobal("fetch", fetchMock);
+
+    const enriched = await enrichOllamaModelsWithContext("http://127.0.0.1:11434", models);
+
+    expect(enriched[0]?.contextWindow).toBe(131072);
+  });
+
+  it("uses positive num_ctx when /api/show omits model context metadata", async () => {
+    const models: OllamaTagModel[] = [{ name: "custom-model:latest" }];
+    const fetchMock = vi.fn(async () =>
+      jsonResponse({
+        model_info: {},
+        parameters: "num_ctx 16384",
+      }),
+    );
+    vi.stubGlobal("fetch", fetchMock);
+
+    const enriched = await enrichOllamaModelsWithContext("http://127.0.0.1:11434", models);
+
+    expect(enriched[0]?.contextWindow).toBe(16384);
+  });
+
   it("sets models with vision capability from /api/show capabilities", async () => {
     const models: OllamaTagModel[] = [{ name: "kimi-k2.5:cloud" }, { name: "glm-5.1:cloud" }];
     const fetchMock = vi.fn(async (input: string | URL | Request, init?: RequestInit) => {
@@ -225,4 +278,11 @@ describe("ollama provider models", () => {
     expect(model.reasoning).toBe(false);
     expect(model.compat?.supportsTools).toBe(false);
   });
+
+  it("parses the last positive Modelfile num_ctx value", () => {
+    expect(parseOllamaNumCtxParameter("num_ctx 8192\nnum_ctx 32768")).toBe(32768);
+    expect(parseOllamaNumCtxParameter("temperature 0.8\nnum_ctx -1\nnum_ctx 0")).toBeUndefined();
+    expect(parseOllamaNumCtxParameter('stop "<|eot_id|>"')).toBeUndefined();
+    expect(parseOllamaNumCtxParameter({ num_ctx: 8192 })).toBeUndefined();
+  });
 });

extensions/ollama/src/provider-models.ts

@@ -95,6 +95,25 @@ function hasCachedOllamaModelShowInfo(info: OllamaModelShowInfo): boolean {
   return typeof info.contextWindow === "number" || (info.capabilities?.length ?? 0) > 0;
 }
 
+export function parseOllamaNumCtxParameter(parameters: unknown): number | undefined {
+  if (typeof parameters !== "string" || !parameters.trim()) {
+    return undefined;
+  }
+
+  let lastValue: number | undefined;
+  for (const rawLine of parameters.split(/\r?\n/)) {
+    const match = rawLine.trim().match(/^num_ctx\s+(-?\d+)\b/);
+    if (!match) {
+      continue;
+    }
+    const parsed = Number.parseInt(match[1], 10);
+    if (Number.isFinite(parsed) && parsed > 0) {
+      lastValue = parsed;
+    }
+  }
+  return lastValue;
+}
+
 export async function queryOllamaModelShowInfo(
   apiBase: string,
   modelName: string,
@@ -119,6 +138,7 @@ export async function queryOllamaModelShowInfo(
       const data = (await response.json()) as {
         model_info?: Record<string, unknown>;
         capabilities?: unknown;
+        parameters?: unknown;
       };
 
       let contextWindow: number | undefined;
@@ -138,6 +158,11 @@ export async function queryOllamaModelShowInfo(
         }
       }
 
+      const paramCtx = parseOllamaNumCtxParameter(data.parameters);
+      if (paramCtx !== undefined && (contextWindow === undefined || paramCtx > contextWindow)) {
+        contextWindow = paramCtx;
+      }
+
       const capabilities = Array.isArray(data.capabilities)
         ? (data.capabilities as unknown[]).filter((c): c is string => typeof c === "string")
         : undefined;