diff --git a/packages/coding-agent/src/extensions/llama/provider.ts b/packages/coding-agent/src/extensions/llama/provider.ts
index f7801d991..1692a7d45 100644
--- a/packages/coding-agent/src/extensions/llama/provider.ts
+++ b/packages/coding-agent/src/extensions/llama/provider.ts
@@ -111,11 +111,20 @@ export function createLlamaProvider(): LlamaProviderController {
 		},
 		getModels: () => models,
 		refreshModels: async (context: RefreshModelsContext): Promise<void> => {
+			const stored = await context.store.read();
+			if (stored) {
+				models = stored.models.filter(
+					(model): model is Model<"openai-completions"> =>
+						model.provider === LLAMA_PROVIDER_ID && model.api === "openai-completions",
+				);
+			}
+
 			if (!context.allowNetwork || context.signal?.aborted || context.credential?.type !== "api_key") return;
 			const serverUrl = credentialServerUrl(context.credential);
 			if (!serverUrl) return;
 			const catalog = await new LlamaClient(serverUrl, context.credential.key).list({ signal: context.signal });
 			setCatalog(catalog, serverUrl);
+			if (!context.signal?.aborted) await context.store.write({ models, checkedAt: Date.now() });
 		},
 		stream: (model, context, options) => stream(model, context, options as ProviderStreamOptions | undefined),
 		streamSimple: (model, context, options) => streamSimple(model, context, options),
diff --git a/packages/coding-agent/test/llama-extension.test.ts b/packages/coding-agent/test/llama-extension.test.ts
index cb286cba0..3af25e563 100644
--- a/packages/coding-agent/test/llama-extension.test.ts
+++ b/packages/coding-agent/test/llama-extension.test.ts
@@ -1,7 +1,7 @@
 import { once } from "node:events";
 import { createServer, type RequestListener, type Server, type ServerResponse } from "node:http";
 import type { AddressInfo } from "node:net";
-import type { AuthContext, AuthPrompt } from "@earendil-works/pi-ai";
+import type { AuthContext, AuthPrompt, ModelsStoreEntry } from "@earendil-works/pi-ai";
 import { afterEach, describe, expect, it } from "vitest";
 import { createEventBus } from "../src/core/event-bus.ts";
 import { createExtensionRuntime, loadExtensionFromFactory } from "../src/core/extensions/loader.ts";
@@ -86,6 +86,50 @@ describe("llama.cpp extension", () => {
 		]);
 	});
 
+	it("persists and restores loaded models for cache-only startup refreshes", async () => {
+		let cachedEntry: ModelsStoreEntry | undefined;
+		const store = {
+			read: async () => cachedEntry,
+			write: async (entry: ModelsStoreEntry) => {
+				cachedEntry = structuredClone(entry);
+			},
+			delete: async () => {
+				cachedEntry = undefined;
+			},
+		};
+		const { url } = await listen((request, response) => {
+			if (request.url === "/models") {
+				json(response, {
+					data: [
+						{ id: "loaded", status: { value: "loaded" }, meta: { n_ctx: 32768 } },
+						{ id: "unloaded", status: { value: "unloaded" } },
+					],
+				});
+				return;
+			}
+			response.writeHead(404).end();
+		});
+
+		const first = createLlamaProvider();
+		await first.provider.refreshModels?.({
+			credential: { type: "api_key", key: "local", env: { LLAMA_BASE_URL: url } },
+			store,
+			allowNetwork: true,
+		});
+		expect(first.provider.getModels().map((model) => model.id)).toEqual(["loaded"]);
+		expect(cachedEntry?.models.map((model) => model.id)).toEqual(["loaded"]);
+
+		const second = createLlamaProvider();
+		await second.provider.refreshModels?.({
+			credential: { type: "api_key", key: "local", env: { LLAMA_BASE_URL: url } },
+			store,
+			allowNetwork: false,
+		});
+		expect(second.provider.getModels()).toEqual([
+			expect.objectContaining({ id: "loaded", baseUrl: `${url}/v1`, contextWindow: 32768 }),
+		]);
+	});
+
 	it("stays dormant until configured and stores URL plus optional key", async () => {
 		const { provider } = createLlamaProvider();
 		const auth = provider.auth.apiKey!;