feat(memory): surface llama.cpp diagnostics

extensions/llama-cpp/index.test.ts

@@ -14,6 +14,7 @@ import { afterEach, describe, expect, it, vi } from "vitest";
 const memoryHostEmbeddingMocks = vi.hoisted(() => ({
   createLocalEmbeddingProvider: vi.fn(),
 }));
+const LOCAL_EMBEDDING_RUNTIME_FACTS = Symbol.for("openclaw.localEmbeddingRuntimeFacts");
 
 vi.mock("openclaw/plugin-sdk/memory-core-host-engine-embeddings", () => ({
   createLocalEmbeddingProvider: memoryHostEmbeddingMocks.createLocalEmbeddingProvider,
@@ -59,7 +60,13 @@ describe("llama.cpp provider plugin", () => {
 
   it("adapts the worker-backed local embedding provider", async () => {
     const close = vi.fn();
-    memoryHostEmbeddingMocks.createLocalEmbeddingProvider.mockResolvedValue({
+    const getRuntimeFacts = vi.fn(() => ({
+      engine: "llama.cpp" as const,
+      state: "ready" as const,
+      backend: "metal" as const,
+      buildType: "prebuilt" as const,
+    }));
+    const workerProvider = {
       id: "local",
       model: DEFAULT_LLAMA_CPP_EMBEDDING_MODEL,
       maxInputTokens: 2048,
@@ -67,7 +74,11 @@ describe("llama.cpp provider plugin", () => {
       embedBatchInputs: vi.fn(async () => [[0.3, 0.4]]),
       embedBatch: vi.fn(async () => [[1, 0]]),
       close,
+    };
+    Object.defineProperty(workerProvider, LOCAL_EMBEDDING_RUNTIME_FACTS, {
+      value: getRuntimeFacts,
     });
+    memoryHostEmbeddingMocks.createLocalEmbeddingProvider.mockResolvedValue(workerProvider);
     const abortController = new AbortController();
 
     const result = await llamaCppEmbeddingProviderAdapter.create({
@@ -89,6 +100,20 @@ describe("llama.cpp provider plugin", () => {
 
     expect(provider.model).toBe(DEFAULT_LLAMA_CPP_EMBEDDING_MODEL);
     expect(provider.maxInputTokens).toBe(2048);
+    const adaptedGetRuntimeFacts = Reflect.get(provider, LOCAL_EMBEDDING_RUNTIME_FACTS);
+    if (typeof adaptedGetRuntimeFacts !== "function") {
+      throw new Error("expected llama.cpp runtime facts carrier");
+    }
+    expect(adaptedGetRuntimeFacts()).toEqual({
+      engine: "llama.cpp",
+      state: "ready",
+      backend: "metal",
+      buildType: "prebuilt",
+    });
+    expect(result.runtime?.cacheKeyData).toEqual({
+      provider: "local",
+      model: DEFAULT_LLAMA_CPP_EMBEDDING_MODEL,
+    });
     expect(close).toHaveBeenCalledTimes(1);
     expect(memoryHostEmbeddingMocks.createLocalEmbeddingProvider).toHaveBeenCalledWith(
       {
@@ -104,9 +129,9 @@ describe("llama.cpp provider plugin", () => {
         nodeLlamaCppImportUrl: expect.stringContaining("node-llama-cpp"),
       },
     );
-    const workerProvider =
+    const createdWorkerProvider =
       await memoryHostEmbeddingMocks.createLocalEmbeddingProvider.mock.results[0].value;
-    expect(workerProvider.embedBatchInputs).toHaveBeenCalledWith([{ text: "doc" }], {
+    expect(createdWorkerProvider.embedBatchInputs).toHaveBeenCalledWith([{ text: "doc" }], {
       signal: abortController.signal,
     });
   });

extensions/llama-cpp/src/embedding-provider.ts

@@ -28,6 +28,7 @@ type LlamaCppEmbeddingProviderRuntimeOptions = {
 };
 
 const LLAMA_CPP_EMBEDDING_PROVIDER_ID = "local";
+const LOCAL_EMBEDDING_RUNTIME_FACTS = Symbol.for("openclaw.localEmbeddingRuntimeFacts");
 export const DEFAULT_LLAMA_CPP_EMBEDDING_MODEL =
   "hf:ggml-org/embeddinggemma-300m-qat-q8_0-GGUF/embeddinggemma-300m-qat-Q8_0.gguf";
 const DEFAULT_LLAMA_CPP_EMBEDDING_MODEL_CACHE_FILE_NAME =
@@ -159,8 +160,18 @@ function resolveNodeLlamaCppImportUrl(): string {
   return pathToFileURL(requireFromPlugin.resolve("node-llama-cpp")).href;
 }
 
+function copyLocalRuntimeFacts(source: object, target: object): void {
+  const getRuntimeFacts = Reflect.get(source, LOCAL_EMBEDDING_RUNTIME_FACTS);
+  if (typeof getRuntimeFacts === "function") {
+    Object.defineProperty(target, LOCAL_EMBEDDING_RUNTIME_FACTS, {
+      enumerable: false,
+      value: getRuntimeFacts,
+    });
+  }
+}
+
 function adaptMemoryEmbeddingProvider(provider: MemoryEmbeddingProvider): EmbeddingProvider {
-  return {
+  const adapted: EmbeddingProvider = {
     id: LLAMA_CPP_EMBEDDING_PROVIDER_ID,
     model: provider.model,
     maxInputTokens: provider.maxInputTokens,
@@ -180,6 +191,8 @@ function adaptMemoryEmbeddingProvider(provider: MemoryEmbeddingProvider): Embedd
     },
     close: provider.close,
   };
+  copyLocalRuntimeFacts(provider, adapted);
+  return adapted;
 }
 
 export async function createLlamaCppMemoryEmbeddingProvider(
@@ -198,6 +211,9 @@ export async function createLlamaCppMemoryEmbeddingProvider(
   );
   const identifiedProvider =
     identity.model === provider.model ? provider : { ...provider, model: identity.model };
+  if (identifiedProvider !== provider) {
+    copyLocalRuntimeFacts(provider, identifiedProvider);
+  }
   return {
     provider: identifiedProvider,
     runtime: createLlamaCppEmbeddingProviderRuntime(identity),