fix(llama-cpp): clarify local model setup

extensions/llama-cpp/index.test.ts

@@ -75,7 +75,7 @@ describe("llama.cpp provider plugin", () => {
     expect(registerProvider).toHaveBeenCalledWith(
       expect.objectContaining({
         id: "llama-cpp",
-        label: "Local model (llama.cpp)",
+        label: "llama.cpp",
         createStreamFn: expect.any(Function),
         normalizeToolSchemas: expect.any(Function),
         inspectToolSchemas: expect.any(Function),

extensions/llama-cpp/index.ts

@@ -24,7 +24,7 @@ export default definePluginEntry({
         {
           id: "local",
           label: LLAMA_CPP_PROVIDER_LABEL,
-          hint: "In-process local GGUF model (about 5.0 GB download; requires 16 GB RAM)",
+          hint: "Run one private GGUF model directly inside this Gateway",
           kind: "custom",
           appGuidedSetup: {
             detect: detectLlamaCppSetup,
@@ -59,15 +59,15 @@ export default definePluginEntry({
         setup: {
           choiceId: LLAMA_CPP_PROVIDER_ID,
           choiceLabel: LLAMA_CPP_PROVIDER_LABEL,
-          choiceHint: "In-process local model (about 5.0 GB download; requires 16 GB RAM)",
+          choiceHint: "Run one private GGUF model directly inside this Gateway",
           groupId: LLAMA_CPP_PROVIDER_ID,
           groupLabel: "Local llama.cpp",
           groupHint: "No API key required",
           methodId: "local",
         },
         modelPicker: {
-          label: "llama.cpp (local GGUF)",
-          hint: "Run a GGUF model in the OpenClaw process",
+          label: "llama.cpp",
+          hint: "Run a GGUF model directly inside OpenClaw",
           methodId: "local",
         },
       },

extensions/llama-cpp/openclaw.plugin.json

@@ -29,9 +29,9 @@
       "method": "local",
       "choiceId": "llama-cpp",
       "appGuidedDiscovery": true,
-      "appGuidedActionLabel": "Review download",
-      "choiceLabel": "Local model (llama.cpp)",
-      "choiceHint": "Downloads an approximately 5.0 GB local model; requires 16 GB RAM",
+      "appGuidedActionLabel": "Set up model",
+      "choiceLabel": "llama.cpp",
+      "choiceHint": "Run one private GGUF model directly inside this Gateway",
       "groupId": "llama-cpp",
       "groupLabel": "Local llama.cpp",
       "groupHint": "No API key required"

extensions/llama-cpp/src/defaults.ts

@@ -6,7 +6,7 @@ import type {
 } from "openclaw/plugin-sdk/provider-model-shared";
 
 export const LLAMA_CPP_PROVIDER_ID = "llama-cpp";
-export const LLAMA_CPP_PROVIDER_LABEL = "Local model (llama.cpp)";
+export const LLAMA_CPP_PROVIDER_LABEL = "llama.cpp";
 const LLAMA_CPP_LOCAL_AUTH_MARKER = "llama-cpp-local";
 const LLAMA_CPP_LOCAL_BASE_URL = "local://llama-cpp";
 

extensions/llama-cpp/src/setup.test.ts

@@ -125,7 +125,7 @@ describe("llama.cpp setup", () => {
 
     await expect(detectLlamaCppSetup({ config: configWithCache(), env: {} })).resolves.toEqual({
       modelRef: DEFAULT_LLAMA_CPP_MODEL_REF,
-      detail: "gemma-4-e4b-it-q4_k_m (downloaded)",
+      detail: "Ready locally",
     });
     expect(nodeLlamaMocks.createModelDownloader).not.toHaveBeenCalled();
     expect(nodeLlamaMocks.resolveModelFile).toHaveBeenCalledWith(
@@ -157,7 +157,7 @@ describe("llama.cpp setup", () => {
 
     await expect(detectLlamaCppSetup({ config, env: {} })).resolves.toEqual({
       modelRef: "llama-cpp/custom",
-      detail: "custom (downloaded)",
+      detail: "Ready locally",
     });
     expect(nodeLlamaMocks.resolveModelFile).toHaveBeenCalledWith(
       "hf:org/repo/model.gguf#release",
@@ -215,7 +215,7 @@ describe("llama.cpp setup", () => {
 
     expect(ctx.prompter.confirm).not.toHaveBeenCalled();
     expect(ctx.prompter.note).toHaveBeenCalledWith(
-      "This machine has 8 GB RAM; the bundled local model needs 16 GB+. Use Ollama/LM Studio with a smaller model, or a cloud provider.",
+      "This Gateway has 8 GB RAM; the recommended model needs 16 GB+. Use Ollama or LM Studio with a smaller model, configure an existing GGUF, or choose a cloud provider.",
       "Setup skipped",
     );
     expect(nodeLlamaMocks.createModelDownloader).not.toHaveBeenCalled();
@@ -240,7 +240,9 @@ describe("llama.cpp setup", () => {
     await expect(runLlamaCppSetup(ctx)).resolves.toEqual({ profiles: [] });
 
     expect(ctx.prompter.confirm).toHaveBeenCalledWith(
-      expect.objectContaining({ message: expect.stringContaining("about 5.0 GB") }),
+      expect.objectContaining({
+        message: expect.stringContaining("run it directly inside this Gateway"),
+      }),
     );
     expect(nodeLlamaMocks.createModelDownloader).not.toHaveBeenCalled();
   });

extensions/llama-cpp/src/setup.ts

@@ -98,7 +98,7 @@ export async function detectLlamaCppSetup(ctx: ProviderAppGuidedSetupContext) {
       }
       return {
         modelRef: `${LLAMA_CPP_PROVIDER_ID}/${candidate.model.id}`,
-        detail: `${candidate.model.id} (downloaded)`,
+        detail: "Ready locally",
       };
     } catch {
       // Discovery is read-only: a missing model or native module is not a setup error.
@@ -148,13 +148,14 @@ export async function runLlamaCppSetup(ctx: ProviderAuthContext): Promise<Provid
     const totalmemBytes = os.totalmem();
     if (!meetsLlamaCppDefaultModelRamFloor(totalmemBytes)) {
       await ctx.prompter.note(
-        `This machine has ${formatRamGb(totalmemBytes)} GB RAM; the bundled local model needs 16 GB+. Use Ollama/LM Studio with a smaller model, or a cloud provider.`,
+        `This Gateway has ${formatRamGb(totalmemBytes)} GB RAM; the recommended model needs 16 GB+. Use Ollama or LM Studio with a smaller model, configure an existing GGUF, or choose a cloud provider.`,
         "Setup skipped",
       );
       return { profiles: [] };
     }
     const consent = await ctx.prompter.confirm({
-      message: "Download Gemma 4 E4B IT Q4_K_M (about 5.0 GB) for local llama.cpp inference?",
+      message:
+        "OpenClaw will download Gemma 4 E4B IT Q4_K_M (about 5.0 GB) and run it directly inside this Gateway. Continue?",
       initialValue: false,
     });
     if (!consent) {