fix(ai): adjusted llama-cpp base url and refactor token validation

- Update `llama.cpp` base URL to remove the `/v1` suffix.
- Consolidate local provider token validation in `coding-agent` using a `Set`.
- Add test coverage to ensure catalog model IDs are preserved verbatim on the wire.

packages/ai/src/registry/llama-cpp.ts

@@ -3,7 +3,7 @@ import type { ProviderDefinition } from "./types";
 
 const PROVIDER_ID = "llama.cpp";
 const AUTH_URL = "https://github.com/ggml-org/llama.cpp#quick-start";
-const DEFAULT_LOCAL_BASE_URL = "http://127.0.0.1:8080/v1";
+const DEFAULT_LOCAL_BASE_URL = "http://127.0.0.1:8080";
 const DEFAULT_LOCAL_TOKEN = "llama-cpp-local";
 
 export async function loginLlamaCpp(options: OAuthController): Promise<string> {

packages/catalog/test/wafer.test.ts

@@ -7,10 +7,19 @@
  * catalog contract and the case-sensitive id pass-through against the wire.
  */
 import { describe, expect, it } from "bun:test";
+import { streamOpenAICompletions } from "@oh-my-pi/pi-ai/providers/openai-completions";
+import type { Context, FetchImpl, Model } from "@oh-my-pi/pi-ai/types";
 import { createModelManager } from "@oh-my-pi/pi-catalog/model-manager";
 import { getBundledModel } from "@oh-my-pi/pi-catalog/models";
 import { waferServerlessModelManagerOptions } from "@oh-my-pi/pi-catalog/provider-models/openai-compat";
-import type { FetchImpl, Model } from "@oh-my-pi/pi-catalog/types";
+
+function sseResponse(events: unknown[]): Response {
+	const payload = `${events.map(e => `data: ${typeof e === "string" ? e : JSON.stringify(e)}`).join("\n\n")}\n\n`;
+	return new Response(payload, {
+		status: 200,
+		headers: { "content-type": "text/event-stream" },
+	});
+}
 
 describe("Wafer Serverless provider", () => {
 	it("ships the documented Serverless catalog (GLM-5.1, Qwen3.5, Qwen3.6, Qwen3.7-Max, Kimi-K2.6, DeepSeek V4 Flash/Pro)", () => {
@@ -70,6 +79,37 @@ describe("Wafer Serverless provider", () => {
 		expect(dsPro.reasoning).toBe(true);
 		expect(dsPro.compatConfig?.thinkingFormat).toBeUndefined();
 	});
+
+	it("preserves the catalog id verbatim on the wire (no rewrite, case-sensitive)", async () => {
+		const model = getBundledModel<"openai-completions">("wafer-serverless", "GLM-5.1");
+		const captured: { url: string | null; body: string | null } = { url: null, body: null };
+		const fetchMock: FetchImpl = async (input: string | URL | Request, init?: RequestInit) => {
+			captured.url = typeof input === "string" ? input : input instanceof URL ? input.toString() : input.url;
+			captured.body = typeof init?.body === "string" ? init.body : null;
+			return sseResponse(["[DONE]"]);
+		};
+
+		const context: Context = {
+			systemPrompt: ["t"],
+			messages: [{ role: "user", content: "hi", timestamp: Date.now() }],
+		};
+		const stream = streamOpenAICompletions(model as Model<"openai-completions">, context, {
+			apiKey: "wfr_test",
+			fetch: fetchMock,
+		});
+		for await (const _event of stream) {
+			/* drain */
+		}
+
+		expect(captured.url).toBe("https://pass.wafer.ai/v1/chat/completions");
+		expect(captured.body).not.toBeNull();
+		const parsed = JSON.parse(captured.body ?? "{}") as { model?: unknown };
+		// Wafer's docs note model names are case-insensitive on input, but the
+		// canonical id has mixed case; we must round-trip it unchanged so users
+		// who pin `GLM-5.1` don't end up with usage rows under `glm-5.1` or
+		// hitting the upstream 404 path.
+		expect(parsed.model).toBe("GLM-5.1");
+	});
 });
 
 describe("Wafer dynamic discovery mapper", () => {

packages/coding-agent/src/config/model-registry.ts

@@ -24,13 +24,6 @@ import {
 	resolveVariantAlias,
 } from "@oh-my-pi/pi-catalog/variant-collapse";
 
-// Sentinels for local-only OAuth tokens — declared inline to avoid loading
-// provider modules at startup. Must match packages/ai/src/registry/llama-cpp.ts,
-// packages/ai/src/registry/lm-studio.ts, and packages/ai/src/registry/vllm.ts.
-const DEFAULT_LLAMA_CPP_LOCAL_TOKEN = "llama-cpp-local";
-const DEFAULT_LOCAL_TOKEN = "lm-studio-local";
-const DEFAULT_VLLM_LOCAL_TOKEN = "vllm-local";
-
 const SPECIAL_MODEL_MANAGER_PROVIDER_IDS: readonly string[] = [
 	"google-antigravity",
 	"google-gemini-cli",
@@ -42,6 +35,11 @@ const STARTUP_MODEL_CACHE_PROVIDER_IDS: readonly string[] = [
 	...SPECIAL_MODEL_MANAGER_PROVIDER_IDS,
 ];
 
+// Sentinels for local-only OAuth tokens — declared inline to avoid loading
+// provider modules at startup. Must match packages/ai/src/registry/llama-cpp.ts,
+// packages/ai/src/registry/lm-studio.ts, and packages/ai/src/registry/vllm.ts.
+const LOCAL_PROVIDER_PLACEHOLDERS = new Set<string>(["llama-cpp-local", "lm-studio-local", "vllm-local"]);
+
 import type { ApiKeyResolver, FetchImpl } from "@oh-my-pi/pi-ai";
 import { registerOAuthProvider, unregisterOAuthProviders } from "@oh-my-pi/pi-ai/oauth";
 import type { OAuthCredentials, OAuthLoginCallbacks } from "@oh-my-pi/pi-ai/oauth/types";
@@ -86,12 +84,7 @@ export function isAuthenticated(apiKey: string | undefined | null): apiKey is st
 }
 
 function isDiscoveryBearerApiKey(apiKey: string | undefined | null): apiKey is string {
-	return (
-		isAuthenticated(apiKey) &&
-		apiKey !== DEFAULT_LLAMA_CPP_LOCAL_TOKEN &&
-		apiKey !== DEFAULT_LOCAL_TOKEN &&
-		apiKey !== DEFAULT_VLLM_LOCAL_TOKEN
-	);
+	return isAuthenticated(apiKey) && !LOCAL_PROVIDER_PLACEHOLDERS.has(apiKey);
 }
 
 /** Provider override config (baseUrl, headers, apiKey, compat, transport) without custom models */