fix(onboard): remove experimental vLLM label on managed platforms

docs/inference/local-compatible-inference-setup.mdx

@@ -220,7 +220,7 @@ On supported Linux hosts with NVIDIA GPUs, the onboard wizard can also install o
 
 ### Use an Existing vLLM Server
 
-For an already-running vLLM server, run `$$nemoclaw onboard` and select **Local vLLM [experimental]** from the provider list.
+For an already-running vLLM server, run `$$nemoclaw onboard` and select **Local vLLM** from the provider list. The entry retains an experimental label on generic hosts, but not on DGX Spark or DGX Station.
 If vLLM is already running, NemoClaw detects the running model and validates the endpoint.
 When vLLM exposes runtime metadata such as `max_model_len`, NemoClaw uses that value for the configured context window unless you set `NEMOCLAW_CONTEXT_WINDOW` yourself.
 

docs/inference/tool-calling-reliability.mdx

@@ -104,7 +104,7 @@ These routes therefore keep direct structured tool calling until search, describ
 
 ## Recommended Fix
 
-For persistent NemoClaw use, start vLLM with auto tool choice and the parser that matches your model family, then rerun onboarding and select **Local vLLM [experimental]** or **Other OpenAI-compatible endpoint**.
+For persistent NemoClaw use, start vLLM with auto tool choice and the parser that matches your model family, then rerun onboarding and select **Local vLLM** or **Other OpenAI-compatible endpoint**. The Local vLLM entry retains an experimental label on generic hosts, but not on DGX Spark or DGX Station.
 
 ### Configure vLLM
 

src/lib/onboard/vllm-menu.test.ts

@@ -20,7 +20,7 @@ describe("buildVllmMenuEntries", () => {
     assert.deepEqual(entries, []);
   });
 
-  it("returns the running entry when vLLM is reachable on localhost", () => {
+  it("marks the running entry experimental on generic hosts", () => {
     const entries = buildVllmMenuEntries({
       vllmRunning: true,
       vllmProfile: null,
@@ -35,6 +35,27 @@ describe("buildVllmMenuEntries", () => {
     assert.match(entries[0].label, /running/);
   });
 
+  for (const [platform, hostLabel] of [
+    ["spark", "Spark"],
+    ["station", "Station"],
+  ] as const) {
+    it(`does not mark the running entry experimental on DGX ${hostLabel}`, () => {
+      const entries = buildVllmMenuEntries({
+        vllmRunning: true,
+        vllmProfile: null,
+        experimental: false,
+        platform,
+        hasVllmImage: false,
+        log: () => {},
+        env: {},
+      });
+      assert.equal(entries.length, 1);
+      assert.equal(entries[0].key, "vllm");
+      assert.doesNotMatch(entries[0].label, /experimental/);
+      assert.match(entries[0].label, /running/);
+    });
+  }
+
   it("returns the install entry when a profile matches and EXPERIMENTAL is set", () => {
     const entries = buildVllmMenuEntries({
       vllmRunning: false,

src/lib/onboard/vllm-menu.ts

@@ -62,10 +62,12 @@ export function buildVllmMenuEntries(opts: BuildVllmMenuOptions): VllmMenuEntry[
         `  Note: NEMOCLAW_PROVIDER=install-vllm requested, but vLLM is already running on localhost:${VLLM_PORT} — selecting the running instance.`,
       );
     }
+    const experimentalLabel =
+      opts.platform && MANAGED_VLLM_DEFAULT_PLATFORMS.has(opts.platform) ? "" : " [experimental]";
     return [
       {
         key: "vllm",
-        label: `Local vLLM [experimental] (localhost:${VLLM_PORT}) — running (suggested)`,
+        label: `Local vLLM${experimentalLabel} (localhost:${VLLM_PORT}) — running (suggested)`,
       },
     ];
   }