fix: restore routed inference and connect UX

bin/lib/onboard.js

@@ -1955,7 +1955,6 @@ async function setupNim(gpu) {
       if (ollamaRunning) suggestions.push("Ollama");
       if (suggestions.length > 0) {
         console.log(`  Detected local inference option${suggestions.length > 1 ? "s" : ""}: ${suggestions.join(", ")}`);
-        console.log("  Select one explicitly to use it. Press Enter to keep NVIDIA Endpoints.");
         console.log("");
       }
 

bin/nemoclaw.js

@@ -562,8 +562,6 @@ function listSandboxes() {
 
 async function sandboxConnect(sandboxName) {
   await ensureLiveSandboxOrExit(sandboxName);
-  // Ensure port forward is alive before connecting
-  runOpenshell(["forward", "start", "--background", "18789", sandboxName], { ignoreError: true });
   const result = spawnSync(getOpenshellBinary(), ["sandbox", "connect", sandboxName], {
     stdio: "inherit",
     cwd: ROOT,

scripts/nemoclaw-start.sh

@@ -207,18 +207,20 @@ PYAUTOPAIR
 
 # ── Proxy environment ────────────────────────────────────────────
 # OpenShell injects HTTP_PROXY/HTTPS_PROXY/NO_PROXY into the sandbox, but its
-# NO_PROXY is limited to 127.0.0.1,localhost,::1 — missing inference.local and
-# the gateway IP.  Without these entries, LLM inference requests are routed
-# through the egress proxy instead of going direct, and the proxy gateway IP
-# itself gets proxied (potential infinite loop).
+# NO_PROXY is limited to 127.0.0.1,localhost,::1 — missing the gateway IP.
+# The gateway IP itself must bypass the proxy to avoid proxy loops.
+#
+# Do NOT add inference.local here. OpenShell intentionally routes that hostname
+# through the proxy path; bypassing the proxy forces a direct DNS lookup inside
+# the sandbox, which breaks inference.local resolution.
 #
 # NEMOCLAW_PROXY_HOST / NEMOCLAW_PROXY_PORT can be overridden at sandbox
 # creation time if the gateway IP or port changes in a future OpenShell release.
 # Ref: https://github.com/NVIDIA/NemoClaw/issues/626
 PROXY_HOST="${NEMOCLAW_PROXY_HOST:-10.200.0.1}"
 PROXY_PORT="${NEMOCLAW_PROXY_PORT:-3128}"
 _PROXY_URL="http://${PROXY_HOST}:${PROXY_PORT}"
-_NO_PROXY_VAL="localhost,127.0.0.1,::1,inference.local,${PROXY_HOST}"
+_NO_PROXY_VAL="localhost,127.0.0.1,::1,${PROXY_HOST}"
 export HTTP_PROXY="$_PROXY_URL"
 export HTTPS_PROXY="$_PROXY_URL"
 export NO_PROXY="$_NO_PROXY_VAL"

test/cli.test.js

@@ -147,6 +147,64 @@ describe("CLI dispatch", () => {
     expect(fs.readFileSync(markerFile, "utf8")).toContain("logs alpha --follow");
   });
 
+  it("connect does not pre-start a duplicate port forward", () => {
+    const home = fs.mkdtempSync(path.join(os.tmpdir(), "nemoclaw-cli-connect-forward-"));
+    const localBin = path.join(home, "bin");
+    const registryDir = path.join(home, ".nemoclaw");
+    const markerFile = path.join(home, "openshell-calls");
+    fs.mkdirSync(localBin, { recursive: true });
+    fs.mkdirSync(registryDir, { recursive: true });
+    fs.writeFileSync(
+      path.join(registryDir, "sandboxes.json"),
+      JSON.stringify({
+        sandboxes: {
+          alpha: {
+            name: "alpha",
+            model: "test-model",
+            provider: "nvidia-prod",
+            gpuEnabled: false,
+            policies: [],
+          },
+        },
+        defaultSandbox: "alpha",
+      }),
+      { mode: 0o600 }
+    );
+    fs.writeFileSync(
+      path.join(localBin, "openshell"),
+      [
+        "#!/usr/bin/env bash",
+        `marker_file=${JSON.stringify(markerFile)}`,
+        "printf '%s\\n' \"$*\" >> \"$marker_file\"",
+        "if [ \"$1\" = \"sandbox\" ] && [ \"$2\" = \"get\" ] && [ \"$3\" = \"alpha\" ]; then",
+        "  echo 'Sandbox:'",
+        "  echo",
+        "  echo '  Id: abc'",
+        "  echo '  Name: alpha'",
+        "  echo '  Namespace: openshell'",
+        "  echo '  Phase: Ready'",
+        "  exit 0",
+        "fi",
+        "if [ \"$1\" = \"sandbox\" ] && [ \"$2\" = \"connect\" ] && [ \"$3\" = \"alpha\" ]; then",
+        "  exit 0",
+        "fi",
+        "exit 0",
+        ].join("\n"),
+      { mode: 0o755 }
+    );
+
+    const r = runWithEnv("alpha connect", {
+      HOME: home,
+      PATH: `${localBin}:${process.env.PATH || ""}`,
+    });
+
+    expect(r.code).toBe(0);
+    const calls = fs.readFileSync(markerFile, "utf8").trim().split("\n").filter(Boolean);
+    expect(calls).toContain("sandbox get alpha");
+    expect(calls).toContain("sandbox connect alpha");
+    expect(calls.some((call) => call.startsWith("forward start --background 18789"))).toBe(false);
+  });
+
   it("removes stale registry entries when connect targets a missing live sandbox", () => {
     const home = fs.mkdtempSync(path.join(os.tmpdir(), "nemoclaw-cli-stale-connect-"));
     const localBin = path.join(home, "bin");

test/onboard-selection.test.js

@@ -100,7 +100,6 @@ const { setupNim } = require(${onboardPath});
     assert.match(payload.messages[0], /Choose \[/);
     assert.match(payload.messages[1], /Choose model \[1\]/);
     assert.ok(payload.lines.some((line) => line.includes("Detected local inference option")));
-    assert.ok(payload.lines.some((line) => line.includes("Press Enter to keep NVIDIA Endpoints")));
     assert.ok(payload.lines.some((line) => line.includes("Cloud models:")));
     assert.ok(payload.lines.some((line) => line.includes("Responses API available")));
   });

test/service-env.test.js

@@ -161,13 +161,13 @@ describe("service environment", () => {
       expect(vars.HTTPS_PROXY).toBe("http://10.200.0.1:8080");
     });
 
-    it("NO_PROXY includes loopback and inference.local", () => {
+    it("NO_PROXY includes loopback only, not inference.local", () => {
       const vars = extractProxyVars();
       const noProxy = vars.NO_PROXY.split(",");
       expect(noProxy).toContain("localhost");
       expect(noProxy).toContain("127.0.0.1");
       expect(noProxy).toContain("::1");
-      expect(noProxy).toContain("inference.local");
+      expect(noProxy).not.toContain("inference.local");
     });
 
     it("NO_PROXY includes OpenShell gateway IP", () => {
@@ -180,7 +180,7 @@ describe("service environment", () => {
       expect(vars.http_proxy).toBe("http://10.200.0.1:3128");
       expect(vars.https_proxy).toBe("http://10.200.0.1:3128");
       const noProxy = vars.no_proxy.split(",");
-      expect(noProxy).toContain("inference.local");
+      expect(noProxy).not.toContain("inference.local");
       expect(noProxy).toContain("10.200.0.1");
     });
 
@@ -211,11 +211,11 @@ describe("service environment", () => {
         expect(bashrc).toContain("export HTTP_PROXY=");
         expect(bashrc).toContain("export HTTPS_PROXY=");
         expect(bashrc).toContain("export NO_PROXY=");
-        expect(bashrc).toContain("inference.local");
+        expect(bashrc).not.toContain("inference.local");
         expect(bashrc).toContain("10.200.0.1");
 
         const profile = readFileSync(join(fakeHome, ".profile"), "utf-8");
-        expect(profile).toContain("inference.local");
+        expect(profile).not.toContain("inference.local");
       } finally {
         try { unlinkSync(tmpFile); } catch { /* ignore */ }
         try { execFileSync("rm", ["-rf", fakeHome]); } catch { /* ignore */ }
@@ -306,10 +306,10 @@ describe("service environment", () => {
           "# nemoclaw-proxy-config begin",
           'export HTTP_PROXY="http://10.200.0.1:3128"',
           'export HTTPS_PROXY="http://10.200.0.1:3128"',
-          'export NO_PROXY="localhost,127.0.0.1,::1,inference.local,10.200.0.1"',
+          'export NO_PROXY="localhost,127.0.0.1,::1,10.200.0.1"',
           'export http_proxy="http://10.200.0.1:3128"',
           'export https_proxy="http://10.200.0.1:3128"',
-          'export no_proxy="localhost,127.0.0.1,::1,inference.local,10.200.0.1"',
+          'export no_proxy="localhost,127.0.0.1,::1,10.200.0.1"',
           "# nemoclaw-proxy-config end",
         ].join("\n");
         writeFileSync(join(fakeHome, ".bashrc"), bashrcContent);
@@ -323,8 +323,8 @@ describe("service environment", () => {
           'echo "no_proxy=$no_proxy"',
         ].join("; ")], { encoding: "utf-8" }).trim();
 
-        expect(out).toContain("NO_PROXY=localhost,127.0.0.1,::1,inference.local,10.200.0.1");
-        expect(out).toContain("no_proxy=localhost,127.0.0.1,::1,inference.local,10.200.0.1");
+        expect(out).toContain("NO_PROXY=localhost,127.0.0.1,::1,10.200.0.1");
+        expect(out).toContain("no_proxy=localhost,127.0.0.1,::1,10.200.0.1");
       } finally {
         try { execFileSync("rm", ["-rf", fakeHome]); } catch { /* ignore */ }
       }