diff --git a/bin/lib/onboard.js b/bin/lib/onboard.js
index 23f19b01a..709c3dc51 100644
--- a/bin/lib/onboard.js
+++ b/bin/lib/onboard.js
@@ -221,10 +221,10 @@ function getNonInteractiveProvider() {
   const providerKey = (process.env.NEMOCLAW_PROVIDER || "").trim().toLowerCase();
   if (!providerKey) return null;
 
-  const validProviders = new Set(["cloud", "ollama", "vllm", "nim"]);
+  const validProviders = new Set(["cloud", "ollama", "vllm", "nim", "dynamo"]);
   if (!validProviders.has(providerKey)) {
     console.error(`  Unsupported NEMOCLAW_PROVIDER: ${providerKey}`);
-    console.error("  Valid values: cloud, ollama, vllm, nim");
+    console.error("  Valid values: cloud, ollama, vllm, nim, dynamo");
     process.exit(1);
   }
 
@@ -473,6 +473,22 @@ async function setupNim(sandboxName, gpu) {
   const vllmRunning = !!runCapture("curl -sf http://localhost:8000/v1/models 2>/dev/null", { ignoreError: true });
   const requestedProvider = isNonInteractive() ? getNonInteractiveProvider() : null;
   const requestedModel = isNonInteractive() ? getNonInteractiveModel(requestedProvider || "cloud") : null;
+
+  // Non-interactive Dynamo provider: handle before the interactive options flow
+  if (isNonInteractive() && requestedProvider === "dynamo") {
+    const dynamoEndpoint = (process.env.NEMOCLAW_DYNAMO_ENDPOINT || "").trim();
+    if (!dynamoEndpoint) {
+      console.error("  NEMOCLAW_DYNAMO_ENDPOINT is required when NEMOCLAW_PROVIDER=dynamo.");
+      process.exit(1);
+    }
+    console.log(`  [non-interactive] Using Dynamo provider`);
+
+    provider = "dynamo";
+    model = (process.env.NEMOCLAW_DYNAMO_MODEL || "").trim() || "dynamo";
+    registry.updateSandbox(sandboxName, { model, provider, nimContainer });
+    return { model, provider };
+  }
+
   // Build options list — only show local options with NEMOCLAW_EXPERIMENTAL=1
   const options = [];
   if (EXPERIMENTAL && gpu && gpu.nimCapable) {
@@ -645,7 +661,23 @@ async function setupNim(sandboxName, gpu) {
 async function setupInference(sandboxName, model, provider) {
   step(5, 7, "Setting up inference provider");
 
-  if (provider === "nvidia-nim") {
+  if (provider === "dynamo") {
+    // Dynamo: external vLLM endpoint (e.g., K8s service)
+    const dynamoEndpoint = process.env.NEMOCLAW_DYNAMO_ENDPOINT;
+    // Use shellQuote for shell-safe escaping of the endpoint URL
+    run(
+      `openshell provider create --name dynamo --type openai ` +
+      `--credential OPENAI_API_KEY=dummy ` +
+      `--config OPENAI_BASE_URL=${shellQuote(dynamoEndpoint)} 2>&1 || ` +
+      `openshell provider update dynamo --credential OPENAI_API_KEY=dummy ` +
+      `--config OPENAI_BASE_URL=${shellQuote(dynamoEndpoint)} 2>&1 || true`,
+      { ignoreError: true }
+    );
+    run(
+      `openshell inference set --no-verify --provider dynamo --model ${shellQuote(model)} 2>/dev/null || true`,
+      { ignoreError: true }
+    );
+  } else if (provider === "nvidia-nim") {
     // Create nvidia-nim provider
     run(
       `openshell provider create --name nvidia-nim --type openai ` +
@@ -850,6 +882,7 @@ function printDashboard(sandboxName, model, provider) {
   if (provider === "nvidia-nim") providerLabel = "NVIDIA Cloud API";
   else if (provider === "vllm-local") providerLabel = "Local vLLM";
   else if (provider === "ollama-local") providerLabel = "Local Ollama";
+  else if (provider === "dynamo") providerLabel = "Dynamo vLLM";
 
   console.log("");
   console.log(`  ${"─".repeat(50)}`);