feat: openshell improvements (#33)

* Add rebrand for welcome UI, include launch.sh script

* Remove BASH_SOURCE dependency

* Address silent fail on launch.sh

* Add favicon, handle ghcr.io login if params present, fix logo

* Init LiteLLM implementation

* LiteLLM working

* Update welcome UI icon assets

* Add on-demand nemoclaw build; improve auto-pair

* Logo fixup, improve auto-approve cycle, NO_PROXY for localhost

* Bump defualt context window, set NO_PROXY widely

* Extend timer for device auto approval, minimize wait

* Reload dashboard once after pairing approval

* Revert nemoclaw runtime back to inference.local

* Keep pairing watcher alive until approval

* Add proxy request tracing for sandbox launch

* Add override to skip nemoclaw image build

* Add revised policy and NO_PROXY

* Fix unconditional chown

* Added guarded reload for pairing; ensure custom policy.yaml bake-in

* Add console logging for device pairing; extend NO_PROXY

* Handle context mod for inference.local

* Fix k3s image import on build; force reload on first pass timeout

* Revise Brev README

* Cleanup Brev section

* Revert policy.yaml to orig

* Added policy recommender

* nemoclaw: inference UX overhaul, inline API keys, partner logos, denial watcher, preview tooling

Made-with: Cursor

---------

Co-authored-by: JR Morgan <liveaverage@gmail.com>
Co-authored-by: OpenShell-Community Dev <dev@openshell-community.local>

Author: 3 people

.gitignore

@@ -1,2 +1,2 @@
 node_modules/
-AGENTS.md
+AGENTS.md

brev/launch.sh

@@ -568,12 +568,12 @@
   set_inference_route() {
     log "Configuring inference route..."
 
-    if "$CLI_BIN" inference set --provider nvidia-endpoints --model qwen/qwen3.5-397b-a17b >/dev/null 2>&1; then
+    if "$CLI_BIN" inference set --provider nvidia-endpoints --model minimaxai/minimax-m2.5 >/dev/null 2>&1; then
       log "Configured inference via '$CLI_BIN inference set'."
       return
     fi
 
-    if "$CLI_BIN" cluster inference set --provider nvidia-endpoints --model qwen/qwen3.5-397b-a17b >/dev/null 2>&1; then
+    if "$CLI_BIN" cluster inference set --provider nvidia-endpoints --model minimaxai/minimax-m2.5 >/dev/null 2>&1; then
       log "Configured inference via legacy '$CLI_BIN cluster inference set'."
       return
     fi
@@ -706,12 +706,6 @@
     import_nemoclaw_image_into_cluster_if_needed
 
     step "Configuring providers"
-    run_provider_create_or_replace \
-      nvidia-inference \
-      --type openai \
-      --credential OPENAI_API_KEY=unused \
-      --config OPENAI_BASE_URL=https://inference-api.nvidia.com/v1
-
     run_provider_create_or_replace \
       nvidia-endpoints \
       --type nvidia \

brev/welcome-ui/SERVER_ARCHITECTURE.md

@@ -148,7 +148,6 @@ main()
   ├── 1. _bootstrap_config_cache()
   │       If /tmp/nemoclaw-provider-config-cache.json does NOT exist:
   │         Write defaults for:
-  │           - nvidia-inference → OPENAI_BASE_URL=https://inference-api.nvidia.com/v1
   │           - nvidia-endpoints → NVIDIA_BASE_URL=https://integrate.api.nvidia.com/v1
   │       If it already exists: skip (no-op)
   │
@@ -436,13 +435,12 @@ Step 10: Cleanup temp policy file
 ```
 Step 1: Log receipt (hash prefix)
 Step 2: Run CLI command:
-          nemoclaw provider update nvidia-inference \
-            --type openai \
-            --credential OPENAI_API_KEY=<key> \
-            --config OPENAI_BASE_URL=https://inference-api.nvidia.com/v1
+          nemoclaw provider update nvidia-endpoints \
+            --credential NVIDIA_API_KEY=<key> \
+            --config NVIDIA_BASE_URL=https://integrate.api.nvidia.com/v1
         Timeout: 120s
 Step 3: If success:
-          - Cache config {"OPENAI_BASE_URL": "https://inference-api.nvidia.com/v1"} under name "nvidia-inference"
+          - Cache config under name "nvidia-endpoints"
           - State → "done"
         If failure:
           - State → "error" with stderr/stdout message
@@ -537,10 +535,10 @@ Step 3: Merge with config cache values
 The CLI outputs text like:
 ```
 Id:              abc-123
-Name:            nvidia-inference
-Type:            openai
-Credential keys: OPENAI_API_KEY
-Config keys:     OPENAI_BASE_URL
+Name:            nvidia-endpoints
+Type:            nvidia
+Credential keys: NVIDIA_API_KEY
+Config keys:     NVIDIA_BASE_URL
 ```
 
 Parsing rules:
@@ -560,11 +558,11 @@ After parsing, if the provider name has an entry in the config cache, a `configV
     "providers": [
         {
             "id": "abc-123",
-            "name": "nvidia-inference",
-            "type": "openai",
-            "credentialKeys": ["OPENAI_API_KEY"],
-            "configKeys": ["OPENAI_BASE_URL"],
-            "configValues": {"OPENAI_BASE_URL": "https://inference-api.nvidia.com/v1"}
+            "name": "nvidia-endpoints",
+            "type": "nvidia",
+            "credentialKeys": ["NVIDIA_API_KEY"],
+            "configKeys": ["NVIDIA_BASE_URL"],
+            "configValues": {"NVIDIA_BASE_URL": "https://integrate.api.nvidia.com/v1"}
         }
     ]
 }
@@ -671,7 +669,7 @@ nemoclaw cluster inference get
 
 **Output Parsing (`_parse_cluster_inference`):**
 ```
-Provider:  nvidia-inference
+Provider:  nvidia-endpoints
 Model:     meta/llama-3.1-70b-instruct
 Version:   2
 ```
@@ -688,7 +686,7 @@ Version:   2
 ```json
 {
     "ok": true,
-    "providerName": "nvidia-inference",
+    "providerName": "nvidia-endpoints",
     "modelId": "meta/llama-3.1-70b-instruct",
     "version": 2
 }
@@ -703,7 +701,7 @@ Version:   2
 **Request Body:**
 ```json
 {
-    "providerName": "nvidia-inference",
+    "providerName": "nvidia-endpoints",
     "modelId": "meta/llama-3.1-70b-instruct"
 }
 ```
@@ -721,7 +719,7 @@ nemoclaw cluster inference set --provider <name> --model <model>
 ```json
 {
     "ok": true,
-    "providerName": "nvidia-inference",
+    "providerName": "nvidia-endpoints",
     "modelId": "meta/llama-3.1-70b-instruct",
     "version": 3
 }
@@ -925,7 +923,7 @@ The `nemoclaw provider get` CLI only returns config **key names**, not their val
 - Read on every `GET /api/providers` request
 - Written on every `POST` (create) and `PUT` (update) that includes config values
 - Cleaned up on `DELETE`
-- Bootstrapped at server startup with a default for `nvidia-inference`
+- Bootstrapped at server startup with a default for `nvidia-endpoints`
 
 ---
 
@@ -948,8 +946,8 @@ Output is parsed the same way as provider detail (line-by-line, prefix matching,
 **Format:**
 ```json
 {
-    "nvidia-inference": {
-        "OPENAI_BASE_URL": "https://inference-api.nvidia.com/v1"
+    "nvidia-endpoints": {
+        "NVIDIA_BASE_URL": "https://integrate.api.nvidia.com/v1"
     },
     "my-custom-provider": {
         "CUSTOM_URL": "https://example.com"
@@ -1237,7 +1235,7 @@ This means it uses a local sandbox name rather than a container image reference.
 
 ### 18.10 Inject Key Hardcodes Provider Name
 
-The `_run_inject_key` function hardcodes `nvidia-inference` as the provider name. This is not configurable via the API.
+The `_run_inject_key` function hardcodes `nvidia-endpoints` as the provider name. This is not configurable via the API.
 
 ### 18.11 Error State Truncation
 

brev/welcome-ui/__tests__/cli-parsing.test.js

@@ -33,7 +33,7 @@ describe("parseProviderDetail", () => {
     const result = parseProviderDetail(FIXTURES.providerGetOutput);
     expect(result).toEqual({
       id: "abc-123",
-      name: "nvidia-inference",
+      name: "nvidia-endpoints",
       type: "openai",
       credentialKeys: ["OPENAI_API_KEY"],
       configKeys: ["OPENAI_BASE_URL"],
@@ -63,7 +63,7 @@ describe("parseProviderDetail", () => {
   it("TC-CL09: ANSI codes in output are stripped before parsing", () => {
     const result = parseProviderDetail(FIXTURES.providerGetAnsi);
     expect(result).not.toBeNull();
-    expect(result.name).toBe("nvidia-inference");
+    expect(result.name).toBe("nvidia-endpoints");
     expect(result.type).toBe("openai");
   });
 });
@@ -72,7 +72,7 @@ describe("parseClusterInference", () => {
   it("TC-CL10: parses Provider, Model, Version lines", () => {
     const result = parseClusterInference(FIXTURES.clusterInferenceOutput);
     expect(result).toEqual({
-      providerName: "nvidia-inference",
+      providerName: "nvidia-endpoints",
       modelId: "meta/llama-3.1-70b-instruct",
       version: 2,
     });

brev/welcome-ui/__tests__/cluster-inference.test.js

@@ -40,7 +40,7 @@ describe("GET /api/cluster-inference", () => {
     const res = await request(server).get("/api/cluster-inference");
     expect(res.status).toBe(200);
     expect(res.body.ok).toBe(true);
-    expect(res.body.providerName).toBe("nvidia-inference");
+    expect(res.body.providerName).toBe("nvidia-endpoints");
     expect(res.body.modelId).toBe("meta/llama-3.1-70b-instruct");
     expect(res.body.version).toBe(2);
   });
@@ -95,7 +95,7 @@ describe("GET /api/cluster-inference", () => {
 
     const res = await request(server).get("/api/cluster-inference");
     expect(res.status).toBe(200);
-    expect(res.body.providerName).toBe("nvidia-inference");
+    expect(res.body.providerName).toBe("nvidia-endpoints");
     expect(res.body.modelId).toBe("meta/llama-3.1-70b-instruct");
   });
 });

brev/welcome-ui/__tests__/config-cache.test.js

@@ -19,7 +19,7 @@ describe("config cache", () => {
     bootstrapConfigCache();
     const cache = readCacheFile();
     expect(cache).not.toBeNull();
-    expect(cache["nvidia-inference"]).toBeDefined();
+    expect(cache["nvidia-endpoints"]).toBeDefined();
   });
 
   it("TC-CC02: bootstrapConfigCache is no-op when file already exists", () => {
@@ -29,13 +29,10 @@ describe("config cache", () => {
     expect(cache).toEqual({ custom: { x: 1 } });
   });
 
-  it("TC-CC03: default bootstrap content seeds both NVIDIA inference providers", () => {
+  it("TC-CC03: default bootstrap content seeds NVIDIA endpoints provider", () => {
     bootstrapConfigCache();
     const cache = readCacheFile();
     expect(cache).toEqual({
-      "nvidia-inference": {
-        OPENAI_BASE_URL: "https://inference-api.nvidia.com/v1",
-      },
       "nvidia-endpoints": {
         NVIDIA_BASE_URL: "https://integrate.api.nvidia.com/v1",
       },

brev/welcome-ui/__tests__/inject-key.test.js

@@ -145,7 +145,7 @@ describe("inject-key background process", () => {
     execFile.mockClear();
   });
 
-  it("TC-K10: updates both default inference providers with the submitted key", async () => {
+  it("TC-K10: updates default NVIDIA endpoints provider with the submitted key", async () => {
     execFile.mockImplementation((cmd, args, opts, cb) => {
       if (typeof opts === "function") { cb = opts; opts = {}; }
       cb(null, "", "");
@@ -161,15 +161,9 @@ describe("inject-key background process", () => {
     const updateCalls = execFile.mock.calls.filter(
       (c) => c[0] === "nemoclaw" && c[1]?.includes("update")
     );
-    expect(updateCalls.length).toBeGreaterThanOrEqual(2);
+    expect(updateCalls.length).toBeGreaterThanOrEqual(1);
 
-    const inferenceArgs = updateCalls.find((c) => c[1].includes("nvidia-inference"))?.[1] || [];
     const endpointsArgs = updateCalls.find((c) => c[1].includes("nvidia-endpoints"))?.[1] || [];
-
-    expect(inferenceArgs).toContain("nvidia-inference");
-    expect(inferenceArgs.some((a) => a.startsWith("OPENAI_API_KEY="))).toBe(true);
-    expect(inferenceArgs.some((a) => a.includes("inference-api.nvidia.com"))).toBe(true);
-
     expect(endpointsArgs).toContain("nvidia-endpoints");
     expect(endpointsArgs.some((a) => a.startsWith("NVIDIA_API_KEY="))).toBe(true);
     expect(endpointsArgs.some((a) => a.includes("integrate.api.nvidia.com"))).toBe(true);
@@ -245,7 +239,7 @@ describe("key hashing", () => {
     expect(hashKey("abc")).toBe(hashKey("abc"));
   });
 
-  it("TC-K16: provider updates cover both nvidia-inference and nvidia-endpoints", async () => {
+  it("TC-K16: provider updates cover nvidia-endpoints", async () => {
     execFile.mockImplementation((cmd, args, opts, cb) => {
       if (typeof opts === "function") { cb = opts; opts = {}; }
       cb(null, "", "");
@@ -260,7 +254,6 @@ describe("key hashing", () => {
     const updateCalls = execFile.mock.calls.filter(
       (c) => c[0] === "nemoclaw" && c[1]?.includes("update")
     );
-    expect(updateCalls.some((c) => c[1].includes("nvidia-inference"))).toBe(true);
     expect(updateCalls.some((c) => c[1].includes("nvidia-endpoints"))).toBe(true);
   });
 });

brev/welcome-ui/__tests__/providers.test.js

@@ -41,7 +41,7 @@ describe("GET /api/providers", () => {
     execFile.mockImplementation((cmd, args, opts, cb) => {
       if (typeof opts === "function") { cb = opts; opts = {}; }
       if (args?.[1] === "list") {
-        return cb(null, "nvidia-inference\n", "");
+        return cb(null, "nvidia-endpoints\n", "");
       }
       if (args?.[1] === "get") {
         return cb(null, FIXTURES.providerGetOutput, "");
@@ -54,7 +54,7 @@ describe("GET /api/providers", () => {
     expect(res.body.ok).toBe(true);
     expect(Array.isArray(res.body.providers)).toBe(true);
     expect(res.body.providers.length).toBe(1);
-    expect(res.body.providers[0].name).toBe("nvidia-inference");
+    expect(res.body.providers[0].name).toBe("nvidia-endpoints");
   });
 
   it("TC-PR02: provider list CLI failure returns 502", async () => {

brev/welcome-ui/__tests__/setup.js

@@ -35,11 +35,11 @@ function readCacheFile() {
 // CLI output fixtures matching the nemoclaw CLI text format
 
 const FIXTURES = {
-  providerListOutput: "nvidia-inference\ncustom-provider\n",
+  providerListOutput: "nvidia-endpoints\ncustom-provider\n",
 
   providerGetOutput: [
     "Id:              abc-123",
-    "Name:            nvidia-inference",
+    "Name:            nvidia-endpoints",
     "Type:            openai",
     "Credential keys: OPENAI_API_KEY",
     "Config keys:     OPENAI_BASE_URL",
@@ -55,19 +55,19 @@ const FIXTURES = {
 
   providerGetAnsi:
     "\x1b[32mId:\x1b[0m              abc-123\n" +
-    "\x1b[32mName:\x1b[0m            nvidia-inference\n" +
+    "\x1b[32mName:\x1b[0m            nvidia-endpoints\n" +
     "\x1b[32mType:\x1b[0m            openai\n" +
     "\x1b[32mCredential keys:\x1b[0m OPENAI_API_KEY\n" +
     "\x1b[32mConfig keys:\x1b[0m     OPENAI_BASE_URL\n",
 
   clusterInferenceOutput: [
-    "Provider:  nvidia-inference",
+    "Provider:  nvidia-endpoints",
     "Model:     meta/llama-3.1-70b-instruct",
     "Version:   2",
   ].join("\n"),
 
   clusterInferenceAnsi:
-    "\x1b[1;34mProvider:\x1b[0m  nvidia-inference\n" +
+    "\x1b[1;34mProvider:\x1b[0m  nvidia-endpoints\n" +
     "\x1b[1;34mModel:\x1b[0m     meta/llama-3.1-70b-instruct\n" +
     "\x1b[1;34mVersion:\x1b[0m   2\n",