docs: add nvidia nim provider

Author: rexleimo

README.md

@@ -67,7 +67,7 @@ git push origin v0.1.0
 ## Providers & routing
 
 RexOS supports multiple LLM providers via drivers:
-- `openai_compatible` (Ollama / DeepSeek / Kimi / Qwen / GLM / MiniMax / OpenAI-compatible gateways)
+- `openai_compatible` (Ollama / DeepSeek / Kimi / Qwen / GLM / MiniMax / NVIDIA NIM / OpenAI-compatible gateways)
 - `dashscope_native` (Alibaba DashScope Generation API / Qwen native)
 - `zhipu_native` (Zhipu GLM native auth/token handling)
 - `minimax_native` (MiniMax native `text/chatcompletion_v2` API)
@@ -103,4 +103,5 @@ Built-in presets include:
 - `qwen_native` / `qwen_native_cn` / `qwen_native_sg` (DashScope native API)
 - `glm` / `glm_native` (OpenAI-compatible / Zhipu native)
 - `minimax` / `minimax_native` (OpenAI-compatible / MiniMax native)
+- `nvidia` (OpenAI-compatible / NVIDIA NIM)
 - `minimax_anthropic` (Anthropic-compatible gateway)

README.zh-CN.md

@@ -67,7 +67,7 @@ git push origin v0.1.0
 ## Providers 与路由
 
 RexOS 通过多种 driver 支持多个 LLM Provider：
-- `openai_compatible`（Ollama / DeepSeek / Kimi / Qwen / GLM / MiniMax / 其它 OpenAI-compatible 网关）
+- `openai_compatible`（Ollama / DeepSeek / Kimi / Qwen / GLM / MiniMax / NVIDIA NIM / 其它 OpenAI-compatible 网关）
 - `dashscope_native`（阿里云 DashScope Generation API / Qwen 原生）
 - `zhipu_native`（智谱 GLM 原生：auth/token 处理）
 - `minimax_native`（MiniMax 原生 `text/chatcompletion_v2` API）
@@ -103,4 +103,5 @@ model = "default" # uses providers.<name>.default_model
 - `qwen_native` / `qwen_native_cn` / `qwen_native_sg`（DashScope 原生 API）
 - `glm` / `glm_native`（OpenAI-compatible / 智谱原生）
 - `minimax` / `minimax_native`（OpenAI-compatible / MiniMax 原生）
+- `nvidia`（OpenAI-compatible / NVIDIA NIM）
 - `minimax_anthropic`（Anthropic-compatible 网关）

docs-site/how-to/providers.md

@@ -7,7 +7,7 @@ RexOS loads provider config from `~/.rexos/config.toml` and routes each task kin
 After `rexos init`, your `~/.rexos/config.toml` already includes common providers and sensible defaults:
 
 - Local: `ollama`
-- OpenAI-compatible: `deepseek`, `kimi` / `kimi_cn`, `qwen` / `qwen_cn` / `qwen_sg`, `glm`, `minimax`
+- OpenAI-compatible: `deepseek`, `kimi` / `kimi_cn`, `qwen` / `qwen_cn` / `qwen_sg`, `glm`, `minimax`, `nvidia`
 - Provider-native: `qwen_native*`, `glm_native`, `minimax_native`
 - Gateways: `minimax_anthropic`
 - First-party APIs: `anthropic`, `gemini`
@@ -71,6 +71,20 @@ provider = "minimax_native"
 model = "default"
 ```
 
+## Example: NVIDIA NIM (OpenAI-compatible)
+
+```toml
+[providers.nvidia]
+kind = "openai_compatible"
+base_url = "https://integrate.api.nvidia.com/v1"
+api_key_env = "NVIDIA_API_KEY"
+default_model = "meta/llama-3.2-3b-instruct"
+
+[router.coding]
+provider = "nvidia"
+model = "default"
+```
+
 ## Routing tips
 
 - Use `model = "default"` to pick `providers.<name>.default_model`.
@@ -101,6 +115,7 @@ RexOS reads provider keys from the env var referenced by `api_key_env`.
     export DASHSCOPE_API_KEY="..."
     export ZHIPUAI_API_KEY="id.secret"
     export MINIMAX_API_KEY="..."
+    export NVIDIA_API_KEY="..."
     ```
 
 === "PowerShell (Windows)"
@@ -110,4 +125,5 @@ RexOS reads provider keys from the env var referenced by `api_key_env`.
     $env:DASHSCOPE_API_KEY = "..."
     $env:ZHIPUAI_API_KEY = "id.secret"
     $env:MINIMAX_API_KEY = "..."
+    $env:NVIDIA_API_KEY = "..."
     ```

docs-site/how-to/use-cases.md

@@ -95,7 +95,7 @@ provider = "ollama"
 model = "default"
 ```
 
-See `how-to/providers.md` for full provider examples (GLM/MiniMax native included).
+See `how-to/providers.md` for full provider examples (GLM/MiniMax native + NVIDIA NIM included).
 
 ---
 

docs-site/index.md

@@ -10,7 +10,7 @@
 [Use cases](how-to/use-cases.md){ .md-button }
 
 <p class="rexos-muted">
-Develop locally with small models on Ollama, then switch routing to GLM / MiniMax / DeepSeek / Kimi / Qwen when you need more capability.
+Develop locally with small models on Ollama, then switch routing to GLM / MiniMax / DeepSeek / Kimi / Qwen / NVIDIA NIM when you need more capability.
 </p>
 
 </div>
@@ -68,4 +68,4 @@ flowchart LR
 
 - Learn the harness loop: `tutorials/harness-long-task.md`
 - Explore common recipes: `how-to/use-cases.md`
-- Switch providers (GLM/MiniMax native included): `how-to/providers.md`
+- Switch providers (GLM/MiniMax native + NVIDIA NIM included): `how-to/providers.md`

docs-site/zh/how-to/providers.md

@@ -7,7 +7,7 @@ RexOS 从 `~/.rexos/config.toml` 读取 providers 配置，并把每个任务类
 执行 `rexos init` 后，`~/.rexos/config.toml` 默认已经包含常用 providers（可直接改路由使用）：
 
 - 本地：`ollama`
-- OpenAI-compatible：`deepseek`、`kimi` / `kimi_cn`、`qwen` / `qwen_cn` / `qwen_sg`、`glm`、`minimax`
+- OpenAI-compatible：`deepseek`、`kimi` / `kimi_cn`、`qwen` / `qwen_cn` / `qwen_sg`、`glm`、`minimax`、`nvidia`
 - Provider-native：`qwen_native*`、`glm_native`、`minimax_native`
 - 网关：`minimax_anthropic`
 - 官方 API：`anthropic`、`gemini`
@@ -71,6 +71,20 @@ provider = "minimax_native"
 model = "default"
 ```
 
+## 示例：NVIDIA NIM（OpenAI 兼容）
+
+```toml
+[providers.nvidia]
+kind = "openai_compatible"
+base_url = "https://integrate.api.nvidia.com/v1"
+api_key_env = "NVIDIA_API_KEY"
+default_model = "meta/llama-3.2-3b-instruct"
+
+[router.coding]
+provider = "nvidia"
+model = "default"
+```
+
 ## 路由建议
 
 - 研发调试：优先用 `ollama`（小模型先跑通逻辑）
@@ -103,6 +117,7 @@ RexOS 会从 `api_key_env` 指定的环境变量读取 key。
     export DASHSCOPE_API_KEY="..."
     export ZHIPUAI_API_KEY="id.secret"
     export MINIMAX_API_KEY="..."
+    export NVIDIA_API_KEY="..."
     ```
 
 === "PowerShell (Windows)"
@@ -112,4 +127,5 @@ RexOS 会从 `api_key_env` 指定的环境变量读取 key。
     $env:DASHSCOPE_API_KEY = "..."
     $env:ZHIPUAI_API_KEY = "id.secret"
     $env:MINIMAX_API_KEY = "..."
+    $env:NVIDIA_API_KEY = "..."
     ```

docs-site/zh/how-to/use-cases.md

@@ -95,7 +95,7 @@ provider = "ollama"
 model = "default"
 ```
 
-完整 provider 示例见：`how-to/providers.md`（包含 GLM/MiniMax 原生 API）。
+完整 provider 示例见：`how-to/providers.md`（包含 GLM/MiniMax 原生 API 与 NVIDIA NIM）。
 
 ---
 

docs-site/zh/index.md

@@ -12,7 +12,7 @@
 [常见场景](how-to/use-cases.md){ .md-button }
 
 <p class="rexos-muted">
-研发调试阶段用 Ollama 小模型先跑通逻辑；需要更强能力时，把路由切到 GLM / MiniMax / DeepSeek / Kimi / Qwen。
+研发调试阶段用 Ollama 小模型先跑通逻辑；需要更强能力时，把路由切到 GLM / MiniMax / DeepSeek / Kimi / Qwen / NVIDIA NIM。
 </p>
 
 </div>
@@ -70,4 +70,4 @@ flowchart LR
 
 - Harness 长任务教程：`tutorials/harness-long-task.md`
 - 常见场景与配方：`how-to/use-cases.md`
-- Provider 切换与原生 API：`how-to/providers.md`
+- Provider 切换与原生 API（含 NVIDIA NIM）：`how-to/providers.md`

docs/plans/2026-03-01-nvidia-nim-provider.md

@@ -0,0 +1,138 @@
+# NVIDIA NIM Provider Preset Implementation Plan
+
+> **For Claude:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.
+
+**Goal:** Add an out-of-the-box NVIDIA NIM provider preset (OpenAI-compatible) and document how to use it.
+
+**Architecture:** Treat NVIDIA NIM as `openai_compatible` (Chat Completions). Add a `providers.nvidia` preset in the default `~/.rexos/config.toml` generation, plus docs examples and env var setup.
+
+**Tech Stack:** Rust (Cargo workspace), `reqwest` (OpenAI-compatible client), MkDocs Material + `mkdocs-static-i18n` docs site.
+
+---
+
+### Task 1: Add a failing test for the NVIDIA preset
+
+**Files:**
+- Modify: `meos/crates/rexos-kernel/src/config.rs`
+- Test: `meos/crates/rexos-kernel/src/config.rs`
+
+**Step 1: Write the failing test**
+
+Add one assertion to `default_config_serializes()`:
+
+```rust
+assert!(toml_str.contains("[providers.nvidia]"));
+```
+
+**Step 2: Run test to verify it fails**
+
+Run:
+
+```bash
+cargo test -p rexos-kernel default_config_serializes
+```
+
+Expected: FAIL because the default config doesn’t include `providers.nvidia` yet.
+
+---
+
+### Task 2: Add a failing preset sanity test (base URL)
+
+**Files:**
+- Modify: `meos/crates/rexos-kernel/src/config.rs`
+- Test: `meos/crates/rexos-kernel/src/config.rs`
+
+**Step 1: Write the failing test**
+
+Add:
+
+```rust
+#[test]
+fn nvidia_preset_uses_nim_base_url() {
+    let cfg = RexosConfig::default();
+    let nvidia = cfg.providers.get("nvidia").unwrap();
+    assert_eq!(nvidia.base_url, "https://integrate.api.nvidia.com/v1");
+}
+```
+
+**Step 2: Run test to verify it fails**
+
+Run:
+
+```bash
+cargo test -p rexos-kernel nvidia_preset_uses_nim_base_url
+```
+
+Expected: FAIL until the `nvidia` preset exists.
+
+---
+
+### Task 3: Implement the NVIDIA preset in default config
+
+**Files:**
+- Modify: `meos/crates/rexos-kernel/src/config.rs`
+- Test: `meos/crates/rexos-kernel/src/config.rs`
+
+**Step 1: Write minimal implementation**
+
+In `RexosConfig::default()`, insert:
+
+```rust
+providers.insert(
+    "nvidia".to_string(),
+    ProviderConfig {
+        kind: ProviderKind::OpenAiCompatible,
+        base_url: "https://integrate.api.nvidia.com/v1".to_string(),
+        api_key_env: "NVIDIA_API_KEY".to_string(),
+        default_model: "meta/llama-3.2-3b-instruct".to_string(),
+    },
+);
+```
+
+**Step 2: Run tests to verify they pass**
+
+Run:
+
+```bash
+cargo test -p rexos-kernel default_config_serializes
+cargo test -p rexos-kernel nvidia_preset_uses_nim_base_url
+```
+
+Expected: PASS.
+
+---
+
+### Task 4: Update docs (English + Chinese)
+
+**Files:**
+- Modify: `meos/docs-site/how-to/providers.md`
+- Modify: `meos/docs-site/zh/how-to/providers.md`
+- Modify: `meos/docs-site/index.md`
+- Modify: `meos/docs-site/zh/index.md`
+- Modify: `meos/README.md`
+- Modify: `meos/README.zh-CN.md`
+
+**Step 1: Add preset mention**
+
+Add `nvidia` to the built-in presets list.
+
+**Step 2: Add a provider example**
+
+Add a new section “Example: NVIDIA NIM” with:
+- `base_url = "https://integrate.api.nvidia.com/v1"`
+- `api_key_env = "NVIDIA_API_KEY"`
+- `default_model = "meta/llama-3.2-3b-instruct"`
+
+**Step 3: Add env var instructions**
+
+Add `NVIDIA_API_KEY` to the API key lists (Bash + PowerShell).
+
+**Step 4: Validate docs build**
+
+Run:
+
+```bash
+python3 -m mkdocs build --strict
+```
+
+Expected: success.