Direct, zero-markup access to NVIDIA NIM's premier open & proprietary reasoning models right inside GitHub Copilot Chat.
Install • Key Features • Supported Models • Quick Start • FAQ • Commands
Free Access: NVIDIA provides free developer API at build.nvidia.com. You can use DeepSeek V4, Nemotron, and Kimi directly inside Copilot without any monthly provider subscriptions.
🚀 Why NVIDIA NIM Agent?
Modern software engineering demands high-precision reasoning, long context understanding, and reliable autonomous tool execution. NVIDIA NIM Agent connects VS Code directly to NVIDIA's high-throughput NIM Cloud infrastructure, unlocking flagship models with minimal latency, zero intermediary proxy overhead, and production-grade resilience.
✨ Key Features
🧠 Native Deep Reasoning
- Collapsible Thinking Blocks: Fully supports VS Code's native
LanguageModelThinkingPartAPI. Reasoning output from models like DeepSeek V4, Kimi K2.6, and Nemotron 3.5 renders as clean, collapsible thinking sections rather than cluttering your chat stream. - Granular Effort Control: Select reasoning modes (
None,On,Medium,High,Max) directly from the Copilot model picker dropdown.
🛡️ Zero-Downtime Smart Failover
- Transient Turn-Level Failover: If an active model experiences rate limits (
HTTP 429/529), temporary unavailability (404), an empty stream, or a slow first-token timeout (TTFT), the extension instantly routes the current turn to a lightning backup model (e.g.nvidia/nemotron-3.5-lightning-30b-a3b). - Automatic Recovery: Automatically restores your preferred primary model on the subsequent turn.
- Transparent Notifications: Clear in-chat callout badges (
> ⚡ **NVIDIA NIM Fallback:** ...) and status notifications keep you informed without interrupting workflow.
🛠️ Self-Healing Agentic Tool Execution
- Tag-Stack XML Scanner: Single-pass streaming scanner compatible with Hermes, Nemotron, Anthropic, Qwen, and DeepSeek syntax.
- Self-Healing Repair: Automatically rectifies unescaped strings, trailing JSON commas, and parameter aliases (
filePath⇄path⇄file). - Loop Prevention: Safely suppresses identical consecutive read-only operations while permitting intentional terminal command retries.
🗜️ Dedicated Context Auto-Compaction
- Decoupled Summarization: Leverages a dedicated, ultra-fast model (
context.summarizationModel) to condense lengthy dialogue history without altering your primary model configuration. - Dynamic Safety Margins: Automatically allocates safety buffers for large context windows ($\ge 256\text{K}$) to prevent sudden payload overflow errors.
📊 Real-Time Token & Latency Telemetry
- Rich Status Bar Breakdown: Real-time token utilization widget displaying exact breakdowns across system prompts, tool schemas, user history, thinking tokens, and completions.
- Developer Metrics: Optional millisecond-level TTFT and generation tokens-per-second logs for benchmarking model performance.
🌟 Supported Models
The extension connects to official NVIDIA NIM endpoints (https://integrate.api.nvidia.com/v1) and provides model-specific prompt adapters:
| Model | Context Window | Reasoning Modes | Tool Calling | Vision | Best For |
|---|---|---|---|---|---|
| DeepSeek V4 Flash 0731 | 1M | None, High, Max |
✅ Yes | ❌ | Deep algorithm design, code architecture, complex refactoring |
| Kimi K2.6 (Deprecated) | 262K | None, On |
✅ Yes | ✅ Yes | Long-context repository comprehension, multimodal review (auto-fails over to MiniMax M3) |
| Nemotron 3.5 Lightning 30B | 1M | None, Medium, High, XHigh |
✅ Yes | ❌ | Ultra-fast responses, agentic tool workflows, summarization |
| Nemotron 3 Ultra 550B | 1M | None, Medium, High |
✅ Yes | ❌ | Heavy multi-step reasoning, deep technical documentation |
| MiniMax M3 | 1M | None, On, Adaptive |
✅ Yes | ✅ Yes | Multimodal code generation, full-stack tasks |
| GLM 5.2 | 1M | None, On |
✅ Yes | ❌ | Precise instruction following, rigorous logic |
| Step 3.7 Flash | 262K | Always On |
✅ Yes | ✅ Yes | Rapid thinking loops, interactive pair programming |
| Inkling | 1M | None to Max (7 levels) |
✅ Yes | ✅ Yes | Ultra-deep analytical inspection |
| Muse Glimmer | 131K | None to XHigh |
✅ Yes | ✅ Yes | Visual UX/UI analysis and front-end generation |
⚡ Quick Start
1. Install
- Install from the VS Code Marketplace, or
- Quick Open (
Ctrl + P/Cmd + P):ext install neuraldock.nvidia-nim-agent
2. Requirements
- VS Code 1.125.0 or later
- GitHub Copilot extension installed and active
- NVIDIA NIM API Key (Get free credits at build.nvidia.com)
3. Configure Your API Key
- Open Copilot Chat (
Ctrl + Alt + I/Cmd + Alt + I). - Click the model selector dropdown $\rightarrow$ Manage Models $\rightarrow$ NVIDIA NIM.
- Paste your NVIDIA NIM API key (
nvapi-...).
(Alternatively, run NVIDIA NIM: Manage NVIDIA NIM API Key from the Command Palette).
4. Start Chatting & Coding
Select any NVIDIA NIM model in Copilot Chat or Copilot Agent Mode and start building!
⌨️ Extension Commands
Access these commands anytime from the VS Code Command Palette (Ctrl + Shift + P / Cmd + Shift + P):
| Command | Identifier | Description |
|---|---|---|
| Manage API Key | nvidia-nim.manage |
Store or update your NVIDIA NIM API key. |
| Refresh Models | nvidia-nim.refreshModels |
Force re-sync available models from your NVIDIA account. |
| Toggle Debug Logging | nvidia-nim.toggleDebugLogging |
Enable/disable verbose diagnostic logs. |
| Open Debug Log | nvidia-nim.openDebugLog |
Reveal the dedicated NVIDIA NIM Output Channel. |
🔒 Privacy & Security
- Direct Communication: Requests flow strictly between your VS Code client and the official NVIDIA NIM API (
https://integrate.api.nvidia.com/v1). There are zero third-party telemetry servers or proxy gateways. - Secure Secret Storage: API keys are encrypted and stored inside VS Code's native OS-level credential vault (
SecretStorage). - No Data Retention by Extension: The extension retains no chat logs, file contents, or personal credentials.
❓ Frequently Asked Questions
Q: Do I need a paid NVIDIA subscription?
A: No. NVIDIA provides free API on build.nvidia.com for developers.
Q: Does it work with Copilot Agent Mode and Tools?
A: Yes! Supported chat models are tool-capable and support autonomous file editing, terminal execution, and MCP tools with automatic JSON repair.
📄 License
This project is licensed under the MIT License.