NIM Model Bridge
Use the NVIDIA NIM model catalog from Claude Code, Codex CLI, GitHub Copilot Chat and any OpenAI- or Anthropic-compatible AI tool, all at once, through one local bridge.
Your own Claude, OpenAI and Copilot models keep working exactly as before: NIM models are added to each tool's model list, never swapped in.
Features
- Several NIM models per tool, each selectable inside the tool itself:
- Claude Code: NIM models appear as rows in the
/model picker.
- Codex CLI: one profile per model, e.g.
codex --profile nim-z-ai-glm-5-3-flash.
- Copilot Chat: NIM models appear in the Copilot model picker.
- Local bridge on
127.0.0.1 that speaks the OpenAI Chat Completions, OpenAI Responses and Anthropic Messages APIs, with streaming, tool calls and images.
- Your NVIDIA key stays private: stored in VS Code SecretStorage and never shown to the tools.
- Works in VS Code-based editors (VS Code, Cursor, Windsurf, Trae, Positron, VSCodium and others), plus a standalone
nim-bridge CLI for editors that can't load VS Code extensions.
- Clean undo: stopping the bridge removes only what it added to your Claude Code and Codex settings.
Quick start
- Get an API key at build.nvidia.com.
- Click Set API Key in the start-up prompt, or run NIM Bridge: Set NVIDIA API Key from the Command Palette. If the status bar still shows NIM off, run NIM Bridge: Start Bridge Server.
- Run NIM Bridge: Select NIM Models (All Providers), pick your models, choose the tools and pick a default.
- Use them:
| Tool |
Where the NIM models appear |
| Claude Code |
/model picker. Restart Claude Code after selecting. |
| Codex CLI |
codex --profile nim-<model>. NIM Bridge: Open Status Dashboard lists the exact profile names. |
| Copilot Chat |
Model picker → Manage Models… → NVIDIA NIM |
| Any other tool |
Base URL http://127.0.0.1:47821/v1, API key = bridge token (from NIM Bridge: Copy Provider Settings for Any IDE), model = a NIM model id |
Commands
| Command |
What it does |
| NIM Bridge: Start / Stop / Restart Bridge Server |
Control the local bridge |
| NIM Bridge: Select NIM Models (All Providers) |
Pick models for several tools at once |
| NIM Bridge: Select NIM Models for Claude Code / Codex CLI / Copilot Chat |
Pick models for one tool |
| NIM Bridge: Open Status Dashboard |
Health, models per tool, live request log |
| NIM Bridge: Copy Provider Settings for Any IDE |
Base URL, token and model ids for other tools |
| NIM Bridge: Set / Clear NVIDIA API Key |
Manage your key |
Key settings
| Setting |
Default |
Purpose |
nimBridge.server.port |
47821 |
Local bridge port |
nimBridge.models.claudeList / codexList / copilotList |
[] |
NIM models offered to each tool (empty Copilot list = whole catalog) |
nimBridge.models.claude / codex / copilot |
"" |
Each tool's default NIM model |
nimBridge.claude.behavesAs |
claude-sonnet-4-5 |
Lets Claude Code show NIM models it doesn't recognise; the NIM model is still the one used |
nimBridge.baseUrl |
NVIDIA hosted API |
Point at a self-hosted NIM endpoint instead |
Requirements
- VS Code 1.82 or a compatible editor. The Copilot picker needs VS Code 1.104+.
- Claude Code 2.1.242+ for the
/model rows; Codex CLI 0.134+ for profiles.
- An NVIDIA API key. The standalone CLI needs Node.js 18+.
Security notes
- The bridge only listens on
127.0.0.1. The bridge token is not an access control: any program on your machine that can reach the port can use your NVIDIA quota. Don't expose the port.
- Requests for non-NIM models (
claude-*, gpt-*) pass straight through to their own provider with your own credentials.
Known limitations
- Each NIM model has its own context and output limits; prefer models with large output limits for agentic work.
- Free-tier NIM models can take a while to answer the first request (cold starts).
- Editors other than VS Code are supported through the same extension API but are not individually tested.
MIT licensed. Not affiliated with NVIDIA, Anthropic, OpenAI or GitHub. NVIDIA and NIM are trademarks of NVIDIA Corporation.
| |
