AI Coding Gateway Documentation
Connect your coding tools to NVIDIA models. Find the right setup, understand what the service handles, and keep control of your credentials.
Create a free connection, choose your client, then download private agent access. Your local agent can handle setup and guide you through attaching an authorized NVIDIA key.
Start the setup guide →Find your guide
MCP helper
Add model assistance to an existing local agent and retain its tools.
Read the guide ↗ CLIENT SETUPClaude Code & Desktop
Choose the terminal client or Desktop’s Code tab and verify your connection.
Read the guide ↗ CLIENT SETUPCodex & OpenCode
Select the installed client version and use a reversible private setup.
Read the guide ↗ SERVICE REFERENCEArchitecture & data
Understand the route, credential boundaries and manual connection options.
Read the guide ↗ COSTS & REQUIREMENTSPricing & limits
Free routing, provider costs, optional tips and separate project listings.
Read the guide ↗ 52-SECOND WALKTHROUGHWatch setup
Follow manual key attachment and the private handoff, with captions.
Watch the video ↗Troubleshooting
- My connection is active, but a request fails.
- Check the attached provider key and selected model. The authenticated readiness check identifies missing setup requirements; it does not call NVIDIA or prove inference. Preserve the reported error and test the actual client.
- The first-key permission expired.
- Download fresh private access with the permission selected, or attach the key manually in the workspace. This permission lasts 30 minutes and cannot replace an existing key.
- My client rejects the configuration.
- Confirm the client and version selected in the workspace match the installed app. Give your local agent the error and version, with credentials removed. Use the saved Undo instructions if you need to restore the previous setup.
- A request is rate limited or times out.
- Honor the provider’s Retry-After response and reduce concurrent calls. A timeout can leave completion unknown. Inspect the original request before deciding to retry; do not automatically replay it or rotate keys to bypass limits.
- My listing payment is pending.
- Refresh your listings and check the recorded payment state before paying again. A browser redirect is not payment confirmation. Contact support with the listing or order reference.
Contact support
Email support@routemyagent.dev for setup, billing, privacy or listing concerns. Include your client and version, the approximate time and timezone, what you expected, and the error with sensitive details removed. Include a request or order reference when available.
Keep API keys, router tokens, private access files, configuration bundles and private prompts out of support messages. For a security concern, describe the affected feature and impact without including credentials or other customers’ data.
For automated agents
Start with agent instructions. Use service discovery and the public agent reference for endpoints and requirements. Client configuration is delivered through the authenticated private handoff.
Policies
Read our service terms, data and privacy notice, and project listing policy. RouteMyAgent is independent of NVIDIA and client vendors.
Hermes Agent
Select Hermes Agent CLI on the connection page. Your local agent fetches a private setup bundle, adds a named provider and creates Start and Undo commands. Your normal Hermes default, memory, skills and tools stay in place. Desktop, bots and running gateways need separate setup and testing.