Some coding harnesses (Claude Code, certain IDE plugins) let you set the API base URL and key, but pin the model name and will not let you change it. That makes it hard to choose which Northern Inference route your traffic uses.
NI commands solve this. You type a command as your prompt, and the gateway answers directly with a normal assistant message. These commands never call a model and are never billed. They act only on your own account.
Commands
| Command | What it does |
|---|---|
ni-help | Lists the available commands. |
ni-models | Lists the model routes your key can call, numbered. |
ni-map-model | Lists routes, then you reply to pick one to map your pinned model to. |
ni-map-model 2 | Maps your current (pinned) model to route number 2 from the list. |
ni-map-model <route-id> | Maps your current model to an exact route id. |
ni-unmap | Removes the mapping for your current model. |
ni-status | Shows your active model mappings. |
How to map a model
- In your harness, send
ni-map-modelas the prompt. You get back a numbered list
of routes your key can call.
- Reply with
ni-map-model <number>(for exampleni-map-model 3), or paste an
exact route id with ni-map-model bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0-us.
- The gateway confirms the mapping is saved. From then on, every request your
harness sends with its pinned model name is routed to the route you chose. No change to the harness config is needed.
To undo it, send ni-unmap. To see what is mapped, send ni-status.
Notes
- Mappings are saved to your account (team scope), so they apply to every key on
the account for that pinned model name.
- The command must be the start of your message (for example
ni-help). A normal
prompt that merely mentions a command is not intercepted.
- These commands work the same whether your harness streams responses or not.
- If your administrator has disabled NI commands, the prompt is sent to the model
as normal text.