Models & providers
Niah can talk to your organization's models, a model server running on your own machine, or a provider you hold the key for. You choose per conversation from the provider and model picker in the sidebar.
Niah Cloud
The default, and the one to use unless you have a reason not to. Signing in with your Niah account enables the Niah Cloud provider, which routes chat through your organization's approved models and gives agents access to your organization's knowledge base.
Because model access is granted centrally, you do not manage any API keys yourself: your administrator decides which models the organization can use, and usage is metered against your plan. See the Quickstart for the sign-in steps.
Local models
Run a model on your own hardware and keep prompts on your machine. Niah supports the common local servers:
| Ollama | Defaults to http://localhost:11434. |
|---|---|
| LM Studio | Point Niah at the server LM Studio exposes. |
| vLLM | For a self-hosted inference server. |
Set the endpoint in Niah Settings. Local models are a good fit for routine edits on sensitive code; larger agent tasks generally do better on a frontier model.
Bring your own key
You can also point Niah at a provider you already pay for. Supported providers include:
| Direct | Anthropic, OpenAI, Google Gemini, Mistral, DeepSeek, Groq, Grok (xAI), Cerebras, Fireworks AI, GitHub Models |
|---|---|
| Cloud platforms | AWS Bedrock, Google Vertex AI, Microsoft Azure OpenAI |
| Gateways | OpenRouter, LiteLLM, and any OpenAI-compatible endpoint |
Choosing and configuring
Open Niah Settings from the Command Palette
(Preferences: Open Niah Settings) to add credentials, set
endpoints, and see which providers are configured. The provider and model
picker in the chat sidebar switches between them for a given conversation.
Plan & usage
The usage badge in the sidebar shows where you stand against your plan. Click it, or open the Plan & Usage tab in Niah Settings, to see current consumption. Usage applies to Niah Cloud — local models and your own provider keys are billed by whoever runs them.