Providers and models

AVAILABLE

Naagmani has provider adapters for OpenAI, Anthropic, Google Gemini and DeepSeek. Configure credentials in Organization → Providers & Health. For an OpenAI-compatible model server, configure the compatible endpoint supported by your deployment.

Choose a model #

Model availability depends on the provider account, configured credentials, endpoint compatibility and the gateway release. Use the provider catalog and the gateway's GET /v1/models response instead of assuming every provider model or API is supported.

The chat API accepts direct model identifiers and configured routing-policy aliases. A name such as smart is not automatically a working model: create the policy first. Optional tools, streaming and reasoning fields depend on the adapter and model.

Credentials and health #

Provider keys belong in the provider configuration, not in your application. Applications use a Naagmani API key or Project Service Token. Check provider health, quotas and attempt logs when requests fail.

A compatible chat endpoint does not imply support for embeddings, realtime audio, every multimodal feature or arbitrary provider APIs. The public gateway currently exposes chat completions, Responses and model listing.

See routing, provider setup and API overview.