Reference

`intelligence.provider`

domain: intelligence

A PLUGGABLE model provider (LLM/embedding), as governed data — adding a provider is a record, not code. The model-gateway reads these, routes a request by model name to the matching provider, injects that provider's key from Vault (referenced by `key_secret`, never stored here), and speaks its OpenAI-compatible API. Replaces a hardcoded llm-proxy: one gateway, many providers (Airawat-hosted, OpenAI, a local vLLM/Ollama, …), all metered.

Field	Type	Required	Description
`schema`	`—`	✓
`id`	`string`	✓
`provider_id`	`string`	✓
`display_name`	`string`
`kind`	`string`	✓	the wire protocol (OpenAI /embeddings + /chat/completions)
`base_url`	`string`	✓	e.g. https://llm.airawat.org/v1
`models`	`array<object>`	✓	[{name, task: embed\|chat, dim?}]
`key_secret`	`string`		Vault secret name holding this provider's API key (the gateway resolves it; never stored here)
`default_embed`	`boolean`
`default_chat`	`boolean`
`enabled`	`boolean`
`tenant`	`string`