/settings/ai. One card per provider.
3 supported providers
| Provider | Auth | Models |
|---|---|---|
| OpenAI | API key (sk-...) | Chat (gpt-4o*, o1*) + Embed (text-embedding-3-*) |
| Anthropic | API key (sk-ant-...) | Chat (claude-sonnet-4-5, claude-haiku-4-5, claude-opus-4-7). No embed. |
| Bedrock | AWS access key + secret + region | Chat (Anthropic via Bedrock), Embed (Cohere, Titan) |
Config drawer
Click a provider card → drawer opens with two tabs:- Credentials
- Info
- List of credentials already connected for this provider
- + Add credential — connect modal
- Per-credential actions: test, reveal secret (admin), delete
Chat & Embed slots
Configure the slots under/settings (user-level):
- Chat slot — autoreply bot, AI campaign generation
- Embed slot — document indexing (RAG), semantic search
An Anthropic-only workspace needs an extra provider (OpenAI or Bedrock) for the embed slot — Anthropic doesn’t have an embed model.
Provider model row
A table of models available for the provider, tagged with capability badges (chat / embed). Click a row → quick-assigns the model to the Chat or Embed slot.
AI Usage page (cost/latency/error analytics), provider fallback (primary → backup), per-org token cap, alerts at 80% of cap — coming soon.