/settings/ai. One card per provider.
AI models

​
Providers

Amazon Bedrock is the provider you can connect today, using an AWS access key, secret and region. It serves the Claude models — Opus 4.5 and 4.6, Sonnet 4.5 and 4.6, Haiku 4.5 — each tagged with what it can do. OpenAI, Anthropic direct and Google AI appear on the page as cards marked coming soon.

​
The config drawer

Clicking a provider card opens a drawer with two tabs. Credentials lists what you have connected and adds more, with per-credential test, reveal and delete. Info carries the pricing reference, the documentation link, and the region. The model rows underneath carry capability tags. Clicking one assigns that model to a slot.

​
Chat and embed slots

Slots are set under /settings, and each is a pair: a provider account and a model.
SlotUsed by
ChatThe inbox AI agent, AI Generate, the AI nodes, lead scoring
EmbedIndexing knowledge base documents so the agent can search them
The model list comes live from the provider, so it reflects what your account can actually call. When the provider cannot list models, the picker falls back to a short preset list and lets you type a model id directly.
Anthropic has no embedding model. A workspace using Anthropic for chat still needs Bedrock or OpenAI for the embed slot, or knowledge-base search has nothing to index with.

​
Cost

API keys are encrypted at rest, but EDGR cannot cap what you spend at the provider. Set a budget limit in the provider’s own console — AWS Budgets for Bedrock — before pointing a campaign at a few thousand customers.