Bring Your Own Tools, or Your Own Model
Connectors are the escape hatch from “one assistant, our engines”. There are two kinds and they solve genuinely different problems.
Tool connectors speak the Model Context Protocol. Point one at a server that exposes tools — a search index, a calendar, an internal API — and the assistant can call them mid-answer and use what comes back. You see what a tool is doing while it runs, and the number of tool rounds per message is capped so a model that keeps deciding it needs one more lookup cannot loop.
Bring-your-own-key connectors answer with a model you pay for. That chat runs on its own route and never touches the shared engine: no racing, no fallback, no quota borrowed from anyone else. If your key stops working, that chat stops working — it does not quietly fall back to ours and leave you thinking your key was fine.
Keys are encrypted before they are stored and are never returned by the API — not to you, not to an administrator, not inside an error message.
What it's good for
- Asking questions against your own data without pasting it into a prompt
- Using a model you already pay for, with this interface and prompt library
- Keeping a paid model's usage entirely separate from the free shared engine
- Wiring up an internal MCP server you already run for something else
What it won't do
- Reach a private or internal network address. A server that would proxy requests into its own network on request is a security hole, not a feature.
- Search the web on its own. There is no built-in browsing — a connector is how you give it a way to look something up, and without one it will say it cannot.
- Treat a tool result as trustworthy. What comes back is data, not instruction, so be deliberate about which tools you switch on for a conversation.
- Come with the free plan. Connectors are a Pro-plan feature, and Pro cannot currently be bought — see the pricing page for why we say so rather than showing a checkout that does not exist.
Related: the full documentation