Brain · Your model key
Your model key (hosted)
Hosted Ask runs on your key, not ours. You bring a key from Anthropic, OpenAI or OpenRouter, we pass the request through to that provider, and we do not store your question or the answer. Until you add one, hosted Ask says so rather than answering.
Why it works this way
The Privacy Policy says your prompts and the contents of your brain are not ours to see. On the desktop app that has always been literally true: the model call goes straight from your machine on your key and never touches our servers.
On the hosted tiers the call has to be made by a server, because that is where your hosted brain lives. What we changed is whose key it spends. It is yours. We hold it encrypted, use it for the request you asked for, and store neither the prompt nor the completion. That keeps the promise true rather than narrowing it to the desktop.
Add your key
Pick a provider
- Anthropic runs the same Claude models the desktop app uses. If you want hosted Ask to behave exactly like local Ask, pick this.
- OpenAI runs GPT models through the Chat Completions API.
- OpenRouter is one key for many models, and routes to Claude by default. Pick this if you would rather not open an Anthropic account.
Create an API key
console.anthropic.com/settings/keysplatform.openai.com/api-keysopenrouter.ai/keys
Paste it into My Brain
Open My Brain and ask a question. With no key stored, the connect panel appears where the answer would be. Choose the provider, paste the key, save.
Ask again
The answer streams as normal. Usage is billed to you by your provider.
What the key can and cannot do
- It is stored encrypted, and no page or API ever reads it back. You will only ever see a mask like
sk-ant-••••3f9a. Replace it or remove it any time. - It is yours alone. A key you add is never visible to your teammates, never shared across a team, and never used for anyone else's request.
- One key at a time. Switching provider replaces the key rather than keeping two, so there is never a question about which one is being spent.
- RepoOps spends it only on the request you made: your Ask, your AI entity pass, and reading text out of a photo you captured.
Without a key
- Ask asks for a key instead of answering. Search and the source cards still work; it is the synthesized answer that needs a model.
- The AI entity pass asks for a key. Your entity chips stay based on your file names, which is what they were before the pass ran.
- Photo capture still saves. It falls back to the caption you typed rather than reading the image, so nothing you captured is lost.
- The web-search connector returns no results. It uses a search tool only Anthropic serves, so it needs an Anthropic key specifically.
Last updated