AI and data
Updated 26 October 2026.
Some words on Terminal are written by a language model. The model writes a short note from facts the software collected, and it answers chat. The software chooses the facts. The model chooses the sentences. Chat is a model reply. It is not a person at Terminal.
Where the product shows a model-written note, it labels that note as AI-written and says the numbers come from market data. A phone notification carries facts the software rendered. It does not carry model prose. If a number is not in the data for that run, treat the note as incomplete.
What we send
When a note or a chat reply is produced, we send the model provider the text needed to write it:
- Your instructions for that agent
- Market facts for that run, such as prices and headlines the data source returned
- Notes already stored for that agent
- The names of accounts you connected
- Text read from a connected account when that run uses it
- The chat message you typed
We do not add your password to that prompt. We do not add a card number. We do not add the secret tokens for connected accounts on purpose. Text that came back from a connected account can still contain whatever that account returned.
Which provider writes the text depends on whether your account pays:
- Paying accounts. Anthropic (
api.anthropic.com), with its Claude model. Anthropic is based in the United States. - Everyone else. This covers free accounts, trials, and accounts with a beta or complimentary grant. For now, a Grok program on our server sends the prompt to xAI, under our own xAI subscription. xAI is based in the United States. This is a temporary arrangement, and we will update this notice when it changes.
A non-paying account's text is never sent to Anthropic. On a host set up another way, Groq (api.groq.com) or xAI's API (api.x.ai) can take either place. Both are based in the United States. An xAI API request sets store to false, which turns off saved conversation state on that call. That flag is not, by itself, a promise that xAI keeps no copy.
Each provider receives the text listed above. That is the alert facts and your question. It never receives your password, your API keys, or your payment details. If no provider is available for your account, notes show the market facts only and chat says answers are not available. [Counsel will attach each provider's retention terms when the production contract is signed.] We do not sell that text. We do not use your instructions or notes to train a model of our own.
What the model is not
The model can be wrong. It can omit a fact that was in the input. It can add a sentence the facts do not support. We do not have a person review each note before you see it. The product is supposed to show a missing figure as unavailable rather than as zero.
A model note is a description of the input. It is not a prediction we stand behind. It is not investment advice. Read the not investment advice notice.
We do not use the model to decide a trade for you, and we do not send an order to an exchange for you.
Safety check
When the safety check is configured, Typesafe (api.typesafe.ai) receives the text being checked. That text is a chat message, or an agent's name and instructions. Typesafe returns a decision. We store the decision, the stage, and how long the check took. We delete that log when you delete the account. If the safety check is not configured, we do not send that text to Typesafe.
The check can block a message or an agent setup. It does not decide credit, housing, employment, or insurance. Write to us if you want a person to review a block.
Your choices
Do not put secrets, passwords, or other people's private data into instructions or chat. Disconnect an account on the connections page if you do not want later runs to read it. We do not keep a chat transcript. We do keep the safety-check decision until you delete the account. Deleting the account also removes your instructions, notes, and connection tokens. The Privacy Policy states the rest.