# AI models

Claude or GPT with your own API key. @ it in a thread and it answers with a note.

## Add a key

AI models need Pro with Sync turned on. In **Settings → Connections**, open **AI models**, choose **Claude** or **GPT**, paste an API key from Anthropic or OpenAI, and choose **Connect**. ThreadPage checks the key with the provider and asks you to confirm it’s you.

Only an API key works. A Claude or ChatGPT subscription can’t be used here.

## Ask it in a thread

Write a note to `@Claude` or `@GPT` and choose **Send**. The model answers with a note in the same thread, which names the model: **From @Claude · Claude Sonnet 5.5**. Only you see **Also read** beside it, which opens the searches it ran and the notes it read. **Stop this request** ends an answer while it runs.

## What it reads

A model reads only what everyone who will see its answer can already read. In a thread only you see, that is everything its access allows. In a thread you share, it answers only after you choose **Allow here**, and then reads only what is shared there.

It reads and searches your notes to answer, and its answer is the only thing it writes. ThreadPage sends what it reads to Anthropic or OpenAI with your key.

## Your key

ThreadPage keeps the key encrypted, and only the server that calls the provider can read it. Settings shows the provider and the key’s last four characters, never the key.

**Replace key…** changes it. **Disconnect…** deletes it.

## Costs and limits

The provider bills your key for every answer. **Recent activity** on the model’s page shows where it answered, what it read and the tokens it used. A daily token limit stops answers for the rest of the day once it is reached; you set it in the model’s details.

When the provider refuses an answer, the note shows why, such as **Key not accepted**. [Troubleshooting connections](https://thread.page/help/connections/troubleshooting.md) lists what to do.