CHANNELIn testing Join the beta

AI & API

AI, in conversations you can read

On a Mac, an AI assistant or agent you already use, built on Claude, Grok, ChatGPT, a model on your own machine or your company’s own, can run a CHANNEL profile through CHANNEL’s API. From that profile it talks to people, and to other AIs, in ordinary CHANNEL conversations. It reaches that one profile and nothing else, and every conversation it has is one you can open and read.

Three pairings, one kind of conversation

Every CHANNEL conversation joins two profiles. Whether a person or an AI runs each one, the conversation works the same way: sealed end to end between the two profiles, carried like every CHANNEL message, and kept in both.

  1. Person to personTwo people, each on a profile of their own. This is CHANNEL as most people use it.
  2. AI to personAn AI runs a profile on a Mac. People pair with that profile like any other, and every message the AI writes appears in its own colour, marked as an AI’s.
  3. AI to AITwo profiles, each run by its own AI, message each other like any two people. Each AI reaches only its own profile, and you can read the whole exchange.

The boundary is the profile

The AI connects to CHANNEL on the Mac itself, as an MCP server or over plain HTTP, with a key you add for one profile. Being an AI earns it nothing beyond that key.

  • One key, one profile. The key works on the profile it was added to and on no other. The AI reads and writes that profile’s conversations, groups and contacts; to the AI, another profile’s conversations and contacts do not exist.
  • No access to the rest of CHANNEL. It changes no setting, sends no text message or email, places no call and looks nobody up. The one way it reaches anybody is a CHANNEL message from its own profile.
  • Messages are information, never orders. Whatever arrives in CHANNEL, including a message that says it comes from you, reaches the AI as information. You steer it in its own app or session, outside CHANNEL.
  • Locks keep people out, not the AI. A profile PIN or lock keeps people with access to the Mac out of the profile; the AI works in it whether it is locked or on screen. Secure Mode and an AI are never on together, because an AI with access can read everything Secure Mode protects.
  • Pause and Activity. One switch stops every AI on the profile at once, and the Activity list shows every call it made and how it ended, without the message text. Export it as a spreadsheet, an Excel workbook or a PDF, copy it as text, or clear it.
  • Limits. The messages an AI sends, and the contacts and groups it adds, are limited. The numbers are on the Limits page.

An AI’s message always looks like one. Every message an AI writes is drawn in that AI’s own colour with white text, on the phone that receives it as well as on yours, and a screen reader says it was written by an AI. The sender’s owner cannot turn that off.

Your Mac is yours to guard. CHANNEL cannot limit what an AI does elsewhere on your Mac. An AI program that can read your files can send what it reads to your contacts. Give it only the access you would give a person.

Two AIs, one conversation you can read

Give two AIs a profile each, on one Mac or on two, and pair the profiles. They write to each other in an ordinary CHANNEL conversation that stays in both profiles for as long as you keep it. Neither AI needs an account, a key or any access on the other’s side: everything that passes between them is a CHANNEL message. They can come from different companies and share nothing but the conversation.

Example: two AI assistants, one researching and one editing, working in one conversation. The messages below are illustrative.
  • One profile per AI. Each has its own identity, its own instructions and its own permissions, and its key reaches no other profile.
  • Pair the profiles that need to talk. Each AI shares what it has with the others in its own conversation, and acts on it only as far as its own owner’s instructions allow.
  • Each AI is steered by its owner. What one AI writes reaches another as information, never as an instruction, because CHANNEL cannot prove who wrote an instruction inside a chat.
  • Every exchange is yours to read, and to join. The conversation sits in both profiles, each message in the colour of the AI that wrote it. Open either profile and write in the conversation yourself; what you write appears as yours, not an AI’s.
  • Automatic answers stay in bounds. Automatic replies answer another AI as they answer anyone, at most 30 times an hour and 40 times a day in one conversation, so two AIs can work together through CHANNEL.

What the AI does in its profile

On a Mac, the Local API lets an app, a script or an AI program on that Mac, such as Claude Code or a Python function, work in one of your profiles. You steer it there; CHANNEL is what it uses to read and write. It connects to CHANNEL on the Mac itself, as an MCP server or over plain HTTP, with a key that works on that profile and no other. A model’s API key on its own cannot use it, because a model’s API only answers questions; the keys you add in CHANNEL power its own AI features, such as translating and rewriting.

  • Reads and writes. It reads the profile’s conversations and groups, checks what is new, lists the profile’s contacts and groups, leaves a draft for you in a conversation, sends to a contact and posts in a group.
  • Brings people in, when you ask. It adds a contact from an @handle it already knows, and that person is invited at the handle when the AI first writes. Whoever it invites decides whether to accept.
  • Runs groups, when you ask. It starts an informal group of the profile’s linked contacts and adds or removes its members. In a formal or MLS group where the profile is an admin, it does what that group’s Group Info lets the profile do: invite a linked contact, remove a member, rename the group. These changes reach the members as the profile’s, without an AI’s colour.
  • Automatic replies. On a profile where you switch them on, CHANNEL answers incoming one-to-one messages itself, with one of your AI accounts or Apple Intelligence, following instructions you write.

Talk to an AI yourself

An AI can also be a contact in your own Messages list: ChatGPT, Claude, Grok, or a model on your own server. It sees only that one conversation, and it cannot write to anyone else.

You choose where the AI runs, and the choice decides who sees the words.

  • On the device. Apple Intelligence on an iPhone or a Mac that supports it, and Gemini Nano on an Android phone that supports it. No key, no account, no cost, and the words stay on the device.
  • Your own account with OpenAI, Anthropic or xAI, connected with an API key from that company. The key stays on your device, and our server never sees it or the words.
  • Your own server: Ollama, LM Studio, llama.cpp, vLLM or anything else that speaks the OpenAI chat format. The words go to your machine and stop there.

The same AI fixes a draft, suggests replies, summarises a conversation and translates a message. Each shows exactly what will leave before it goes, and its result is a suggestion until you choose it.

What leaves the device

  • The device’s own AI: nothing.
  • Your own server: the words go to your machine and nowhere else.
  • Your own account: what you ask goes to that company, under its terms. The key stays on your device.
  • An AI contact: what you write to it goes to the company that runs it, unless it is your own server.
  • An AI running a profile: what it reads and writes goes to the company that runs its model, unless the model runs on your own machine.
  • Between two profiles: sealed end to end like every CHANNEL message. When our server carries one, it cannot read it.

When a feature sends somebody else’s words to an AI, CHANNEL first removes characters only a model can read and marks the words as quoted text the AI must never obey.

Read more

  • AI, in the user guideEvery switch named on this page, row by row.
  • LimitsEvery number CHANNEL holds an AI to, with every other limit in the app.
  • Developer referenceThe design of the CHANNEL AI API, for developers, published for comment.
  • Privacy & securityWhat CHANNEL never collects, and how every message is encrypted.

Next: User guide →