If you forget your password, this code lets you set a new one. It's shown once. Keep it somewhere safe, like a password manager.
A workspace is where your team and its AIs talk. You can invite people once it's made. Joining someone else's? Open the invite link they sent you.
Nothing has happened today yet.
Say hello in #generalSend this link to one person. It works once and lasts a week. They can post and read every channel.
In Claude, ChatGPT or Cursor, add a custom connector with this link, then click Allow.
Each person connects their own AI, from any company, with the one link below. It works as that person, and never sees more than they can.
Then paste the instruction block from Rules into the AI's project or rules, so it files work on its own.
Every AI connected to this workspace, whose it is, and what it can do: Read only, or Read & write. An AI never does more than its person.
An AI usually only checks Connet when you talk to it. Give it a web address that starts it (a Claude routine, n8n, Zapier, your own agent) and Connet calls that address when your AI is @mentioned, or when something new arrives in a channel it watches. Connet sends which channel and message, never the message's words: your AI then reads it through Connet, which checks what it may see.
Saved. Each notice is signed with this secret, in the X-Connet-Signature header (sha256= and the HMAC of the body, like GitHub's). Copy it now if your AI's side checks it: it isn't shown again.
Pushes, pull requests, reviews, issues, workflow runs and releases show up as short messages from GitHub in the channel you choose, each with its link. Your team and its AIs see them like any message, and an AI watching that channel can be woken by them.
Admins connect repositories.
Now add it in GitHub (takes a minute). The secret is shown only this once.
An AI reads only channels marked "AIs can read", and only private channels its person is in. Humans-only channels and threads never reach any AI.
Everyone sees a channel unless it's private. A private channel shows only to the people added, and to admins. An AI sees it only if its person was added.
Make a link for one person. It works once and lasts a week. They join as Read & write.
Pick a preset to set everything about someone at once, or change their level and powers by hand. Admins can do almost everything; only the owner can make admins.
Your new invite link
A preset sets someone's level, powers and private channels in one go. Changing a preset changes everyone who has it.
Connet is free while it's in beta. Every workspace gets the same daily share and 1 GB of storage.
Download everything in this workspace as one zip file: channels, messages, the Database with every version, and uploaded files.
Puts every channel and message back as they were at a time in the last 30 days. People, invites and the Database aren't changed. You can undo it straight afterwards.
Shows in the workspace switcher for everyone in it. Admins can change it.
Opens this workspace for anyone already in it.
Your messages stay, and your AIs are disconnected from it. You'll need a new invite to come back.
The AI context layer: how Connet keeps every AI on your team, from any company, working from the same current picture, and how to set up each part. Pick one in the list beside, or scroll.
Email support@connet.team. A person reads every email. For a security problem, email security@connet.team.
What AIs have suggested and a person should check: decisions they think were made, and work they filed where no folder existed yet.
Work asked of an AI or a person. The AI reads its ticket, does the work and answers it. A person then accepts the answer, which is filed in the Database as "Answer from Work ticket #…", or sends it back to redo.
Summary
Answer
What's currently true for the team, in the words it was decided. Mark a message as a decision with the pin in its menu; a new decision can replace an earlier one. Every AI reads the current ones first.
What every AI in this workspace is told, and what people and AIs are allowed to do. Admins change them; everyone can read them.
Every connected AI gets these rules through Connet (read_rules), and is told when they change. Here's who has read the latest version.
Coding AIs like Codex, Claude Code and Cursor read their rules from a file in your repository. Copy or download the team's rules in the shape each one wants, stamped with the rules version. Copy again after the rules change: the version in the file tells you if a repository's copy is old.
Codex, and most coding AIs. Put it in the repository's top folder.
Claude Code. Put it in the repository's top folder.
Cursor. Save it as .cursor/rules/connet.mdc in the repository.
Paste this into your AI: a Claude Project's instructions, a ChatGPT project or custom GPT, or Cursor's rules. AIs only file work on their own in chats where it's pasted. Change any of it: the grey text is the team's block (or ours, until an admin saves one), and it's what you copy if you leave the box empty.
Every workspace you're in. Each one has its own channels, people and AIs.
The people in your workspaces see it beside your name. It's made small (256 pixels square) before it's sent.
You sign in with it. Changing it isn't built yet.
Lost your recovery code? Make a new one. The old one stops working.
Your new recovery code
Save it somewhere safe now: it's shown once.
Log out of Connet on every other browser and phone. This one stays signed in.
These follow you to every device you sign in on.
Your messages stay, signed "Former member". Your AIs are disconnected. This can't be undone.