Work as a team
Connect your AI assistant
Connect the AI assistant in your editor with a key of your own, let it read and draft, and review its drafts before anything changes.
Open AI assistants in the appWhat the assistant does
Claude Code, Claude Desktop, VS Code, Cursor or any tool that speaks HTTP can work in Rabbinatix. Your assistant reads your systems, APIs, logic, flows and whiteboards, checks what a change would break, runs tests, and drafts new work. Everything it drafts waits for you to review. Nothing changes until you accept it, and what you accept is saved as your own change.
There is one URL, /ai. It serves a guide that tells the assistant everything it needs, and it is also an MCP server. The full guide, as your assistant reads it, is on Your AI assistant.
Connect your assistant
- Open Integrations → AI assistants and click Connect your assistant.
- Pick your tool.
- Under It can reach, choose The whole organization or Only what I choose, then pick systems, flows, logic or whiteboards.
- Under It can, choose Read and draft changes for me to review or Only read. Tick Also draft deletions only if you need it. Tick Report findings when your assistant should scan your code for security issues and report what it finds (see Security checks).
- Under It expires after, pick a time. The default is 1 hour, and the longest is 30 days.
- Click Make the key.
- Copy the snippet into your tool. Your key is inside it, and it is shown only now.
The dialog shows Connected when your assistant makes its first call. It then suggests a few things to ask.
Keys and their limits
Every key belongs to one person and does at most what that person may do now. If your role changes, your key changes with it. A viewer's key only reads.
- Your keys lists each key, what it reaches, when it expires and its Activity, the log of its calls.
- Change narrows or widens a key. Revoke stops it at once: an assistant or pipeline still using it is told it was revoked. Revoke all stops every key you have. A revoked or expired key offers Remove, which takes it off the list with its activity.
- People who manage integrations see Everyone’s active keys and can revoke them.
- A key makes at most 60 calls a minute. The whole workspace makes up to its plan's assistant calls a day (100 on Free, 100 on Standard, 200 on Premium, 300 on Enterprise); owners and admins see how many are used today in Usage and billing → Plan and usage.
- People who manage the organization's settings can switch assistants off for everyone in Settings → General → AI assistants.
- The AI assistant in chat is separate: it uses your own Claude, ChatGPT or Gemini key (connected under Use your own AI in chat on the same tab) and is off until someone who manages the organization turns on Settings → General → AI assistant in chat. It's on the Standard plan and up. It can also call other apps you connect, like Jira, under Other apps for your AI: see Chat.
Review drafts
Drafts arrive in batches on Changes, in its From your AI assistant view. The sidebar's count includes the drafts waiting for you.
- Open a batch. Read The assistant’s note: and each change, with its diff against what is live, its warnings and the contracts it affects.
- Click Accept all, or select some changes and accept those.
- Or click Reject and give a reason under Why? (optional). Your assistant reads the reason the next time it asks about its drafts.
Undo takes an accept back for 30 seconds. If someone changed an object after your assistant read it, choose Re-apply on the latest or Use the draft anyway. A logic change can be opened with Open in the logic editor and adjusted there. Saving it accepts it.
Your batches are yours alone. Nobody else can read them, not even owners.
AI prompts
Pages where your assistant can help have an AI prompts button (in an editor, it is AI prompts… in the … menu). Pick a prompt and it is copied: paste it into your assistant. Each prompt is short, because your assistant reads the rest from Rabbinatix. Not connected yet: picking one opens Connect your assistant.
- Security checks: scan the code for security issues, apply the fixes marked Apply fix, fix the design’s security findings. A system’s Security tab has the same, for that system only.
- Systems: put your repo on the map, scan the code, answer a question. A system’s … menu: put its code on the map, scan it, bring it in step, explain it, fix its design’s findings.
- Logic and a logic: turn your code into logic, write its code, bring your code in step, change it, suggest improvements, explain it.
- Flows and a flow: turn a journey in your code into a flow, write its code, add test cases, change it, explain the journey.
- A ticket’s … menu: work on it, plan it, what it affects.
- Changes → From your AI assistant: redo the rejected or the outdated batches.
It builds your way
Your assistant has no memory of its own, so Rabbinatix tells it how your organization or project builds things on every call: each system's API prefix, how operations are named, which functions to reuse, and how flows are tested. Before it drafts, it can check its work. The check reports problems, contracts whose status would change, and hints such as a step that repeats an existing function or an API outside its system's prefix. Rabbinatix also refuses a second copy of something that already exists.
Suggestions
Your assistant can ask for suggestions on a logic, flow, system or API. These come from fixed rules, not a model, so the app and the assistant always agree. Examples are a logic with problems, a contract that doesn't match, a flow with no test cases, an API that nothing uses, and an API without logic. Where a fix exists, the assistant can draft it for you to review.
What it can't do
- Change anything live. It only drafts.
- See secret values, keys, members, roles, settings or integrations.
- Reach the network from a test. Tests are test runs on your test data.
- Go past its key. Every call is checked and logged.
Read more on Your AI assistant.
Related
- AI assistant referenceConnect the assistant in your editor, review its drafts, and the guide it reads.
- IntegrationsSend contract and ticket events to your own endpoint or an email address with webhooks, and see each delivery and its retries.
- LogicBuild what a system does by connecting nodes on a canvas: endpoint logic for an API, event logic for a user action, and functions you reuse.
- FlowsBuild a flow from the logic it starts at, see it cross your systems in lanes, and move it from draft to verified.
- Flow Pseudocode (FPC)Text steps, the plain-text form logic reads as: reference, cheat sheet, grammar and a validator.
