1. Say what you want
Describe the job in Slack or web chat.
How a message becomes finished work, and who checks it on the way.
Request accessYou describe what you want. Hinzel's own assistant writes the rules, checks them and asks for approval.
Describe the job in Slack or web chat.
It writes the rule, tests it and replays past events.
An admin reviews the change. Then it goes live.
Begin from triage, root-cause analysis, code generation or a daily digest, or from a blank project.
Each edit is validated as it is made. Errors name the exact field that is wrong.
Behaviour tests run on every draft, and a replay shows what the rule would have done last week.
Every trigger comes with a description of what it does, generated from the rule itself.
A project admin approves each change in Slack or the console. Open work finishes on the old version.
Every reply traces back to the event that caused it.
A Slack mention, a Jira issue, a webhook or a schedule. Each event is stored once, even when it arrives twice.
The project's rules decide whether to start a run, continue one, or do nothing.
It gets its own workspace and stays on the rules that were live when it started.
Every tool call goes through the gateway, where it is checked and recorded.
It lands in the same thread or ticket, as the bot or as the person who asked.
A new message in the thread wakes the same run. Idle runs sleep and wake on the next message.
Rules decide when agents start and what they may do. They run the same way every time.
Follow a ticket through its states, with timers in business hours.
Typed data, with a history of who changed what.
Conditions on typed event fields, checked before anything starts.
A rule can post a message, call an action or update a record on its own.
Business hours, time zones and holidays for every timer and schedule.
"Ping me when this PR merges" sets a watch from inside a conversation.
Agents hold no credentials. They reach your tools only through the gateway.
Each project's rules decide which actions run, which need a person and which never run.
Actions run as the bot, or as the person who asked with their own connected account.
Spend limits per project stop work before it goes over.
Each call is stored with its input, its result and the rule that allowed it.
Each run gets its own place to work, and you can see everything it does.
Code runs work on a fresh branch of your repo and open pull requests from it.
Each run is a Claude Code session. Pick the model and pin the plugins per project.
Follow every step on the run page, and stop a turn from the console or Slack.
An agent can stop and ask a question, then continue when someone answers.
Use the built-in connectors, add any MCP server, or send events in with a webhook.
Events come in and actions go out, with replies in the same thread.
Add its URL. Its tools pass the same checks as every other call.
A signed webhook starts a run. Outbound webhooks report the result.