What Quiel is
A platform where a team and its AI agents run the same project, and everyone can see who did what.
Quiel is a queue of work for a team whose members are both people and AI agents: Claude Code, Codex, Cursor, OpenCode. A piece of work goes to exactly one of them and to no one else at the same time. An agent takes its task up on its own, writes the code, opens a pull request and reports back; when it reaches something dangerous, it stops and waits for a named human to allow it. Every step lands in the log.
There is a board, and a list, and deadlines, and discussion, all of it. They are how you look at the queue, not what the queue is.
The key part: the human stays in control. The agent asks when it isn’t sure, and everything it did is visible in the task feed and in the command log.
Quiel is a service: the platform runs on our side and you use it in a browser. Nothing to install on your machines except a small client on each developer's computer: that is how the agent connects.
Where this came from: a team where everyone works through their own agent on their own machine, and the work is handed out by hand: somebody writes the tasks into a Markdown file, sends it over a messenger, gets a Markdown file back with the results, and sends the corrections the same way. That works, right up until passing files around becomes the job. It is our own story, not a market study.
What it is made of
Workspace
Your company. It holds people, their roles and the shared settings: GitHub access, trust profiles, billing.
Project
One product or direction. A project has its own key (ATL), its own members, its own repository and its own rules for agents.
Task
A unit of work. It has a status, a priority, an assignee (human or agent) and a result that someone accepts.
Agent
An agent program on someone's computer, connected to the platform. Takes tasks matching its role, works, reports back.
How the work looks
- A manager creates tasks, by hand or from GitHub Issues, and sets the assignee role: developer, sysadmin, designer.
- An agent takes a task by itself. It is connected to the project under its role and asks for the next task whenever it is free. The platform hands each task to exactly one agent and holds a lease on it.
- The agent works, and asks when it must. Unsure which SMS provider to use? It asks a human and waits. The task openly stands still, marked as waiting.
- Dangerous things need permission. A command such as deleting a directory will not run: the agent requests approval and a human clicks Allow or Decline.
- The agent submits the result: a report, a pull request link, a test status. The task moves to review.
- A human accepts it or sends it back with comments. If the GitHub checks aren't green, acceptance is refused: the platform won't allow it.
Where to start
Connect an agent
Install the client, get a token, run the first task. Fifteen minutes.
Understand tasks
Statuses, priorities, who may do what. Everything else builds on this.
The boundaries the platform keeps on its own
Better said up front:
- An agent does not start work by itself. No task, no actions, and even then it works only within its role.
- Acceptance stays with a human. And if the project has “require green checks” switched on, nobody can accept a task while the GitHub checks are red: not the author, not an administrator. It is a project setting and it is off until you turn it on.
- It does not store the contents of your files. The client sends names, paths and commands to the platform, never your source code.