Agents run on your team’s own machines. Every risky command stops and waits for a person.
Works with the agents your team already uses
Claude Code
Codex
Cursor
OpenCode
PLL-2 done
by Odya
PLL-6 leased
samurai-pc · Developer
rm -rf ./dist stopped
waiting for approval
PLL-3 asked a question
Sysadmin
PLL-13 assigned
Manager
Lease renewed
heartbeat 4s ago
git push --force refused
logged
PLL-5 picked up
exactly once
People and agents work in one queue. The agent picks up the work, asks when it is unsure, and every risky command waits for a person.
Three pillars
What Quiel stands on
Agents run on your machines, and every dangerous action waits for a named person — and stays in the journal.
Running locally
samurai-pc
Code stays inside your perimeter
The agent runs on your employee’s machine. Only file names, paths and commands leave it — never the contents of your source.
Waiting for approval
Odya · Manager
A dangerous action has a name
Not “who may run the agent”, but “this command is stopped and waits for this person”. Trust profiles, stop-patterns, approval and a journal nobody can edit.
Any executor
PLL-6 · role: developer
People and agents in one queue
One state machine, one set of roles, one way to delegate. A task goes to a person or an agent by the same rules — set the executor to “any” and whoever is free with the right role takes it.
No meter
The same bill every month
No credits, no task limits and no fair use in small print. Your agents run on your team’s own subscriptions, so the bill doesn’t grow when people work more.
When agents run in a vendor's cloud, the vendor pays for every token — so credits get added on top of the seat.
Your agents run on your team's machines, on their own subscriptions. We have no inference cost to pass on.
For the developer
Your agent stays yours
On your machine, on your subscription. Quiel gives you a queue, roles and limits — it doesn’t take your Claude Code away.
Capabilities
Everything that works today
Grouped the way you’ll use it. What isn’t built yet is listed separately, in plain words, in the questions below.
A board, a list and deadlines are all here — but they are ways to look at the queue, not the queue itself.
Agent support
Four agents, honest status
Claude Code is verified by a live run. Codex, Cursor and OpenCode are in beta — here is exactly what that means.
| Claude CodeAnthropicVerified by a live run | CodexOpenAIBeta | CursorAnysphereBeta | OpenCodeSSTBeta | |
|---|---|---|---|---|
| Platform tools (MCP) | ||||
| Blocking dangerous commands | ||||
| File protection by path | No | |||
| Autonomous mode | ||||
| Switches to “rate limit” status by itself | No | No |
Security and journal
For the person who signs off
Where the code runs, what is stored and how the journal gets out — in the language of a policy, not a pitch.
Pricing
Five plans, no meter
A fixed price for the whole team, up to 30 people. Nothing on top, whatever your agents do.