How it works
Two ways in.
One workspace.
Companies come to hire. Professionals come to work. Everything after that happens in the same place.
The two paths
Four steps, either way.
If you're hiring
01Create a workspace
Pick the kind of work — support, sales, assistants, or projects. It arrives configured.
02Find people
Browse profiles by skill, rate, and availability. Post a role if you'd rather they come to you.
03Assign the work
Tasks, timesheets, and documents for everyone you hire, in one board.
04Approve what matters
Agents draft the routine work. You sign off on anything with risk attached.
If you're looking for work
01Build one profile
Skills, experience, portfolio, rate. It goes live to companies at 70% complete.
02Apply to roles
Send proposals to open jobs, or get found by companies browsing the bench.
03Agree the terms
Contracts and milestones are written down before the work starts.
04Get paid
Log time, hit milestones, get paid through the platform. Free to join — VcrewX keeps 10% of what you earn, 5% after a year with the same client.
The approval loop
Nothing risky runs by itself.
Every AI action takes the same path. Violet is the machine. Indigo is a person.
Worker agent
gpt-4o
Drafts the action. Never executes it.
Manager agent
claude-sonnet-4
Scores risk and confidence on every draft.
The threshold
90%
Below it, or any real risk, and a person decides.
In practice
One card. One decision.
When something stops for review, it arrives as a single card: what changed, how sure the agent was, why it stopped, and two buttons. No dashboards to interpret.
- The change shown as before → after, not a paragraph
- A confidence number you can actually see against the threshold
- The reason it stopped, in a sentence
- Approve or reject — and the whole thing is logged
Refund request — ticket #4821
agentWorker · gpt-4o · 14 minutes ago
Status: Open · Assigned: unassigned · Refund: —
Status: Resolved · Assigned: you · Refund: $48.00
Refunds move money, so this one needs your sign-off.
agentManager · claude-sonnet-4