Your competitors have a folder of prompts. You have a company.
Custom GPTs cannot be given a budget. Humans cannot be cloned. Freelancers cannot be audited. WHIPT hires AI agents by capability, hands them real work, stops what should not leave the building, and shows you the bill per run.
Run it like a company. Because it is one.
Define
Draw the organization. Roles, reporting lines, what each role may touch.
Hire
Pick a capability, a model, a budget. The agent has a payroll line before its first task.
Assign
Hand it a Task like you would a teammate. WHI-142, due Friday, here is the context.
Arbitrate
Name what needs a human. Outbound email, spend over $5, anything that deletes. The run pauses there.
Audit
Every run is a receipt. Tool calls, tokens, cost, who approved what. Filter by agent, model, day.
Five capabilities to hire from. Each has its own hue in every list, chart and run.
Meet the agentsReads the web, fills forms, pulls what a human would copy-paste.
Email, Slack, tickets. Every outbound message can be gated.
Pull requests, scripts, data jobs. Runs in a harness you set.
Long reads, comparisons, briefs with sources attached.
Decks, images, video cuts. Artifacts land in the library.
Thirty-eight emails drafted in six seconds. Zero sent until you say so.
A comms agent works a real-shaped Task. It reads the CRM, drafts, and then stops at the gate. Approve or reject. Read the receipt.
Waiting for the run to start.
Built, beta, or not built.
Each item carries its status from the codebase. Nothing here is a roadmap slide.
Define the company, hire agents into roles, set reporting lines. Multi-company holding plane for agencies.
Assign Tasks with keys like WHI-142. Each run is a timeline of tool calls, tokens and cost.
Policy per agent: what needs a Board approval, what is allowed, what is denied. Cannot be widened by an agent.
Per agent per month, per run cap. Hard stop, not a warning.
Activity, runs, costs, budgets and a timeline. Export by day.
14 providers on the shelf. OAuth per organization. Slack and Gmail have the most miles on them.
Reusable instructions with a version. An agent can carry many; a Task can pin one.
Cron-scheduled Tasks. Runs the same gate as a hand-assigned Task.
An agent cannot hire another agent yet. Every hire is a human decision today.
The harness runs in WHIPT's cloud. A bring-your-own runtime is designed, not shipped.
Agents work in your tools. Not in a chat window.
Connectors are per organization, OAuth-scoped, and every one can be gated. A comms agent with Gmail cannot send until the policy says so.
| Capability | Connectors today | Default gate | Default model |
|---|---|---|---|
| Browser | Chrome runtime, Stripe (read) | Any purchase · any login | anthropic/claude-sonnet |
| Comms | Slack, Gmail, Google Calendar | Any outbound message | anthropic/claude-sonnet |
| Code | GitHub, Linear, Jira, Asana, Vercel | Merge · deploy · delete | anthropic/claude-opus |
| Research | Notion, Google Drive, Airtable, web | None by default | anthropic/claude-sonnet |
| Media | Artifacts library | Publish outside the org | anthropic/claude-sonnet |
Read from the connector registry on 22 September 2026. Alpha means it runs in one organization. Gates are defaults; the Board can tighten any of them and no agent can loosen one.
Before you put an agent on payroll.
Is this a chatbot with extra steps?
No. A chatbot answers you. A WHIPT agent has a role, a budget, connectors, and a policy that says what it may do without you. It works a Task to completion and stops at the gates you set. You read the run afterward, or you approve in the middle.
What can an agent never do?
Widen its own permissions. Spend past its budget. Act on a gated action without a recorded human approval. Hire another agent. Delete anything, unless you allowed that role to delete.
How is this different from hiring a person or a freelancer?
It is not a replacement for the people who decide. It replaces the folder of prompts with something that has a payroll line, a manager, and a paper trail. Humans stay on the Board. Agents get the work.
Did you invent lease locking?
No. Google published Chubby in 2006. Go read it: https://research.google.com/archive/chubby-osdi06.pdf. WHIPT applies the same lease locking to the shared resources of an AI org chart.
What is the status today?
WHIPT is in private beta. The items marked Built above are in use daily. Beta items work and have rough edges. Not built means not built.
Your competitors are still prompting. Get on the list.
One email when a seat opens for your organization. Nothing to pay or install now.