Private pilot · in development

The control layer for AI agents.

Define what each agent can do, approve what matters, and keep a record of everything. Built for agencies running AI agents for several client businesses.

We're working with a small number of agencies to shape the first version.

Why

Agents have stopped just answering. They're acting.

Agents now buy software, pay vendors, issue refunds and change customer records. When you run them for several clients, the hard questions are about control: what is each agent allowed to do, for which client, up to what amount, and who signs off when it matters? And when something goes wrong, can you show what happened?

How it works

Every consequential action gets one of three answers, with a reason.

This is the approach we're building and testing with pilot agencies.

01

The agent asks first

Before a protected action, the agent tells Fisher what it wants to do, through an API or MCP tool.

02

Explicit rules decide

Allowed, denied, or needs approval. Rules are written and testable; no AI model makes the allow/deny call, and anything not permitted is denied.

03

The right person approves

Risky actions go to the approver for that client, who sees what the agent wants to do and why.

04

Everything is recorded

Each request, decision and approval is logged, so you can show a client exactly what their agents did.

Buy software, $40Dental clinic · ops agent · within its limit Allowed
Pay a new vendor, $2,400Supply company · payables agent · unknown vendor over $500 Needs approval
Delete a customer recordSupply company · payables agent · never permitted Denied

Illustrative examples, not customer data.

For agencies

One place to control every agent, for every client.

  • Separate permissions and spending limits for each client and each agent
  • Approvals routed to each client's own approver
  • Pause or revoke any agent instantly
  • A per-client record you can share with the client

Our aim: give agents more useful autonomy without giving them more authority than they should have, and save you from building and maintaining this control layer yourself.

What it controls

Clear about what Fisher can and can't control.

Fisher only controls actions that go through it. If an agent has its own direct access to a system and doesn't ask Fisher first, Fisher can't stop that action. Part of working with us is deciding together which actions to route through Fisher.

Fisher is in a private pilot and isn't generally available. It does not move money or issue cards today.

Running AI agents for clients? Let's talk.

We're looking for a few agencies to be design partners: tell us how you control your agents today, and help shape what we build.

Email hello@fisherpay.ca