AgentWorks is the delivery platform for AI agencies and consultants. Build an agent in your own workspace, hand it over into your client’s, keep it running, and get paid for it every month — instead of shipping a one-off project and losing the thread.
01 — The delivery loop
Most platforms let you build an agent. The hard part of agency work is everything after that: handing it over without handing over control, keeping the client’s credentials out of your hands, and still being able to improve the thing next month. That loop is what AgentWorks is built around.
Compose the agent the way you always do — instructions, tools, knowledge, guardrails. Start from one of the 50+ templates or from a blank agent, and iterate in chat until it behaves.
Connections live per workspace, so you test against your own mailbox, drive and test CRM. Nothing you connect while building ever reaches into a client’s systems.
Hand the agent over as a managed agent: it runs inside the client’s own isolated workspace, but you stay its maintainer. The client uses it; only you can change it.
They authorise their own Google, Microsoft, or CRM accounts in their own workspace. You never hold their production credentials — which is usually the moment their IT lead relaxes.
Models improve, processes change, edge cases show up. You tune instructions, swap models, add tools — and that ongoing care is what turns a delivered agent into a monthly retainer.
02 — Hand-over
Two ways to put an agent in a client’s hands. The one you pick decides who can change it afterwards — so pick it deliberately.
The agent lives in the client’s workspace but stays yours to change. Ideal for bespoke work and for anything you support under a monthly agreement.
Push an agent, workflow, skill or rule to many client workspaces at once. Each client gets an independent copy they are free to adapt.
03 — Day to day
Agency console
Each client gets a separate, isolated workspace — and you get a single console across all of them: who is using what, what it costs, and which agents are actually earning their keep.
Agent library
Start a client engagement from a working agent instead of a blank page: templates across support, sales, marketing, finance, HR and operations, all included for every client workspace you open.
04 — Getting paid
Some clients want AI spend on their own books. Others want one predictable number from you. You do not have to pick one model for your whole portfolio.
Usage draws on the client’s own balance and you invoice for your work. The cleanest split when the client wants their AI spend on their own books.
The agent is free at the client’s end and its usage draws on your agency balance. This is what makes a flat monthly “agent as a service” price possible.
Scheduled and automated runs can draw on a dedicated automations budget, so an overnight job can never quietly eat the balance the client’s team uses during the day.
05 — Partner pricing
Before you book a call, here is the shape of it — so you can work out whether the model fits your practice without having to ask.
You run on an ordinary AgentWorks plan, the same as any other customer. On top of it sits a separate agency balance — the one that pays for whatever you choose to sponsor on behalf of clients.
A client workspace is billed to that client, on the plan that fits them. Their subscription is theirs and independent of yours, so nothing about your agreement with us is entangled with theirs.
Charge it to the client’s own balance, sponsor it from your agency balance so the agent is free at their end, or point it at a separate automations budget for unattended runs.
Whichever balance is paying is charged at our published per-model rate for the model that ran — the same rate card as every other customer, visible per run.
Workflow automation is counted in runs, not tokens: every plan includes a monthly run allowance, and extra runs are bought in packages when a workload outgrows it. A step that also calls a model pays the published model rate on top.
You set your own price per agent and invoice your client yourself. The platform can record that price next to what the agent actually cost you, per client, per month. The partner program includes revenue sharing on top of that — so growing your client base grows what we both earn.
Partner terms — including your revenue-sharing percentage — are agreed on the qualification call, once we both know what your delivery looks like. See the per-model rates.
06 — The upside
Put your logo and accent colour on the workspaces your clients log into, so the platform reads as part of your practice rather than a tool you resold.
Maintenance, sponsored usage and per-client pricing are first-class, so the work you do after go-live is something you can bill for instead of absorb. The partner program also shares in that revenue with you — details on the qualification call.
Every client’s data stays in the EU, every agent run is logged, risky steps can require a human approval, and each agent carries a risk classification — the artefacts your client’s DPO asks for.
Playbooks, office hours and a direct line to the product team — so you are not working out delivery, pricing and governance on your own.
07 — The program
We keep the program small enough to stay useful. Partners we work best with tend to recognise themselves here.
08 — Apply
Tell us what you build and who you build it for. We review every application and reply within a few business days.
Send us your agency profile and a sense of the clients you work with.
A short call on fit: what you deliver today, where agents would land, and what your clients need on compliance.
We set up your agency workspace and branding, and walk your team through the delivery loop on a real first agent.
Your first client workspaces open, with support channels in place and a plan for the next ones.
09 — Questions
The client relationship is yours. You hold the contract, set your own prices, and are the first line of support. We work with you, not around you — that is the point of the program.
They can use it and see what it does. A managed agent stays read-only for the client, so nobody accidentally rewrites the instructions you tuned. If a client should be able to adapt it, distribute it as a template copy instead — then it is fully theirs to edit.
No. You build and test against your own connected accounts in your own workspace. When the agent is delivered, the client authorises their own systems in their own workspace. Connections never cross the boundary between workspaces.
You choose, per agent. Usage can be charged to the client’s own balance, sponsored from your agency balance so the agent is free at their end, or drawn from a separate automations budget for unattended runs.
Their workspace and its data are theirs. A managed agent you maintain can be unmanaged and handed over, or removed — and because each client is an isolated workspace, offboarding one never touches the others.
No infrastructure to run and nothing to host. Agents are built in the browser; a developer on your team can go further with the CLI and the agent SDK, but it is not a prerequisite for delivering client work.
Agency Partner Program — onboarded via a short call
Try the platform on a free account to see how an agent is built, or apply to the program and start delivering into client workspaces.
