White-label AI agents
Your client wants an AI agent. You quote it, we build it, they never hear our name.
We build AI agents and chatbots under your agency's brand - support agents, internal copilots, triage and research agents - in your repositories, on fixed scope and fixed price, with a mutual NDA and non-solicit signed before scoping.
Our cost is $4,500 to $14,000 depending on shape, against resale prices that typically leave you above 50%. Every build includes the parts that decide whether an agent survives production: a narrow capability boundary, confidence thresholds, escalation to a human, full action logging and an evaluation harness.
What sells, and what it costs you
| Agent | You quote | Our cost | Your margin | Timeline |
|---|---|---|---|---|
| Support agent on the client's documentation | $8,000-12,000 | $4,500-5,500 | ~50% | 3-4 weeks |
| Triage and routing agent | $10,000-15,000 | $4,500-7,000 | ~54% | 4-5 weeks |
| Internal copilot across 3-5 systems | $18,000-30,000 | $8,000-14,000 | ~54% | 6-10 weeks |
| Agent platform with its own admin | $30,000-60,000 | $14,000-26,000 | ~56% | 10-16 weeks |
Model and infrastructure spend sits on your client's own accounts at cost and is quoted separately - typically $200 to $800 a month for a support agent at real volume.
What is in every agent we build
- 01A narrow capability boundary, agreed in writing before any code - what the agent may do, what it must never do, what needs a human.
- 02Confidence thresholds and a clean escalation path, so low-confidence cases reach a person rather than being guessed at.
- 03An explicit refusal when retrieval finds nothing. An agent that always answers is one that sometimes invents.
- 04Full action logging, reproducible after the fact, which is what lets you answer a client's question about what happened.
- 05An evaluation harness scoring tool selection and outcomes, with a regression set of previously-fixed failures.
- 06A kill switch: a feature flag that stops the agent per account without a deploy.
The engineering detail behind each of these is on our AI agent development page.
What to tell your client, and what not to promise
The fastest way to lose money on an agent project is to sell autonomy the build cannot safely deliver. Three promises worth avoiding at the pitch:
- 01"It will handle everything."Sell a deflection rate instead - the share of cases it closes without a human. That number is measurable, defensible and improves over time.
- 02"It will never be wrong."Sell the escalation path. What matters to their operations lead is what happens on the 8% it cannot handle, not the 92% it can.
- 03"It will act on their systems from day one."Sell a shadow run first. The agent proposes, a human approves, and after two weeks of agreement it gets to act. Clients accept this readily and it removes most of the risk from your delivery.
Proof you can point at
The self-healing engineering agent - an agent that generates code, tests it in an isolated sandbox, reads the failure and patches the faulty block.
It is published under our name, so you cannot forward it as your own. What agencies usually do is quote the numbers as their partner's track record, or ask us for an anonymised technical summary they can attach to a proposal - we will write one.
How the engagement runs
- 01The processNDA first, scoping call within two working days, fixed quote within three, then delivery in your repositories with a weekly update in your voice. The full process is on how a white-label engagement works.
- 02Need search instead?If your client's need is search over their documents rather than an agent that acts, see white-label RAG.
Frequently asked questions
Support agents that answer from the client's own documentation and escalate cleanly, internal copilots over their systems, triage agents that route and enrich inbound work, and research agents that gather and summarise. The shapes that sell well are the ones where the client can already name the hours being spent.
$8,000 to $12,000 for a support agent with a knowledge base and one or two integrations, $18,000 to $30,000 for an internal copilot across several systems. Our side is $4,500 to $5,500 and $8,000 to $14,000 respectively, which leaves most agencies a margin above 50%.
The design handles most of it: confidence thresholds, an explicit refusal path and escalation to a human. For the rest, ongoing support can be retained. Your client escalates to you, and you escalate to us - they never see the second hop.
Yes, and it is usually the better part of the deal. Agents need prompt updates, occasional model migrations and eval maintenance. A retainer at your own rate against our monthly cost is a cleaner margin than the build.
It is much better if they do. Their account, their bill, at cost. Reselling API credit with a markup means a busy month becomes your problem, and it gives you an incentive structure your client would not like if they understood it.
Quoting an agent this week?
Send the brief. Fixed number within three working days.