Know what every client costs
Group each client’s projects under the client. Every request through the gateway is logged and priced against a versioned rate table from the provider’s own token counts — not estimated, not sampled — and rolls up per project and per client: requests, spend, and savings, month to date.
Cap each client
Give a client a monthly spend ceiling across all of its projects. The gateway checks it before calling the provider and refuses the request once it’s reached, so a looping agent at one client can’t turn into a surprise on your bill. Per-project limits on rate, prompt size and output size still apply on top.
Cut the bill
- Exact cache. Identical deterministic requests — temperature 0, no tools — are answered from cache. A hit costs nothing.
- Model routing. Small, simple, low-temperature calls on a premium model can move to the same provider’s value model (gpt-4o → gpt-4o-mini, claude-sonnet-5 → claude-haiku-4-5, gemini-2.5-pro → gemini-2.5-flash). Features named legal, medical, financial or security are never moved.
- Replace with code. Calls whose answers follow a rule are found in recorded traffic, and ModelGate writes the code — shown only once it matches at least 95% of 30+ answers it never saw.
Changes in your code, in one click. Connect a client’s GitHub repository and each suggestion points at the exact file and line; model and output-limit changes and code replacements open as a pull request on a new branch for you to review. ModelGate reads only the repositories you choose and never stores the source.
Everything that changes a request is off by default and switched on per project. See what the audit finds.
Prove it with a report
Each client gets a monthly report: AI spend, what the same traffic would have cost without ModelGate, what was saved (and how much of that came from cache), broken down by project and by model. Print it to PDF or download the CSV. It carries your agency’s name, logo and colour — and not ModelGate’s fee, so you set your own price. Read what to put in a client AI report.
Bill it your way
- You pay, itemized. One monthly invoice with the 30% success fee listed per client, so you know what each client’s savings cost you before you rebill.
- The client pays. Send a billing link: the client approves their own PayPal subscription ( $29/month + 30% of their savings) and gets their own invoice. You never see their card, and they stop using one of your seats.
Your team in, your clients read-only
Invite staff as admins (everything except payments) or members (project work). Give a client a login that shows only their own projects, usage, savings and report — read-only, under your brand.
Setup, start to first number
- Sign up and switch on agency mode.
- Add a client and create its project. Store the provider key on the project: the client’s own, or yours.
- Point the client’s app at the gateway: one base URL and a ModelGate key — or connect the client’s GitHub repository and merge the one pull request ModelGate opens for it.
- Watch spend and waste per client from the first request. Start the 14-day free trial to turn the automatic savings on.
What it costs
$199/month with 5 clients included, then $29 per extra client you pay for, plus 30% of realized savings— savings the gateway measured itself on cache hits and automatic downgrades, and on changes you make from our suggestions (for 3 months after each change). Nothing saved, nothing beyond the base.
| Example | Seats you pay for | Base per month |
|---|---|---|
| 3 clients, you pay for all | 3 (0 extra) | $199 |
| 8 clients, you pay for all | 8 (3 extra) | $286 = $199 + 3 × $29 |
| 8 clients, 2 pay ModelGate directly | 6 (1 extra) | $228 = $199 + 1 × $29 |
Plus 30% of the realized savings on the clients you pay for. Clients paying directly cover their own $29/month and 30%. Full detail on the pricing page.
Questions agencies ask
Who pays OpenAI, Anthropic or Google?
Whoever owns the provider key on the project. Each project stores its own provider keys, encrypted — so a client's project can run on the client's own provider account (the provider bills them directly) or on yours. ModelGate never resells model usage.
Do my clients see what ModelGate costs me?
Not on the savings report — it shows their AI spend, what it would have cost without ModelGate, and what was saved, so you can bill them at your own price. Only a client you set to pay ModelGate directly sees ModelGate's fee, on their own invoice.
What does it need from the client's app?
One change: point the OpenAI-compatible base URL at the gateway and use a ModelGate key for that project. The existing OpenAI SDK keeps working. Or connect the repository on GitHub: ModelGate finds every OpenAI client and opens that change as one pull request, with an app key to paste into the environment.
Can ModelGate see our clients' code?
Only if you connect GitHub, and only the repositories you choose. The code is read in memory to find where AI calls are made and isn't stored — ModelGate keeps file paths, line numbers and model names. It opens pull requests only when you click, on a new branch, and never pushes to the main branch or merges. One repository per project.
What does it cost?
Every plan starts with a 14-day free trial when you add a card, and you can cancel anytime. After that the agency plan is $199/month with 5 clients included and $29/month per extra client, plus 30% of the savings the gateway actually realized, plus 30% of the savings from changes you make from our suggestions for 3 months after each change.
What happens when a client leaves?
Archive the client. Its projects stop serving traffic and it stops counting as a seat from the next month; its history and reports stay available, and you can restore it.
Are my clients' prompts stored?
Not by default. Request bodies are stored only if you turn it on per project, for richer recommendations; metadata, token counts and costs are always kept.
Guides for agencies
- How to price AI work for clients when model costs move every month — Retainer, cost-plus or a share of savings — and the token math that decides your margin.
- Passing LLM API costs through to clients: three models that work — The client's own provider account, rebilling with a markup, or a bundled fee — and what each needs.
- What to put in a monthly AI cost report for a client — The seven numbers a client actually reads, and how to get them without a spreadsheet.