AI API Reseller and Partner Program: Selling AI to Your Clients Under Your Own Brand and Margin

An AI API reseller program lets an agency or software house sell AI model access to its own clients under its own brand and its own profit margin. In the Onysoft Partner Program you get a separate partner portal, open a dedicated API key per client, set your own sale price, and manage billing with client-level usage reports; applications go through /partner/apply.

Almost every team that delivers chatbots, document analysis, or content automation to clients hits the same question: who carries the AI cost? Having each client open a foreign account in their own name is often impractical; manually splitting one agency account across clients does not scale. The reseller model unties this knot: the infrastructure stays on the platform, while the client relationship and the price stay with you.

This article covers the real partner portal mechanics (per-client keys, margin freedom, client-level reports), an illustrative revenue calculation built on live catalog prices, the controls that cap your risk, and — what most promotional pages skip — an honest look at who this model is not for.

What Is AI API Reselling, and How Does It Differ from Affiliate Referrals?

AI API reselling means buying model usage at catalog price and selling it to your own clients at a price you set; the core difference from an affiliate arrangement is that the client belongs to you, not to the platform. In an affiliate model you refer the client, collect a fixed commission, and step out of the relationship; in a reseller model you set the price, issue the invoice, and provide the support — and the entire margin stays with you.

The natural candidates are teams already shipping AI-powered work to clients: software houses (chat, summarization, and classification endpoints inside delivered products), digital agencies (content and customer-service automation), vertical SaaS builders (per-tenant AI consumption in a multi-tenant product), and IT consultants. The common thread: the end client cares about a working solution, not an API key — they expect you to own keys, catalog, and cost management.

In Türkiye, teams facing this need have been stuck between two half-solutions: opening a separate foreign account per client (payment and accounting friction repeats with every client) or sharing one account across everyone (usage blurs together, no limits, no invoice separation). A partner portal is the third path that fills exactly this gap.

How Does the Partner Portal Work? Per-Client Keys, Your Own Margin, Client-Level Reports

In the Onysoft Partner Program you get a partner portal separate from the standard user dashboard, and you create your clients there yourself — each client gets its own project and its own API key. The mechanics have four parts:

• A dedicated key and project per client: Each client's usage flows in isolation inside its own project. One client's traffic never blurs into another's report; if a key leaks, you rotate only that client's key.
• You set the profit margin: The platform sells to you at catalog price; what price or package you offer your client is entirely your commercial decision. Usage-based pass-through, a flat monthly package, or bundling into a project fee — all three work.
• One balance, all clients: Your clients' usage is deducted from your partner balance. You do not build per-client payment infrastructure; you collect through your own invoicing.
• Client-level usage reports: The partner panel shows each client's consumption separately, and also includes Chat and documentation.

Behind the keys sits the full catalog: 739+ models from more than 60 providers behind one OpenAI-compatible API. Starting a client on an economy model today and moving to a flagship tomorrow is a one-line model-name change in the code you deliver.

How Do You Build the Revenue Model? An Illustrative Calculation

The core of the revenue model is simple: you buy usage at catalog price and sell to your client at your own price; the difference is your gross margin. The table below prices three typical client scenarios with live catalog prices; the 30% rate is a purely illustrative value chosen to make the math concrete — the partner sets the margin, and this is not an earnings promise:

Client scenario	Model	Monthly usage	Your cost (USD)	Your cost (TRY)	Client invoice at illustrative 30% margin (TRY)
Support chatbot	openai/gpt-5.6-luna	30M input + 6M output	$9.90	₺470.79	₺612.03
Contract/document analysis	anthropic/claude-sonnet-5	10M input + 2M output	$60.00	₺2,853.30	₺3,709.29
Content generation tool	google/gemini-3.6-flash	8M input + 4M output	$63.00	₺2,995.97	₺3,894.76
Total (3 clients)	—	—	$132.90	₺6,320.06	₺8,216.08

Measured: August 5, 2026 — api.onysoft.com live catalog. TRY equivalents calculated at the August 5, 2026 TCMB rate (1 USD = 47.555 TRY). The 30% margin is illustrative; real volumes and margins depend entirely on your scenario.

Two mechanics make this calculation reliable. First, every API response returns that request's actual cost (the cost field, in USD); you base client billing on per-request measured cost, not estimates. Second, the accounting side is local: your expense is the TRY e-invoice Onysoft issues at the TCMB rate, and your own company invoices your client. The reverse-charge VAT and withholding obligations that come with foreign AI subscriptions do not arise on a local invoice — a meaningful simplification for agency bookkeeping. (This section is not legal or tax advice; consult your accountant for your own structure.)

How Do You Control the Risk? Pre-Checks, Key Limits, and Auto-Stop

The main risk of the reseller model is obvious: because client usage draws down your balance, an uncontrolled client could in theory burn through your budget. The platform caps this risk with three technical mechanisms:

• Pre-request balance check: Before every request, the estimated cost — with a 1.2 buffer — is checked against your balance; if it does not cover the request, a 402 is returned before the request ever reaches a provider. Going negative or getting a "surprise invoice" is impossible in this architecture.
• Per-key token limits: You can define per-model token limits on each client key. If you sold a flat-price package, you enforce its ceiling with a technical limit rather than a contract clause — overflow usage is stopped before the request is even made.
• Auto-stop at zero balance: When the balance runs out, the system stops; after a top-up it resumes where it left off.

There is a lever on the margin side too: set the model field to onysoft/auto and OnyRouter routes cheap jobs to cheap models and heavy jobs to the flagship class; routing itself is free. Preventing simple questions from being processed at flagship prices is the single most effective move for protecting a reseller margin.

And when discussing service quality with your client, you hold measurements rather than marketing lines: across 6,959 successful requests through the gateway in the last 14 days, average end-to-end response time was 3.8 seconds, the fastest 0.3 seconds. Long-generation flagship models pull the average up; with lightweight models the time drops below a second.

How It Works on Onysoft: From Application to the First Client Key

The process has four steps, and your first client key is a one-line change away in code:

1. Apply: Submit your company/agency details at /partner/apply.
2. Portal and balance: After approval you sign in to your separate partner portal and top up a TRY balance. Enter your tax number (VKN) in the billing details and click "Query from GİB" — your registered company title auto-fills from the live GİB registry, and an e-fatura or e-arşiv invoice is issued automatically according to your taxpayer status.
3. Create clients: In the portal you open a separate project and an sk-ony- prefixed key for each client, and define per-model token limits on the key if needed.
4. Deliver: Place the key in your client's application or in the solution you build. The endpoint follows the OpenAI schema exactly:

from openai import OpenAI

# Each client uses that client's own key
client = OpenAI(
    base_url="https://api.onysoft.com/v1",
    api_key="sk-ony-CLIENT_A_KEY",
)

response = client.chat.completions.create(
    model="openai/gpt-5.6-luna",  # or "onysoft/auto" to protect margin
    messages=[{"role": "user", "content": "Classify this support ticket."}],
)
print(response.choices[0].message.content)

What remains is monitoring: filter client-level reports in the partner panel by date, model, and status, then download them as PDF or send them as a logo-branded email report directly to your client — no extra spreadsheet needed for month-end reconciliation. Clients who want to try models without writing code can run demos in the in-panel Playground; technical details are in the documentation.

Who Is It For — and Who Is It Not For? An Honest Assessment

The partner program makes sense for teams that regularly deliver AI-powered solutions to clients and want to own the billing relationship; it is not for everyone. The honest picture:

Good fit:

• Software houses and agencies delivering chatbots, automation, or AI-powered products to multiple clients — per-client keys and reports become a natural part of the delivery.
• SaaS teams running multi-tenant products — per-tenant consumption isolation and limits arrive ready-made for the architecture.
• Consultants/MSPs managing clients' IT operations — AI access becomes one more line item on an existing service agreement.

Poor fit:

• A single project for a single client: you do not need the partner layer; a standard free account and one key are enough.
• You do not want the billing and support burden: reselling creates work; referring your client directly to the platform is the more honest solution.
• Data must never leave your own servers: that is a gateway-model discussion; we covered the framework in detail in the LLM gateway guide.

At the decision stage we recommend the same trust signals we would look for ourselves: prices published openly in the catalog, per-request cost breakdowns, and third-party validation. Onysoft has itself passed a partner network's vetting — we are a member of the Anthropic Claude Partner Network. If you want to test the model on your own account first, start with a free account and evaluate the catalog and Playground; when you are ready to resell, the application is at /partner/apply.

Last updated: August 5, 2026 · Data: api.onysoft.com live catalog