Reseller Playbook
Run your own AI API store — you source, you price, you collect
This is the how-to-sell companion to the Reseller Tutorial. The tutorial shows the mechanics — top up, create a Client Key, hand it off. This playbook shows how to turn “people who already use an API” into recurring revenue: who to find, how to price, how to close, and what to say.
1 · What you're actually doing
You're running your own AI API store. You top up your derouter.ai balance (your wholesale cost), mint Client Keys at a price you set, and hand them to customers. Customers call the API through apikey.cloud and never see derouter.ai. You keep the margin.
Three things make this a real business, not just a referral link:
- White-label. Your customer only ever touches
apikey.cloudplus the key you gave them. They can't see — or go around — your supply. - Your margin, your call. You set cost → client price. The ratio is locked at key creation, and renewals reapply it automatically, so you never miscalculate or accidentally reprice.
- Off-platform collection. How the customer pays you is entirely up to you (bank transfer, Alipay, crypto, invoice — anything). The platform never touches that money.
2 · Who to sell to
Go after people who already have an API need — don't educate from zero.
- Solo AI-tool developers — building bots, plugins, automations; want full-spec models without running their own infra. Find them on dev communities, GitHub issues, indie-hacker groups.
- Small overseas / startup teams — need multi-seat access, usage visibility, and a budget cap per project. Find them in startup and remote-work groups.
- Content & design studios — need
gpt-image-2for batch image generation and want to review logs. Find them in AI-art and design communities. - Users in restricted regions — need a stable, reachable endpoint without juggling foreign cards. Find them in student and remote-worker groups.
3 · How to price
Pick a markup, not a discount. You set your cost and the price the client sees; margin is (client price − your cost) / client price.
your cost client price multiplier margin $100 → $115 1.15x 13% thin, high-volume $100 → $130 1.30x 23% recommended starting point $100 → $150 1.50x 33% includes your onboarding + support The multiplier is locked when you create the key. Renewals reuse it automatically — no mental math, no accidental reprice.
Don't compete on raw unit price.What the customer is actually buying is “works out of the box + a stable endpoint + you answer questions in their language.” That's the pitch — not “cheapest per token.”
The exact create-a-key UI (cost field, client-price field, live margin preview) is walked through step by step in the Reseller Tutorial.
4 · Standard delivery flow
Your job isn't “send a key” — it's get the customer's first real request working.
- Top up your balance (start small, verify, then scale).
- Create a Client Key — set your cost + the price the client sees.
- Confirm the multiplier / margin, then create (it locks).
- Send the client their key + the Base URL (template below).
- Have them verify with a model list call first.
- Then one real call — chat or image generation.
- They self-check usage at apikey.cloud; when the balance runs low, you renew.
Hand-off template
Hi, Your API key: sk-ant-1234567890abcdef... Base URL (OpenAI SDK): https://api.apikey.cloud/openai/v1 For image gen / long requests, use the direct endpoint: https://api-direct.apikey.cloud/openai/v1 Check your usage anytime: https://apikey.cloud (sign in with email OTP, bind the key once)
Which Base URL to give
customer scenario Base URL timeout chat / text (<100s) https://api.apikey.cloud/openai/v1 100s image gen / long output https://api-direct.apikey.cloud/... 600s Anthropic-native (Claude Messages API) — swap /openai/v1 for /proxy/v1 Both domains are fully white-label: the customer never sees derouter.ai.
Customer integration example (OpenAI SDK)
from openai import OpenAI
client = OpenAI(
api_key="sk-ant-...", # the key you gave them
base_url="https://api.apikey.cloud/openai/v1",
)
# 1) verify the key + network
client.models.list()
# 2) a real chat call
r = client.chat.completions.create(
model="gpt-5.5",
messages=[{"role": "user", "content": "hello"}],
)
# image generation -> use the direct endpoint (avoids the 100s cap)
img_client = OpenAI(
api_key="sk-ant-...",
base_url="https://api-direct.apikey.cloud/openai/v1",
)
img = img_client.images.generate(model="gpt-image-2", prompt="a cat")Available models: Claude Opus 4.8 / Sonnet 4.6 / Haiku 4.5, GPT-5.4 / GPT-5.5, and gpt-image-2. Anything OpenAI-SDK-compatible usually just needs base_url + api_key changed. Full reference: API Integration Guide.
5 · Copy-paste scripts
30-second intro
I provide full-spec Claude & GPT API access. Change your Base URL to api.apikey.cloud and drop in the key I give you — you can call Claude Opus 4.8, GPT-5.5, and gpt-image-2 from one key. It's OpenAI-SDK compatible, so usually two lines of code, and you can watch your own usage and logs in the portal anytime.
Price pitch
Full-spec Claude/GPT, one key for every model. What you're getting is works-out-of-the-box + a stable endpoint + support in your language — not a pile of setup and pitfalls you have to debug yourself. Don't compare raw per-token price; compare what it costs you to NOT ship.
Tech onboarding
If your project already uses the OpenAI SDK, it's basically two lines: set base_url to https://api.apikey.cloud/openai/v1 and api_key to the key I gave you. Run models.list() to confirm it's live, then make a real request. Image gen and long outputs go to api-direct.apikey.cloud.
Troubleshooting
If a request fails, send me: the time, the model name, the endpoint, the status code, and whether it was streaming. For image gen or any long request, switch to https://api-direct.apikey.cloud/openai/v1 first (the default host caps at 100s and will 524).
6 · Handling objections
“Will you throttle / cut my quota?”
Quota is transparent — they see their own balance, usage and logs at apikey.cloud any time. Nothing is hidden on their side.
“Is migration a lot of work?”
Two lines — change base_url and api_key. Model names stay standard (e.g. gpt-5.5, claude-sonnet-4-6).
“My image / long requests time out.”
Those must use the direct endpoint api-direct.apikey.cloud (600s). The default host is capped at 100s and returns 524 on long calls.
“Why not just buy from the official provider?”
Full-spec, uncut models + one key for everything + you get answers in your own language. For many customers that, plus a reachable endpoint, is the whole reason to buy from you.
7 · Operating cadence
Steady deals come from a rhythm, not the occasional blast.
- Daily: contact 5 people who already use an API; post 1 short thing (a setup tip, a cost comparison, a troubleshooting note).
- Weekly: write up 3 real cases (what they used before, what they saved, what tripped them up); review your numbers — new keys, consumption rate, who's due to renew.
7-day kickoff
- Top up a small amount and create one test Client Key for yourself.
- Run the hand-off template end-to-end on your own machine — confirm the white-label Base URL works.
- Walk one real customer from key →
models.list()→ first real call. - Get them to set up usage tracking at apikey.cloud so they self-serve.
- Pick your starting multiplier (1.30x is a safe default) and write your 30-second intro in your own words.
- Send the intro to 3 people who genuinely use an API.
- Log who replied; turn your most-asked question into a saved reply.
Questions? Reach us on Telegram or email support@derouter.ai.