Sign up
Email and a password. You get a live key immediately, against a zero balance.
OpenAI-compatible · prepaid credits · no surprise invoices
Your agent's model bill, finally under control.
Anastos Harness is one endpoint in front of OpenRouter. You get a key, a balance and per-request metering — the provider key never reaches your code, and no request runs without a balance to pay for it.
curl https://anastosharnessagent.creativemagic.studio/v1/chat/completions \
-H "Authorization: Bearer ah_…" \
-H "content-type: application/json" \
-d '{"model":"anthropic/claude-sonnet-4.5",
"messages":[{"role":"user","content":"Ship it."}],
"stream":true}'Three steps, no sales call
Add credits
Card payment through Stripe. Credits land on your balance the moment Stripe confirms.
Point your app at it
Change one base URL. Streaming, tools, JSON mode — the OpenAI wire format, unchanged.
What you actually get
One key, many models
Claude, GPT, Gemini, DeepSeek, Llama and the rest of OpenRouter's catalogue. Switch model ids per request; nothing else changes.
Metered at cost, per call
Every response is billed from the provider's own usage figure. Your dashboard shows the cost of each request, not a monthly guess.
Balance-checked before it spends
Requests are refused with a clear 402 when the balance is gone. A single call cannot outrun the account.
Keys you can rotate
Issue a new key in one click; the old one stops working immediately. Nothing is shared between customers.
Streaming that streams
Server-sent events pass straight through. No buffering, no "it all arrives at the end".
Provider keys stay ours
Your code holds an ah_… key. The OpenRouter key lives only on the gateway.
Credits, not subscriptions
1 credit = $1 of metered usage. Credits do not expire.
Starter
$20once
20 credits
Enough to build and ship a small agent.
Pro
$100once
110 credits +10 bonus
For daily use. 10% bonus credit, no expiry.
Scale
$500once
600 credits +100 bonus
Heavy agent traffic. 20% bonus credit.
Prices in USD. Usage is billed at the provider's cost plus the platform markup shown at checkout.
Questions
Is it really OpenAI-compatible?
Yes: /v1/chat/completions, /v1/models, /v1/embeddings, streaming and usage blocks. Point the OpenAI SDK's baseURL at us.
What happens when credits run out?
Calls return 402 with insufficient_credit. Nothing is silently queued and no invoice appears later.
Do you train on my prompts?
The gateway is a pass-through with accounting. Prompt routing is pinned to providers that do not retain prompts for training.
Can I get an invoice?
Every payment is a Stripe receipt, and the dashboard keeps the ledger of what each request cost.