Spend is visible before the bill
Quota and balance are checked before the request is sent. You see spend per model, not one total at month end.
Pay per token · No subscriptions
Every model behind one key
GPT, Claude, Gemini, Seedream and Veo through one [OI]-compatible API. One balance, one request log, pay for what you use — no subscription per service.
99.98% uptime
Catalog flagship modelsRate / Context
Claude Opus 4.8claude-opus-4-8
Anthropic·Reasoning
Active
15M cr/M·1M tokens
GPT-5.6 Solgpt-5.6-sol
OpenAI·Reasoning
Active
5M cr/M·1.1M tokens
DeepSeek V4 Prodeepseek-v4-pro
DeepSeek·Reasoning
Active
1.2M cr/M·128K tokens
Gemini 3.5 Flashgemini-3.5-flash
Google·Tools
Active
400K cr/M·1M tokens
40+ models ready to routeView full catalog
Available models for team traffic.
Browse enabled models and context windows before you route a workload.
AI credit packages
Credits are the internal unit of spend. A request's cost depends on the model, token type and caching; it is visible in the log before you pay.
Select your use case
No subscriptions, no expiration
A package funds your AI-credit balance.
Transparent consumption
AI-credit charges depend on the model, token type, and caching.
Connect the tools your team already uses.
One API instead of ten subscriptions
One key, one balance, one request log. You pay for the tokens you actually use, not a fixed subscription per service.
One key for every model
GPT, Claude, Gemini, Seedream, Veo and more through one [OI]-compatible API. Switching models means changing the model field, not your code.
Pay for tokens, not subscriptions
No fixed fee per service. The balance is charged on actual usage: input, output and cached tokens are counted separately.
Balance and quotas under control
Top up with packages and see spend per model. Set limits on a key so one client cannot exhaust the team budget.
Streaming answers and cancellation
The answer arrives as it is generated, not after it finishes. If the client disconnects, the request closes and you pay only for received tokens.
A log for every request
Every request keeps its model, status, latency, token usage and cost. A disputed bill is settled from the log, not screenshots.
Keys and access in one place
Issue a separate key per tool and revoke it in one action. You can see which key was last used and when.
What happens to a request
A request enters once and stays under control: routing, quota and logging are applied at one point.
Change the model without rewriting code
The model field in the request decides which provider answers. Switching models is changing a string, not your integration.
Every request leaves a trace
The record keeps owner, key, model, status, latency, tokens and cost. Investigating an incident does not require collecting data by hand.
POST/v1/chat/completions
+1.2ms overheadYour appgw_live_key
Existing customer request
Budget guard< 0.8ms
Hosted SaaS control layer
Live logSSE Stream
Zero-buffer SSE & live trace log
Model provider200 OK
Only after account checks pass
From sign-up to the first request
Sign up, issue a key, send a request. Everything after that is measured automatically.
Account and key in a couple of minutes
Create an account, top up the balance and issue a gw_ key. From then on the key works like any [OI]-compatible client's API key.
Two lines change
In your SDK, CLI or agent settings, replace the base URL and the key. The request body and response format stay the same, so the integration is not rewritten.
Spend is visible immediately
Every request lands in the log with its model, status, latency, tokens and cost. No manual statistics gathering.
Control over your team's AI spend
You can see who used a model, which model answered and what it cost. No manual statistics gathering.
- Team budgets
- Scoped API keys
- Request history
- Usage visibility
What the log looks like
An anonymized sample log record: owner, key, model, quota and trace on one line.
userusr_7f2...quota ok
keygw_93...rate ok
modelgpt-5.5OpenAI
tracetrc_41...logged
Frequently asked questions
Brief answers on price, data and models.
Start with one key.
Create an account, top up your balance and issue a key. The first request can go through any [OI]-compatible client.