One endpoint,
every model.
A unified API to OpenAI and Anthropic (more soon). Bring your own keys — encrypted at rest — and every call is tracked: tokens and cost. Route by model name, no vendor SDK. The flat-rate alternative to Vercel AI Gateway, OpenRouter and LiteLLM.
Every model has its own SDK, its own billing, its own dashboard
Wiring each LLM provider separately — different SDKs, keys scattered across apps, no unified spend view — slows every AI feature down. koigrid AI Gateway gives you one endpoint: send a model name and messages, and it routes to the right provider using your own key (stored encrypted), records tokens and estimated cost, and returns the completion. Perfect for agents that need to call models with a scoped token.
How koigrid compares
| koigrid | Direct SDKs | |
|---|---|---|
| Pricing | Flat rate + your provider | Per provider |
| One unified endpoint | Yes | No |
| Bring your own keys | Yes (encrypted) | Yes |
| Usage + cost tracking | Built in | DIY |
| Route by model name | Yes | No |
| Agent token + scopes | Native | No |
From public docs, 2026. koigrid routes gpt-*/o1-* to OpenAI and claude-* to Anthropic; more providers on the roadmap. Your keys never appear in logs or events.
The gateway your agents call
Unified chat
One request shape for every model — send messages, get a completion.
BYO keys, encrypted
Your provider keys are envelope-encrypted at rest and never logged.
Usage & cost
Every call records prompt/completion tokens and an estimated cost.
Model routing
gpt-*/o1 → OpenAI, claude-* → Anthropic, automatically.
Agent-native
Agents call models through the gateway with a scoped token.
No lock-in
A provider seam — swap or add providers without touching your code.