Unified AI Gateway
One API key, all AI models.
OpenAI-compatible gateway. Switch providers, track usage, pay as you go.
Drop-in compatible with OpenAI SDKs and tooling (/v1/chat/completions, streaming supported).
Supported providers
OpenAIDeepSeekComing: AnthropicComing: GeminiComing: GLMComing: MCP Tools
Features
OpenAI-compatible integration
Use your existing OpenAI SDKs and ecosystem tools with minimal changes.
- Endpoints:
POST /v1/chat/completions(incl.stream=true),GET /v1/models - Standard error shape with request_id for easy debugging
One key, many upstreams
Create one API key, then select providers by model without wiring multiple vendor SDKs or juggling keys.
- Built to support 2+ upstreams in MVP
- Add/expand providers later without refactoring your app
Transparent usage (and cost-ready)
See exactly what happened on each request so you can trust and operate the system.
- Usage visibility in P0a (recent request logs, latency, tokens when available)
- Wallet + auto-billing in P0b (Stripe top-ups, per-request cost and deductions)
Operability-first defaults
Designed for “ship fast, debug fast” in a serverless setup.
- Structured JSON logs, consistent error codes, request tracing
- Basic safety controls: rate limits, sensible caps (e.g., max_tokens)
Stats
≤ 5 minutes
Target time from sign-up to first successful API call (TTFC).
2+ upstream providers
MVP supports at least two upstreams to reduce lock-in and simplify experiments.
< 150 ms P50 overhead
Target gateway-added latency (excluding upstream inference time).
50–200 recent requests
Minimum “last calls” visibility in Console for quick troubleshooting.
Get to your first call in minutes.
Sign in, generate an API key, and keep using the OpenAI-compatible API you already know. When you’re ready, top up and move to per-request billing with a clear audit trail.