Flat-rate billing
One payment covers the whole pass. Prompts are unlimited, so a long agent run costs exactly what a one-line question costs — nothing extra. No token counting, no overage line, no bill shock at month end.
One API key, every frontier model
ProjectCOZY is a managed gateway that puts Kimi, Claude, DeepSeek, GLM, Qwen and MiniMax behind a single OpenAI-compatible endpoint — billed at a flat rate instead of per token.
Ten-plus frontier models
Kimi K3, Kimi K2.6, Claude 5 Opus, Claude 5 Sonnet, DeepSeek V4 Pro, GLM 5.2, GLM 5.1, MiniMax M3, MiniMax M2.7 and Qwen-3.5-Coder — all on the same key, all at the same price.
OpenAI-compatible API
A drop-in chat-completions endpoint. Change the base URL and the model alias and your existing code, SDKs and tooling keep working — including streaming and tool calling.
One key, every tool
A single cozy_ key drives your editor, your terminal agent and your CI job at once. Revoke and regenerate it from the dashboard any time without touching billing.
Streaming and tool calls
Server-sent events pass straight through, so token-by-token output arrives exactly as your client expects. Function and tool calling are supported on the models that implement them.
Usage analytics
Request counts, token totals and per-model daily trends in the dashboard. You get the visibility of a metered API without being billed like one.
Hashed key storage
Only a SHA-256 hash of your key is stored. The raw value is shown once at generation and never retrievable — if it leaks, revoke and reissue in a click.
Crypto payments
Pay with Binance Pay in any coin — USDT, BTC, ETH, BNB or SOL — plus KHQR where enabled. No card, no billing address, no per-seat minimum.
Instant activation
Your key is live the moment payment clears. No approval queue, no sales call, no waiting list — register and you are making requests in under a minute.
The models you get
Every pass includes every model. Switching is a one-line change to the model alias in your client — no separate key, no tier upgrade, no renegotiation.
Kimi K3
Long-context agentic coding and deep multi-step tool use.
Kimi K2.6
Fast general-purpose coding with a large context window.
Claude 5 Opus
Hardest reasoning, refactors and architecture work.
Claude 5 Sonnet
Balanced speed and quality for day-to-day development.
DeepSeek V4 Pro
Strong reasoning and maths at high throughput.
GLM 5.2
Reliable instruction following and structured output.
MiniMax M3
Long-horizon agent runs and high-volume batch work.
Qwen-3.5-Coder
Code completion, test generation and translation.
Works with the tools you already use
Anything that speaks the OpenAI chat-completions format works unmodified. Set the base URL, paste your cozy_ key, pick an alias.
Claude Code
Cline
OpenClaw
OpenCode
Kilo Code
Roo Code
Cursor- Aider
- Continue.dev
- Any OpenAI SDK
Full setup instructions for each client live in the documentation.
How it works
- 1
Create an account
Sign up with email or Telegram. Every new account starts on the free trial — no card required.
- 2
Generate your API key
One click in the dashboard issues a cozy_ key. It is displayed once, so copy it straight into your tool.
- 3
Point your client at the gateway
Set the base URL to our OpenAI-compatible endpoint and paste the key. No other code changes are needed.
- 4
Pick a model alias
Name the model you want in the request. Switch between Kimi, Claude, DeepSeek, GLM, Qwen and MiniMax whenever you like — same key, same price.
Try it on the free trial
No card required. See the pass lengths and prices on the pricing page, or start with the trial and test the models on your own codebase first.