Documentation
Connect in under two minutes
ProjectCOZY speaks the OpenAI chat-completions format. Point any compatible client at the base URL below, paste your key, and name a model alias. No SDK to install, no code to rewrite.
- Base URL
https://api.projectcozy.me/v1- Auth header
Authorization: Bearer cozy_…Get a key- Model field
"model": "kimi-k2.6"All aliases
Make your first request
Runs as written once you swap in your key.
cURL
curl https://api.projectcozy.me/v1/chat/completions \
-H "Authorization: Bearer cozy_your_key_here" \
-H "Content-Type: application/json" \
-d '{
"model": "kimi-k2.6",
"messages": [
{ "role": "user", "content": "Say hello in one line." }
]
}' Streaming works the same way: set "stream": true and read the server-sent event stream exactly as you would from OpenAI.
Model aliases
Every alias is included in every pass. Switching model is a one-word change.
| Alias | Model | Best for |
|---|---|---|
kimi-k3 | Long-context agentic coding and deep tool use | |
kimi-k2.6 | Fast general-purpose coding, large context | |
kimi-k2.5 | Previous generation, kept for stability | |
claude-5-opus | Hardest reasoning, refactors, architecture | |
claude-5-sonnet | Balanced speed and quality, day to day | |
deepseek-v4-pro | DeepSeek V4 Pro | Reasoning and maths at high throughput |
glm-5.2 | GLM 5.2 | Instruction following, structured output |
glm-5.1 | GLM 5.1 | Previous generation, kept for stability |
minimax-m3 | MiniMax M3 | Long-horizon agent runs, batch work |
minimax-m2.7 | MiniMax M2.7 | Previous generation, kept for stability |
qwen-3.5-coder | Qwen-3.5-Coder | Completion, test generation, translation |
Endpoints
Relative to the base URL above.
| Method | Path | Returns |
|---|---|---|
| POST | /chat/completions | Chat completion, or an SSE stream when "stream": true |
| GET | /models | Every alias your pass can call |
Editor and agent setup
Most tools read these two variables. Set them and restart the tool.
env
OPENAI_BASE_URL=https://api.projectcozy.me/v1
OPENAI_API_KEY=cozy_your_key_hereClaude Code ANTHROPIC_BASE_URL
Cline OpenAI Compatible
OpenClaw OPENAI_BASE_URL
OpenCode OPENAI_BASE_URLKilo Code OpenAI Compatible
Roo Code OpenAI CompatibleCursor Override base URL
- Aider OPENAI_API_BASE
If the first request fails
- 401
- Key rejected. Check for a trailing space when pasting, and that the key starts with cozy_. Keys are shown once, so regenerate if unsure.
- 402
- Pass expired or trial used up. Buy or renew a pass; the same key reactivates.
- 404
- Wrong path. The base URL must end in /v1, and your client appends /chat/completions itself. A doubled /v1/v1 is the usual cause.
- 400
- Unknown model. Use an alias exactly as listed above. Upstream model IDs are not accepted.
DeepSeek V4 Pro
GLM 5.2
MiniMax M3
Qwen-3.5-Coder