CodePass docs

One key. Two models.
Ship code.

Kimi K3 and GLM 5.3 through a standard OpenAI-compatible endpoint. No catalog to study and no provider keys to juggle.

Base URLhttps://api.codepass.ninja/v1
60-second quickstart

Make your first request.

Send your CodePass key as a Bearer token and choose either glm-5.3 or kimi-k3.

TerminalOpenAI Chat Completions
curl https://api.codepass.ninja/v1/chat/completions \
  -H "Authorization: Bearer $CODEPASS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3",
    "messages": [{"role": "user", "content": "Write a hello-world CLI."}],
    "stream": true
  }'
glm-5.3Fast, capable coding intelligence.
kimi-k3Long-context reasoning for ambitious agents.
Quick setup

Configure your coding agent.

Your key never leaves your browser. It is inserted into the command locally and is not sent to CodePass from this page.

1. Choose a client
2. Choose your machine
3. Pick a default model
Copy and pastePowerShell
$env:CODEPASS_API_KEY=''; $p=Join-Path $env:TEMP 'codepass-setup.ps1'; Invoke-WebRequest https://codepass.ninja/install/codepass-setup.ps1 -OutFile $p; & $p -Client opencode -Model glm-5.3 -NoPersist

The installer backs up your existing config and merges CodePass without deleting your other providers.

Verified configuration contract

OpenCode

The setup helper safely merges a codepass provider into ~/.config/opencode/opencode.json, keeps your existing providers, and stores the key outside the JSON as CODEPASS_API_KEY.

  1. Choose OpenCode in the helper above.
  2. Pick Windows PowerShell or Linux / macOS.
  3. Paste your key, copy the generated command, and run it.
  4. Launch opencode. Switch models with /models.
Not enabled yet

Claude Code

Claude Code compatibility is pending live verification. Claude Code speaks Anthropic Messages through /v1/messages, not OpenAI Chat Completions. We will publish this installer after both CodePass models pass streaming and tool-use tests through that exact endpoint.

Not enabled yet

Codex CLI

Codex custom providers require the OpenAI Responses protocol at /v1/responses. We will publish its installer after that endpoint passes a real Codex session with streaming and tool calls. We will not offer a command that only looks correct.

API reference

OpenAI-compatible by design.

MethodEndpointUse
GET/v1/modelsModels available to your key
POST/v1/chat/completionsChat, streaming, and tool calls

Authentication

Every request includes Authorization: Bearer YOUR_CODEPASS_KEY. Keep keys out of source control and browser-side applications.

Common parameters

model and messages are required. stream, max_tokens, temperature, top_p, tools, and tool_choice follow the standard Chat Completions shape.

Errors

Useful failures, not mysteries.

400Invalid request or model name.
401Missing, invalid, or revoked API key.
402Credit balance exhausted.
429Rate or concurrency limit reached; retry with backoff.
5xxTemporary gateway or upstream failure; retry with backoff.