Docs / Client Setup

Different tools need slightly different route shapes. The docs should admit that early.

OpenMagic API needs to meet users where they already work: Claude Code, Codex, Cursor, Cline, SDKs, and lightweight local tools.

Open quick start Open models page

Universal settings

Use these unless a client explicitly asks for a different endpoint field.

Field	Value
Base URL	https://api.openmagic.ai/v1
API key	Your OpenMagic console key
Chat model	gpt-4o-mini or another model listed on /models
Compatibility	OpenAI-compatible chat/completions route

# OpenAI SDK / generic compatible clients
OPENAI_API_KEY=sk-om-...
OPENAI_BASE_URL=https://api.openmagic.ai/v1

Codex CLI

Install with npm install -g @openai/codex. Start with environment variables: OPENAI_API_KEY=sk-om-... and OPENAI_BASE_URL=https://api.openmagic.ai/v1. First task: ask it to inspect a small folder and explain before editing.

OpenCode style CLI

If the CLI supports OpenAI-compatible providers, use the same two fields: API key and base URL. Prefer gpt-4o-mini for the first run, then switch to stronger coding models after usage is visible in Console.

Cline CLI / Extension

Cline CLI supports an OpenAI provider with custom base URL. Use cline auth -p openai -k sk-om-... -b https://api.openmagic.ai/v1. In VS Code, choose OpenAI-compatible provider and fill the same fields.

Roo Code

Open Roo settings, select OpenAI Compatible, set Base URL to https://api.openmagic.ai/v1, API Key to your OpenMagic key, and Model ID to a supported model. If model dropdown is empty, type the model manually.

Continue

Use provider openai and set apiBase to https://api.openmagic.ai/v1. Configure separate models for chat/autocomplete later; first verify chat with one model.

Cursor

Use custom OpenAI-compatible settings if your version exposes Base URL. Paste the OpenMagic key, set model name manually if needed, and run a short prompt before using Composer on a whole repo.

LobeChat / NextChat

For self-hosted web chat, set server-side OpenAI key to your OpenMagic key and OpenAI proxy/base URL to https://api.openmagic.ai/v1. Keep access code enabled if the app is public.

Claude Code

Claude Code installs with Anthropic's package and supports ANTHROPIC_BASE_URL. Use ANTHROPIC_BASE_URL=https://api.openmagic.ai/anthropic and ANTHROPIC_API_KEY=sk-om-.... This route accepts Anthropic Messages format and converts it to the OpenMagic gateway internally.

Windsurf

Some versions expose only official BYOK and not custom Base URL. If your version has custom OpenAI-compatible Base URL, use https://api.openmagic.ai/v1; otherwise it cannot directly route to OpenMagic.

Recommended first settings

Start safe; increase limits only after the first task succeeds.

Setting	Beginner value	Why
Model	gpt-4o-mini	Cheap smoke test before stronger routes.
Max output tokens	1024-2048	Enough for explanations, avoids runaway output.
Temperature	0.2-0.5	More stable for coding tasks.
Context / max input	Use tool default first	Only raise it when the tool reports context overflow.
First task scope	One folder or one bug	Avoid whole-repo rewrites on first run.

curl https://api.openmagic.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-om-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-4o-mini",
    "messages": [{"role": "user", "content": "Reply with only OK."}],
    "max_tokens": 10
  }'

Troubleshooting

Most first-call failures fall into these buckets.

Symptom	Check
401 unauthorized	Confirm the key is copied from OpenMagic console and sent as Bearer token.
404 or wrong route	Use https://api.openmagic.ai/v1, not the marketing homepage URL.
Model not found	Use a model from /models or your assigned route group.
Quota exhausted	Open Console > Billing to recharge or buy a monthly quota plan.

Warning: session_name(): Cannot change session name when headers already sent in /www/wwwroot/api.openmagic.ai/lib/UserSession.php on line 13

Warning: session_start(): Cannot start session when headers already sent in /www/wwwroot/api.openmagic.ai/lib/UserSession.php on line 14