Skip to content

API_KEYS.md

Latest commit

 

History

History
87 lines (58 loc) · 2.44 KB

API_KEYS.md

File metadata and controls

87 lines (58 loc) · 2.44 KB
Raw

API keys — plain English

You only need one key to run OpenWorker. The two names below are roles, not two separate bills.

The short version (OpenRouter)

  1. Get a key from openrouter.ai/keys — it starts with sk-or-
  2. Put it in .env:
OPENROUTER_API_KEY=sk-or-v1-...
  1. Done. Chat and desktop automation both use it.

You can also paste the same key in the web UI under API Key Settings (localhost only unless you set OPENDESKTOP_API_TOKEN).

What are CHAT vs VISION?

Name What it does When it runs
Chat Fast text replies — OpenWorker conversation, intent routing "What can you do?", planning, quick answers
Vision Looks at screenshots of the sandbox and clicks/types Research, forms, desktop automation

Same provider, same key in most setups. We split the env vars so you can use a cheap model for chat and a stronger one for desktop — but you don't have to.

# One key (recommended)
OPENROUTER_API_KEY=sk-or-v1-...

# OR explicitly the same key twice (also fine)
CHAT_API_KEY=sk-or-v1-...
VISION_API_KEY=sk-or-v1-...

# OR split models later (power users)
CHAT_API_KEY=sk-or-v1-...
CHAT_MODEL=openai/gpt-4o-mini
VISION_API_KEY=sk-or-v1-...
VISION_MODEL=anthropic/claude-3.5-sonnet

OpenRouter vs OpenAI

Provider Env var Key shape
OpenRouter (recommended) OPENROUTER_API_KEY sk-or-v1-...
OpenAI direct CHAT_API_KEY / VISION_API_KEY sk-proj-... or sk-...
Generic fallback API_KEY any

OpenRouter routes to OpenAI, Anthropic, Google, etc. with one account — handy for indie budgets.

Local Ollama (no cloud key)

CHAT_API_URL=http://localhost:11434/v1/chat/completions
CHAT_MODEL=llama3.1
VISION_API_URL=http://localhost:11434/v1/chat/completions
VISION_MODEL=llava

Desktop automation quality depends on the vision model — cloud models are more reliable today.

Security

  • Never commit .env or keys to git
  • POST /api/v1/keys/set only works from localhost/LAN, or with OPENDESKTOP_API_TOKEN when the server is exposed
  • For production: set OPENDESKTOP_API_TOKEN and keep keys in .env only

Check it's working

curl http://localhost:8000/api/v1/health | jq '.api_key_configured, .llm_provider'

Expect: true and "openrouter" (or "openai" / "custom").