.env.example

# Required — LLM API key (default provider: OpenRouter):
OPENROUTER_API_KEY=sk-or-...
# Alternatives (set NANOCLAW_PROVIDER accordingly):
# ANTHROPIC_API_KEY=sk-ant-...
# OPENAI_API_KEY=sk-...

# Required — at least one platform adapter (fill in the one you use, comment out the rest):
TELEGRAM_TOKEN=your-telegram-bot-token
# SLACK_BOT_TOKEN=xoxb-...        # Slack needs both tokens
# SLACK_APP_TOKEN=xapp-...
# DISCORD_TOKEN=your-discord-bot-token

# Optional — LLM provider settings (override config.yaml defaults):
# NANOCLAW_MODEL=anthropic/claude-sonnet-4-5
#   Model ID passed to the API. For OpenRouter use "provider/model" format,
#   e.g. "anthropic/claude-opus-4-5", "openai/gpt-5.5-pro", "google/gemini-3.1-pro-preview".
#   For direct Anthropic: "claude-opus-4-8". For OpenAI: "gpt-5.5-pro".
#
# NANOCLAW_PROVIDER=openai_compatible
#   Which provider driver to use: anthropic | openai | openai_compatible
#   Use "openai_compatible" for OpenRouter or any OpenAI-compatible endpoint.
#
# NANOCLAW_BASE_URL=https://openrouter.ai/api/v1
#   API base URL. Only needed for openai_compatible provider.
#   Leave empty when using anthropic or openai directly.

# Optional — agent behavior:
# NANOCLAW_PERMISSION_MODE=default
#   Controls whether the agent can write files and run shell commands:
#     default — prompts the user before any write or shell operation
#     auto    — allows everything without prompting (use in trusted/CI environments)
#     plan    — read-only; all writes and shell commands are blocked
#
# NANOCLAW_MAX_TOKENS=8192
#   Maximum tokens the model generates per turn. Higher = longer replies, higher cost.
#
# NANOCLAW_MAX_ITERATIONS=40
#   Maximum tool-use rounds per user message before the agent gives up.

# Optional — other settings:
# NANOCLAW_WORKSPACE=~/.nanoclaw/workspace
#   Path to workspace directory containing AGENTS.md and USER.md.
#
# NANOCLAW_WEB_PROXY=http://host:port
#   Opt-in HTTP proxy for web_fetch and web_search tools.
#   Must be http:// or https://. Credentials in the URL are rejected.