Skip to content

configuration.md

Latest commit

 

History

History
151 lines (99 loc) · 4.13 KB

configuration.md

File metadata and controls

151 lines (99 loc) · 4.13 KB

Configuration

This page describes how to customize InnoClaw's behavior through configuration options.

Environment-Based Configuration

All configuration is done through environment variables in your .env.local file. See Environment Variables for a complete reference.

AI Provider Settings

Selecting a Provider

InnoClaw supports multiple AI providers. Configure your preferred provider by setting the appropriate API key:

# Use OpenAI
OPENAI_API_KEY=sk-xxxxxxxxxxxxxxxxxxxxxxxx

# Use Anthropic Claude
ANTHROPIC_API_KEY=sk-ant-xxxxxxxxxxxxxxxxxxxxxxxx

# Use Google Gemini
GEMINI_API_KEY=your-gemini-key

# Or configure multiple and switch in the Settings UI

Default Provider and Model

Set the default LLM provider and model used when the application starts:

LLM_PROVIDER=openai       # openai, anthropic, or gemini
LLM_MODEL=gpt-4o-mini     # Any model ID from the selected provider

These defaults can be overridden at any time via the Settings UI.

Using Custom Endpoints

For third-party compatible services, proxies, or self-hosted models:

# Custom OpenAI-compatible endpoint
OPENAI_BASE_URL=https://api.your-provider.com/v1

# Custom Anthropic endpoint
ANTHROPIC_BASE_URL=https://api.your-provider.com

# Custom Gemini-compatible endpoint (OpenAI-compatible proxy)
GEMINI_BASE_URL=https://api.your-provider.com

Separate Embedding Configuration

If your chat model provider doesn't support the embedding endpoint, configure a separate service:

EMBEDDING_API_KEY=sk-your-embedding-key
EMBEDDING_BASE_URL=https://api.your-embedding-provider.com/v1
EMBEDDING_MODEL=text-embedding-3-small

Agent Configuration

Max Tool Steps

Control how many tool-call steps the agent can take per request:

AGENT_MAX_STEPS=10    # Default: 10, range: 1-100

Higher values allow more complex multi-step tasks but consume more tokens.

In-App Settings

Access the settings page at /settings to:

  • Switch AI Provider — Toggle between OpenAI, Anthropic, and Gemini
  • Select Model — Choose the specific model to use for chat
  • View API Status — Check which API keys are configured
  • View Workspace Roots — See the configured workspace root directories
  • Configure HuggingFace Token — Set HF_TOKEN for dataset downloads

Workspace Configuration

Root Directories

The WORKSPACE_ROOTS variable controls which server directories users can create workspaces in:

# Single root
WORKSPACE_ROOTS=/data/workspaces

# Multiple roots (comma-separated)
WORKSPACE_ROOTS=/data/research,/data/projects,/home/user/documents

:::{important}

  • All specified directories must already exist on the server
  • Users can only create workspaces within these root directories
  • Path traversal outside these roots is prevented by security checks :::

Database Configuration

InnoClaw uses SQLite with the database stored at ./data/innoclaw.db by default. You can override this path via the DATABASE_URL environment variable:

# Plain filesystem path (no SQLite URI scheme)
DATABASE_URL=/tmp/innoclaw/innoclaw.db

:::{note} Only plain filesystem paths are accepted (e.g. /data/innoclaw.db). SQLite connection strings like file:./data/innoclaw.db?mode=rwc are not supported — the file: prefix will be stripped automatically. :::

To use a fresh database:

# Remove existing database
rm ./data/innoclaw.db

# Re-run migrations
npx drizzle-kit migrate

Build Configuration

If the project resides on a network filesystem (NFS, CIFS, etc.), the Next.js build cache may fail. Set a local build directory:

NEXT_BUILD_DIR=/tmp/innoclaw-next

Proxy Configuration

For environments that require an HTTP proxy to reach external APIs:

HTTP_PROXY=http://your-proxy:3128
HTTPS_PROXY=http://your-proxy:3128
NO_PROXY=localhost,127.0.0.1,10.0.0.0/8

All outbound fetch() calls (AI API, GitHub, HuggingFace, etc.) will go through the proxy. Hosts listed in NO_PROXY bypass it.

Bot Integration Configuration

See Notifications for configuring Feishu and WeChat bot integrations.