Tool setup
OpenClaw Integration Guide
Connect OpenClaw to SeaLink for multi-step AI orchestration across supported models. Ideal for custom workflows, batch processing, and execution-log analysis.
Prerequisites
- Node.js ≥ 16 (run node -v to check)
- Git (for cloning repositories)
- A valid SeaLink API Key (starts with sk-sealink-)
- Network access to sealink.io (port 443, HTTPS)
Setup Steps
Install OpenClaw
Start by cloning the OpenClaw repository: git clone https://github.com/openclaw/openclaw.git cd openclaw Then install the dependencies as described in the project documentation: npm install Once done, run openclaw --version to confirm everything installed correctly.
Configure SeaLink environment variables
Create or edit a .env.local file in the project root and add: OPENAI_BASE_URL=https://test.sealink.io/v1 OPENAI_API_KEY=sk-sealink-your-key If you prefer system-level environment variables (available to all applications), you can add the export statements to your ~/.bashrc or ~/.zshrc instead.
Create your workflow configuration
Create a config file in the workflows directory (YAML or JSON format): name: my-workflow steps: - name: analyze model: qwen3.5-plus prompt: Extract the key information from this text... - name: summarize model: deepseek-v4-pro prompt: Based on the analysis from the previous step, write a concise summary Each step can use a different model, so you can pick the best model for each type of task in your workflow.
Initialize the database
OpenClaw needs a database to store workflow definitions and execution results. Run: openclaw db:init SQLite is used by default, which works well for development and testing. For production use, you can configure PostgreSQL for better concurrency and reliability.
Start the OpenClaw service
Launch the service with: openclaw server Or use development mode for hot reloading: openclaw dev The service listens on http://localhost:3000 by default. You can change the port via the PORT environment variable. Once it's running, we recommend opening http://localhost:3000 in your browser to confirm the service page loads correctly.
Execute your workflow
You can trigger workflows via CLI or HTTP API. CLI: openclaw run my-workflow --input "your input text" HTTP API: curl -X POST http://localhost:3000/workflows/my-workflow/run \ -H 'Content-Type: application/json' \ -d '{"input": "your input text"}' For your first run, we suggest using a simple test workflow to verify the full pipeline is working end to end.
Core Features
Workflow Orchestration
Define multi-step AI processes with automatic context and state management
Flexible Model Switching
Pick the best model for each step — all usage is tracked in your SeaLink workspace
Batch Processing Engine
Large-scale task processing with built-in scheduling and concurrency control
Execution Logs
Full execution history, combined with SeaLink usage data for model, cost, and failure analysis
API Interface
RESTful API and WebSocket with streaming output support
Extensibility
Custom nodes, plugins, and third-party service integrations to fit your workflow needs
Common Issues
Q: Cannot connect to SeaLink API
A: Let's go through this step by step: 1. Make sure both OPENAI_BASE_URL and OPENAI_API_KEY are properly set 2. Run echo $OPENAI_API_KEY to verify the key is loaded in your environment 3. Confirm your network can reach https://test.sealink.io 4. Test directly with curl: curl -H 'Authorization: Bearer sk-sealink-xxx' https://test.sealink.io/v1/models If all of the above check out but you still can't connect, verify that no proxy or firewall is blocking outbound HTTPS traffic on port 443.
Q: Workflow execution fails or times out
A: Here's a troubleshooting sequence we recommend: 1. Check OpenClaw logs for the specific error message 2. Run openclaw models:list to confirm the model is available 3. Create a minimal single-step workflow to isolate the issue 4. Increase the timeout value in your workflow config 5. Check your SeaLink dashboard to confirm the API Key has enough balance Step 3 usually surfaces the root cause quickly.
Q: Costs are increasing but I'm not seeing results
A: Costs without visible output can come from a few sources: 1. Multiple workflow runs during testing and debugging — each execution counts 2. Higher-priced models in your workflow — check /models and /pricing for per-model rates 3. Requests that were recorded but didn't complete successfully — check logs for failures 4. Run openclaw logs:view to inspect recent executions and find calls without output We recommend using lightweight models like gpt-4o-mini during development, then switching to production models once the workflow is correct.
Q: Database initialization fails
A: Here's what to check: 1. Make sure your user has read/write permissions on the database directory 2. For SQLite: try removing the old database file and reinitializing: rm openclaw.db && openclaw db:init 3. For PostgreSQL in production, verify the connection string: OPENAI_DATABASE_URL=postgresql://user:pass@localhost/openclaw openclaw db:init 4. Confirm the PostgreSQL service is running and the target database exists