OpenClaw

Use a custom model connection

OpenClaw documents custom model connections through models.providers, including an API base URL, credentials, adapter and model list. The official configuration reference distinguishes Chat Completions, Responses and Anthropic Messages adapters. Choose the adapter that matches the selected model's documented API.

Configure the connection

Create a SeedRouter API key and keep it in your local secret environment. For an OpenAI-compatible connection, the API base URL is https://api.seedrouter.ai/v1. Use the exact model identifier from the model directory.

Add the connection to your existing OpenClaw configuration rather than replacing unrelated settings. Set the model's context and input capabilities from its actual documentation. Do not copy another model's context limit or assume vision support because the application accepts image attachments.

OpenClaw's optional capability declarations also depend on the endpoint and model. Follow the official reference for your installed version instead of enabling every capability flag.

Check a bounded task

Record your OpenClaw version, model identifier and configuration before testing. Verify a plain response, then a read-only tool task. For an ecommerce draft, give the agent a small approved product record and inspect the saved output against it.

Check that tool arguments are valid, missing information remains unresolved and the agent stops after the requested deliverable. Add business-system permissions only for the task you actually intend to run.

Measure useful work

OpenClaw manages the agent session and its tools; SeedRouter supplies the selected model calls. Track total requests and revisions for a completed task in your usage history, using current pricing. One short response cannot establish the ongoing cost of an autonomous session.

This guide is based on the configuration reference checked on October 12, 2026. It does not report a new end-to-end OpenClaw benchmark or certify every tool/model combination.