Connect the gateway to agent runtimes (OpenClaw, Hermes, etc.)
Agent runtimes call POST /v1/chat/completions over OpenAI-compatible HTTP, often with tools / tool_calls, SSE, and content: null on assistant messages. For OpenClaw, Hermes Agent, etc., you effectively point them at a custom OpenAI backend (Base URL + Bearer + model).
Each product’s config files and env vars follow its docs; this page only covers HTTP-level commonalities.
Integration checklist
- Base URL:
https://<host>/v1(verify auto-append behavior withGET /v1/models). - API Key: valid tenant
sk-…. - Model: exact
codefromGET /v1/models. - TLS / SSE: cert chain, proxies must not buffer
text/event-stream, timeouts long enough.
OpenClaw
Configure OpenAI-compatible baseUrl + apiKey` (field names vary by version). Upstream: openclaw/openclaw, #3307.
content may be null and other tool-call validation details: see chatCompletionSchema comments and Chat Completions.
Hermes Agent
In Adding Providers / Configuration, point the LLM upstream at this service’s …/v1. Do not confuse with Hermes’ built-in API server (local listener): you want an external OpenAI-compatible backend, not “Hermes only listens on localhost”.
Multimodal
input_file rules: Chat Completions. On Node, @routerbrain/sdk can assemble attachments: chat.messages.