The request enters one endpoint
Codex, OpenClaw, and internal agents only need one OpenAI-compatible address.
Lumoswitch
Scroll to exploreOne endpoint for your AI agents.
Connect Codex and internal agents directly to Lumoswitch while managing models, fixed and failover routing, access limits, billing, and observability in one place.
sk-xxxxxDeepSeeksk-xxxxxMiniMaxsk-xxxxxQwensk-xxxxxGLMsk-xxxxxDeepSeeksk-xxxxxMiniMaxsk-xxxxxQwensk-xxxxxGLMCodex
Hermes Agent
OpenClaw
OpenCode
Codex
Hermes Agent
OpenClaw
OpenCode
Lumoswitch inference runtime
Agents call the runtime directly while the control plane syncs models and limits
Requests go directly to Lumoswitch, where fixed routing, failover, and access limits execute in-path.
The runtime resolves routing policy
Lumoswitch applies fixed models, candidate order, budgets, and access limits in the request path.
The primary fails; fallback takes over
A 429, timeout, or outage does not make clients change endpoints. The runtime tries the next candidate.
The response returns; usage is imported
After completion, Lumoswitch imports tokens, cost, and trace data while inference remains independent.
Clients
Codex
OpenClaw
Hermes
Lumoswitch RuntimeRouting intelligently
Model network
AIExample primary model842ms · primary
Example fallback model Astandby · fallback
Example fallback model Bstandby · fallback
Example route is healthy
The following is product capability demo data. It is not a real model name, live production status, or SLA commitment.
Connect directly
Copy the config. Keep your client.
Pick an agent and copy the unified endpoint and access key. Models, providers, and failover continue to change in the control plane.
01Connect models and credentials02Create a route03Copy the URL and ship
export OPENAI_BASE_URL="https://api.lumoswitch.com/v1"
export OPENAI_API_KEY="sk-lumoswitch-..." Your route is ready
Switch models and providers while agent endpoints stay put
Manage model access, routing, access limits, usage, and cost from Lumoswitch.