Each page targets one error, names the causes in order of likelihood, and gives you a command that proves which one you have. No preamble, no history of AI — the diagnostic is the first thing on the page.
Free endpoints fail in patterns. A 60-second diagnostic to tell provider death from your own misconfiguration, plus the two-endpoint fallback pattern.
authUsually not a bad key. Most often the process never saw your variable, or the endpoint wants a different auth header than your tool sends.
modelsAuth worked, model resolution failed. Stop typing the slug from memory — list what the endpoint actually serves and copy an id verbatim.
limitsFree ways to cut token consumption first, then an honest account of when reducing usage genuinely stops being enough.
modelsSilent fallback chains, auxiliary model slots, and config precedence. A cost bug that produces no error message.
If your exact error isn't listed, this sequence localises almost any coding-agent connection problem in a few minutes.
Every page here follows the same rule: reproduce the failure with a command before changing configuration. Most time lost to these errors is spent editing config files while the actual failure sits one layer away.
Work the diagnostic first — most of these errors are configuration. But a few genuinely resolve to needing an endpoint that serves the model you want, or one you can point every tool at with the same base URL. AgentRouter is OpenAI-compatible with published per-token pricing and documented setup for 15 tools, so it satisfies that case without an adapter and without a subscription window.
Look at the setup guides →Referral link — we may earn a reward if you sign up, at no extra cost to you. This is the same advice we would give with no link at all. Check their current pricing and model list on their own site; we deliberately do not restate numbers that change.
The Setup Doctor walks through your tool, symptom, and setup and points at the likely cause — no account needed.