Find the fix for the error text you actually saw
Most API troubleshooting pages group failures by status code. That is not how people search: they paste the sentence their editor printed. This index maps the literal message to the page that fixes it, because the same 404 can mean a wrong base path, an unknown model ID or a disabled tool-calling route.
Exact error text, mapped to the page that fixes it
Search engines match the string people actually paste, not the HTTP status class. A 404 from a wrong base path and a 404 from an unknown model ID need different fixes, which is why this index is keyed on the message.
| Error text | Where it comes from | Go to |
|---|---|---|
cursor's unification api is unreachable. | Cursor / connectivity | Fix |
this run hit an unexpected error. retry or send a follow-up message | Cursor Agent / tool call | Fix |
langflow_ssrf_allowed_hosts | Langflow / SSRF allowlist | Fix |
401 Unauthorized | Claude Code / headers | Fix |
429 rate limit exceeded | Cursor / quota | Fix |
404 model not found | Cursor / model ID | Fix |
context length exceeded | Any client / context | Fix |
stream disconnected | Codex CLI / streaming | Fix |
models list empty | Cline / discovery | Fix |
base url not working | Any client / routing | Fix |
gemini cli 404 not found | Gemini CLI / path | Fix |
opencode context overflow | OpenCode / context | Fix |
Browse the other clusters
All 21 guides in this cluster
- Base URL not working: seven causes, in order
- Claude API errors: 401, 404, 429 and 529, diagnosed
- Claude Code 401: the header is the difference
- Claude Code 429: which limit you actually hit
- Cline OpenAI Compatible 401: Invalid API Key and Base URL Fixes
- Codex CLI: stream disconnected before completion
- context_length_exceeded on a custom endpoint
- Cursor 429: why agent mode hits rate limits first
- Cursor Agent Errors on a Custom Endpoint: How to Localise Them
- Cursor API key errors — every common failure and the fix
- Cursor model_not_found: the string is the problem
- Cursor request timeout: measure before you raise the limit
- Cursor: what "unification API is unreachable" actually means
- Empty model picker on a custom endpoint: read /v1/models first
- Gemini CLI 404: two causes, both in the URL
- Langflow: when the endpoint host is blocked, not the key
- LiteLLM model names: the prefix is a routing instruction
- LiteLLM retries and timeouts: the budget you did not set
- OpenAI-compatible error codes: 401 vs 404
- OpenCode: prompt is too long, and the field that decides when
- Zed edit predictions stay silent while chat works
FAQ
Why not just look up the HTTP status code?
A status class does not identify the cause. On a custom endpoint, 404 covers an unversioned base URL, a model ID the tool never added, and a route the provider does not expose. Each needs a different change, so the exact message is the useful key.
I see an error that is not listed here. What should I check first?
Confirm the request reaches your own key at all, by checking the usage log on the gateway dashboard. A normal chat reply is not proof that Agent mode, streaming or tool calls are using the same route.
Does an error mean the gateway rate is wrong?
No. Routing errors concern URL shape, headers, model IDs and tool support; prices only decide what a successful call bills. Check the current rate separately before you change a workload.
Check the current rate for your model before you commit a workload.