A timeout or dropped stream tells you that one part of an AI-agent request stopped waiting or lost its connection; it does not, by itself, tell you whether the provider received the request or whether the agent finished its work. Start by matching client logs to provider records, identify the exact timer and transport involved, then inspect the run and any tool side effects before retrying.
First find out where the request failed
Compare your client’s error and timestamp with the provider’s request or error records. Filter the provider dashboard to the affected project and one model at a time so unrelated traffic does not obscure the result. OpenAI’s troubleshooting guidance says that a client-side error with no corresponding Service Health data is likely not to have reached OpenAI; timeouts, proxies and network conditions are among the possible causes. A missing provider record is a clue, not proof of a particular network fault. OpenAI troubleshooting guidance
As an Amazon Associate I earn from qualifying purchases.
Capture the client’s exact timestamp and timezone, model and project, request ID if available, HTTP status or error, configured timeouts, and latency percentiles. Those details help distinguish a local wait limit from a provider-side error and are also useful if you need to escalate an issue. OpenAI recommends examining request errors and latency and supplying request IDs and timezone-qualified timestamps. OpenAI troubleshooting guidance
Identify which timeout actually fired
There may be several independent clocks in an agent workflow. A client or proxy can stop waiting before a model call completes; the model call can have its own attempt limit; a tool can time out separately; and the overall agent run can have a deadline. Retry delays may also extend total elapsed time. Raising one timeout does not necessarily change the others.
#1 Best Overall
Model-call attempt timeout
The OpenAI Agents Python SDK documents ModelSettings.timeout as a limit on one model-call attempt, including transport waits. It does not bound the complete agent run, function-tool execution or retry backoff. When the attempt exceeds that limit, the SDK cancels it and raises ModelTimeoutError after cleanup. SDK-managed retries classify timeout failures and apply replay-safety rules. These details describe that SDK, not every agent framework; check the installed package version and the settings in your own runtime. OpenAI Agents Python SDK configuration
Compare limits across the workflow
Put the client, proxy, model-call, tool and overall-run limits beside the expected duration of each operation. If an upstream client or proxy limit is shorter than a normal model response, it can report failure even when the provider has no matching failure record. If the model call is completing but the workflow stalls, inspect the tool or run deadline rather than increasing the model-call timeout blindly. OpenAI Agents Python SDK configuration OpenAI troubleshooting guidance
Check streaming, idle connections and transport behavior
For a long request, establish whether the client streams, how its read timeout is defined, and whether a proxy or other intermediary closes connections that appear idle. A connection can be lost without a useful response reaching the application. Streaming can suit long requests, but the correct settings depend on the SDK and transport.
HTTP read timeout versus time between content chunks
These timers can measure different things. The LangChain OpenAI reference distinguishes an inter-byte HTTP read timeout from an application-level timeout between parsed content chunks. For example, SSE keepalive comments may reset the HTTP read timer while producing no application content chunk, so one timer can remain satisfied while the other expires. Confirm which setting is firing in the installed integration rather than assuming that any incoming bytes count as model output. LangChain OpenAI reference
Vendor-specific SDK guidance
Anthropic’s Python SDK documentation identifies idle network connections as a possible cause of failed or timed-out requests without a response, discusses TCP keep-alive and recommends streaming for long requests. It documents two timeout retries by default and says a non-streaming request is expected to trigger a ValueError after approximately 10 minutes unless streaming or a timeout override is used. Those are mutable, SDK-specific defaults and behaviors; verify them against the version you run. Anthropic Python SDK: long requests
For the OpenAI Agents Python SDK’s optional Responses WebSocket transport, the documentation advises increasing ping_timeout for long reasoning turns that encounter keepalive timeouts. It describes a shared connection as processing one response at a time and having a 60-minute limit. Fully consume streamed results before the session context exits: leaving while a request is in flight may close the connection. The SDK documentation recommends HTTP/SSE when reliability matters more than WebSocket latency. Treat these as details for that transport and SDK, not rules for all WebSocket implementations. OpenAI Agents Python SDK models and transports
Check whether the agent or its tools already completed work
OpenAI’s Agents API recovery documentation states: “An error event or a disconnected stream doesn’t confirm the turn’s final state.” Retrieve the session or turn status and inspect its saved output and tool results before resubmitting. A turn may have changed files or called external tools before the connection dropped. If it is still active, continue following it; if it completed, use its result. OpenAI API documentation: Errors and recovery — Agents API
In managed environments, a connection event describes connection state; it does not itself restart a killed command or guarantee that a tool succeeded. A mid-turn disconnect may fail a tool even if the overall turn completes, and pending input may not be recovered after a process crash. Check the tool result, final response and durable state before replaying an action. OpenAI Agents API reference
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Retry only when the outcome and replay risk are understood
For a rate limit, overload, timeout or temporary service failure, first check whether the original request or action succeeded. OpenAI advises waiting, limiting retries, honoring Retry-After when supplied, and stopping automatic retries if the error changes or a limit is reached. Invalid input, credentials, permissions and billing limits need a fix rather than repeated attempts. OpenAI API documentation: Errors and recovery — Agents API
Before retrying, consider whether a response had started, whether the operation is safe to replay, whether a tool may already have changed external state, and whether attempts and total time are bounded. The Agents SDK retry policy uses normalized facts such as timeout or network classification, response-start state and replay safety. For state-changing tools, application-level deduplication or reconciliation may be needed: a disconnected client may not know whether the side effect happened. OpenAI Agents Python SDK configuration OpenAI API documentation: Errors and recovery — Agents API
What to include in a useful incident report
- Exact start and failure timestamps with timezone, plus elapsed time to failure.
- Provider or client request ID, and session and turn IDs where applicable.
- Model, project, endpoint or transport, and SDK or package version.
- Timeout values for the client, proxy, model call, tool and overall run.
- HTTP status and error code, stream events or content chunks received, and whether the provider recorded a matching request.
- Affected P50, P90, P95 and P99 latency, the baseline, and error percentage—not raw error counts alone.
- Tool actions or other side effects that may have completed before disconnection.
For OpenAI incidents, its troubleshooting guidance specifically recommends filtering by project and model, reviewing HTTP request errors and latency, and including request IDs and timestamps with timezone when escalating. The Agents API observability material describes session and turn failure or cancellation and environment or error events. OpenAI troubleshooting guidance OpenAI Agents API reference
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
