Codex: Error Running Remote Compact Task
When a Codex conversation gets close to the model's context limit, Codex compacts it: it sends the history to OpenAI's servers to be summarized so the thread can keep going. 'Error running remote compact task' means that summarizing step failed. The part after the colon gives the reason. 'stream disconnected before completion' (often with 'error decoding response body' or 'error sending request for url ... /codex/responses/compact') is a connection that broke during compaction; 'request timed out' or 'timeout' means compaction took too long, which is common on very large threads; 'We're currently experiencing high demand' is OpenAI capacity; and messages like 'unexpected status 404' or 'Unknown parameter' have appeared after client and server versions drifted apart. Because compaction runs exactly when a thread is at its largest, retrying the same compaction often fails again. The dependable fix is to save a short checkpoint, start a fresh thread, and carry over only what the next step needs.
Affiliate link, we may earn a commission at no extra cost to you.
Symptoms You Might See
Codex stops with 'Error running remote compact task' after a long session
The message continues with 'stream disconnected before completion' or 'error sending request for url'
Auto-compaction fails after a long wait, sometimes around two and a half minutes
Running /compact by hand fails the same way
A context-window message ('Codex ran out of room') appears alongside the compact error
Common Causes
The thread is too large to summarize reliably
Compaction has to send and process the whole long history. On very large threads this is slow and is the step most likely to time out or drop, which is why the error shows up at the end of long sessions.
A network break during compaction
The compaction request is a network call to OpenAI's compact endpoint. Proxies, VPNs, unstable Wi-Fi or an OpenAI incident can break it in the same way as a normal Codex stream.
OpenAI capacity or incidents
During high demand the compact task can be rejected with a high-demand message or fail with server errors. These are fixed on OpenAI's side.
Client and server versions out of step
Several reports (404 Not Found, Unknown parameter) were tied to specific Codex app or CLI versions. Updating usually clears these.
Step-by-Step Fixes
Save a checkpoint, then start a new thread
When to try: First, when compaction fails on a long thread
Write down the goal, what is done, what remains and the files that matter (ask Codex for a short handoff if it still responds). Run /new in the CLI or open a new chat in the app, paste the checkpoint, and ask it to inspect only the files it needs. This is the fix SmartScope and many GitHub reporters land on, because it avoids compacting the oversized history at all.
Check the working tree before continuing
When to try: Every time you move to a new thread
Starting a new thread does not undo file changes. Run git status and git diff --stat so the new thread does not repeat edits that already landed.
Retry /compact once, earlier next time
When to try: When you want to keep the same thread
If the thread still responds and continuity matters, try /compact once. Running it before the thread is at its limit gives compaction less to process. If it fails again, do not keep retrying the same compaction; move to a new thread.
Check status.openai.com
When to try: When the error appeared suddenly across threads
If the error mentions high demand, timeouts or server errors, look for a Codex incident. During incidents, waiting is faster than retrying.
Update Codex
When to try: When the error mentions 404 or an unknown parameter
Update the CLI, IDE extension or desktop app and restart it. Version mismatches have caused 404 and unknown-parameter compact errors that disappear after updating.
Check your network path
When to try: When compaction fails on a stable thread size
If the error says 'error sending request for url', try without VPN or proxy, or on another network, to rule out something blocking the compact request.
How to Prevent This
Use one thread per task and start fresh when the task changes
Name the relevant files instead of pasting whole logs into the chat
Keep reusable project guidance in AGENTS.md rather than repeating it in chat
Compact manually before a thread is near its limit, not after
When to Contact Support
Report it when compaction fails on modestly sized threads, fails on every thread, or keeps failing with no incident on the status page. Use /feedback to attach logs and file an issue on github.com/openai/codex with your version, platform and the exact error text.
Frequently Asked Questions
What is the remote compact task in Codex?▼
It is the step where Codex sends a long conversation to OpenAI to be summarized so the thread fits back inside the model's context window. The error means that summarizing call failed.
Does codex resume fix a failed compaction?▼
No. codex resume reopens a previous conversation with its full history, so it does not make the thread smaller. To get past a failed compaction, start a new thread with a short checkpoint.
Will I lose my work if compaction fails?▼
Your files stay as they are. What you lose is the conversation's working memory, which is why you write a short checkpoint before starting the new thread.
Why does it fail after about two and a half minutes?▼
Users on the Codex issue tracker report auto-compaction timing out after roughly 150 seconds on large threads. A timeout means the history was too large or the connection too slow to finish summarizing in time.
How do I avoid it?▼
Keep threads focused on one task, start a new thread when the task changes, and compact before the thread is huge. Smaller threads compact faster and fail less.
Sources and verification
Product behavior and limits can change. These primary sources were used to verify this guide.
Other Common AI Errors
ChatGPT Not Working
Service Issues
🚦Rate Limit Exceeded
Usage Limits
📏Context Length Exceeded
Input Limits
🔥Model Overloaded
Service Issues