My agent has stopped doing background tasks.

[neat zephyr]

My agent has stopped doing the take I tell it to do in the background. It just responds and says it will do it but never completes or does it anymore.

• What you’re trying to do (goal / expected behaviour)
  I want to give it tasks and have it complete them either immediately or in the background.

• What happened instead (exact error message)
  I just sends a response saying I'm doing this or that and ends there.

• What you ran or clicked (commands, config snippet, etc.)
  It's been building things up until the point where it just stopped. I was last building a heartbeat task that prompts it to complete tasks in the background every 10 minutes.

• Your environment (OS, install method, versions)
  OS / access context: Ubuntu 22.04.5 LTS on kernel 6.8.0-100-generic (likely a local workstation).

OpenClaw 2026.4.12 (1c0672b)

Model: openai-codex/gpt-5.4

Fallbacks: nvidia-qwen/qwen/

oauth (openai-codex:[e-mail]

([e-mail]))

qwen3-coder-480b-a35b-instruct, openai/ gpt-5.4, nvidia-kimi/moonshotai/kimi-k2.5, openrouter/qwen/qwen3-coder:free, google/gemini-1.5-pro

Fallback: nvidia-qwen/qwen/

qwen3-coder-480b-a35b-instruct

api-key (models.json) (rate limit)

Tokens: 1.1m in / 1.9k out

$0.0000

Cost:

Cache: 96% hit 22.8m cached, 0 new

Context: 1.1m/262k (409%)

Compactions: 2

Usage: 5h 100% left

0% left 2d Oh

Session:

4h 59m Week

agent:main:telegram

updated just now

Runtime: direct Think: off Queue: collect (depth 0)

[knotty rose]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[vivid hemlock]

Two things jump out here:

1. Heartbeat is not a detached worker
   Heartbeat is just a periodic main-session turn. It does not create background task records by itself. Real detached work shows up in openclaw tasks.
   Docs: https://docs.openclaw.ai/gateway/heartbeat and https://docs.openclaw.ai/automation/tasks

2. Your session looks badly over budget
   Context: 1.1m/262k (409%) is a huge red flag. Even with compaction, once a thread gets this bloated, the model can start doing the annoying “I’ll do it” thing without actually following through.

What I’d try first

In the broken session:

openclaw tasks list
openclaw tasks audit

openclaw system heartbeat last
openclaw logs --follow

And in chat, start fresh:

/new

If you want to keep the thread first, try:

/compact

What to look for in logs

Especially these:

• heartbeat skipped with reason=requests-in-flight
• heartbeat skipped with reason=empty-heartbeat-file
• heartbeat skipped with reason=no-tasks-due
• stale/lost tasks

My best guess

This is mostly a session/context problem, and possibly also a heartbeat misunderstanding:

• Heartbeat = periodic check-in on the main session
• Cron / subagents / ACP = actual detached background work

So if your goal is “run this every 10 minutes and complete work in the background”, cron is usually the better fit than heartbeat.