OpenClaw feels “non-proactive” + agent and subagents stall after ~1 minute
I’m running OpenClaw on a Proxmox LXC and I’m not seeing the same “proactive / keeps working until done” behavior I’m seeing in demos and posts. I know the project is young and bug-prone, but I’m trying to figure out whether this is misconfiguration / environment limits on my side or a known issue that needs a fix.
What’s going wrong (symptoms)
Not proactive like the examples: it often waits for explicit prompts rather than pushing progress updates.
When I ask it to do a task or spin up subagents, they appear to “work” for ~1 minute, then progress stops.
SSH keys don’t persist: I provide SSH keys for server access, but after a new session it behaves as if it never had them.
Webhook/plugin creation often crashes the gateway, sometimes requiring openclaw doctor --fix or manual repair (and doctor doesn’t always recover it).
Even on small tasks (like creating a webhook), it will:
work briefly,
then stall,
and when I ask “status?”, it repeats the same update from minutes ago as if no work happened.
Environment
Host: Proxmox VE 8.4.14
Guest: Ubuntu 24.04 LTS (LXC)
CPU allocated: 16 threads (Intel Xeon E5-2680 v3 @ 2.50GHz)
RAM allocated: 16GB (2133 MHz)
OpenClaw version: 2026.2.2-2
Model: openai/gpt-5
Skills/tools: brave-search
Memory: qmd backend + sessions enabled
Channels tested: Telegram + built-in web GUI
What I’ve tried already
Scraped docs.openclaw.ai and converted to markdown in a local docs cache to help it generate valid configs (it frequently breaks config structure otherwise).
Multiple OpenClaw versions + channels (stable/beta).
Multiple platforms (Telegram + web GUI).
Asked the model to self-diagnose / check for issues.
Config (sanitized)
I’m including my sanitized openclaw.json below. Key points:
memorySearch enabled (sources: memory + sessions)
qmd backend for memory
maxConcurrent: 4 / subagents.maxConcurrent: 8
gateway TLS enabled (autoGenerate)
telegram streamMode: partial
hooks enabled
What’s going wrong (symptoms)
Not proactive like the examples: it often waits for explicit prompts rather than pushing progress updates.
When I ask it to do a task or spin up subagents, they appear to “work” for ~1 minute, then progress stops.
SSH keys don’t persist: I provide SSH keys for server access, but after a new session it behaves as if it never had them.
Webhook/plugin creation often crashes the gateway, sometimes requiring openclaw doctor --fix or manual repair (and doctor doesn’t always recover it).
Even on small tasks (like creating a webhook), it will:
work briefly,
then stall,
and when I ask “status?”, it repeats the same update from minutes ago as if no work happened.
Environment
Host: Proxmox VE 8.4.14
Guest: Ubuntu 24.04 LTS (LXC)
CPU allocated: 16 threads (Intel Xeon E5-2680 v3 @ 2.50GHz)
RAM allocated: 16GB (2133 MHz)
OpenClaw version: 2026.2.2-2
Model: openai/gpt-5
Skills/tools: brave-search
Memory: qmd backend + sessions enabled
Channels tested: Telegram + built-in web GUI
What I’ve tried already
Scraped docs.openclaw.ai and converted to markdown in a local docs cache to help it generate valid configs (it frequently breaks config structure otherwise).
Multiple OpenClaw versions + channels (stable/beta).
Multiple platforms (Telegram + web GUI).
Asked the model to self-diagnose / check for issues.
Config (sanitized)
I’m including my sanitized openclaw.json below. Key points:
memorySearch enabled (sources: memory + sessions)
qmd backend for memory
maxConcurrent: 4 / subagents.maxConcurrent: 8
gateway TLS enabled (autoGenerate)
telegram streamMode: partial
hooks enabled
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
83,815Members
Resources
Similar Threads
Was this page helpful?
