#telegram stopped working

16 messages · Page 1 of 1 (latest)

cunning plover
rare socketBOT
#

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

echo furnaceBOT
untold belfryBOT
#

Let me check the docs against that error so I can give you a clean fix.

#

I checked the gist. This is not a Telegram transport failure, the agent is dying before it can reply.

The key error is:

Embedded agent failed before reply: Sandbox mode requires Docker, but the "docker" command was not found in PATH

So the fix is:

  1. Please paste:
openclaw --version
  1. Then run:
openclaw doctor
# 
openclaw sandbox explain
  1. If you do not want sandboxing right now, the quickest fix is:
openclaw config set agents.defaults.sandbox.mode off
openclaw gateway restart
  1. If you do want sandboxing, install Docker and make sure the gateway service can see docker in its PATH, then restart the gateway.

One important nuance: your failing session is telegram:direct:..., so if you set agents.defaults.sandbox.mode: "non-main", that Telegram session can still be sandboxed.

Docs:
https://docs.openclaw.ai/cli/doctor
https://docs.openclaw.ai/gateway/sandboxing

cunning plover
untold belfryBOT
#

I’m checking the exact config path so I don’t send you on a goose chase.

#

Yep, then your config still has sandboxing enabled somewhere.

From the logs, OpenClaw is trying to sandbox the Telegram session, and since Docker is missing it crashes before replying.

Please run these and paste the output:

openclaw config get agents.defaults.sandbox.mode
openclaw config get agents.list
openclaw sandbox explain --session 'agent:main:telegram:direct:473602828'

Most likely one of these is true:

  • agents.defaults.sandbox.mode is non-main or all
  • your main agent has an override under agents.list[].sandbox.mode
#

If you want sandboxing fully off, the fix is usually:

openclaw config set agents.defaults.sandbox.mode off
openclaw gateway restart

But if agents.list shows an agent-specific override, that also needs to be set to off (agent-specific config wins).

Small gotcha from the docs: non-main is based on the session key, not just “groups”, so a Telegram direct session can still count as non-main and get sandboxed.

Relevant docs:
https://docs.openclaw.ai/gateway/sandboxing
https://docs.openclaw.ai/cli/sandbox
https://docs.openclaw.ai/help/faq

Paste those 3 outputs and I’ll point at the exact bad setting.

cunning plover
untold belfryBOT
#

Nice, we found it.

Your config has:

agents.defaults.sandbox.mode = non-main

and your main agent has no override, so Telegram direct sessions are being sandboxed. Since you do not have Docker, they fail.

You can fix it with:

openclaw config set agents.defaults.sandbox.mode off
openclaw gateway restart
#

Then verify with:

openclaw sandbox explain --session 'agent:main:telegram:direct:473602828'

It should change from:

  • runtime: sandboxed
  • mode: non-main

to:

  • runtime: direct
  • mode: off

Then test Telegram again.