Clawdbot + Ollama: WhatsApp & Telegram respond, but replies are extremely delayed
Iโm running Clawdbot 2026.1.24โ3 on a Hostinger KVM8 VPS (Ubuntu 24.04). Ollama is in Docker with the model qwen2.5:14b-instruct-q4_K_M as the primary agent model. Telegram and WhatsApp are both configured as channels and show as enabled/linked.
What works:
clawdbot channels status reports:
โข Telegram: enabled, configured, running, mode: polling, token from config.
โข WhatsApp: enabled, configured, linked, running, connected, inbound seen recently.
When I DM the Telegram bot or selfโchat the WhatsApp number, Clawdbot:
- Receives the message (I see inbound logs for both channels).
- Starts an embedded run with provider=ollama and model=qwen2.5:14b-instruct-q4_K_M.
- Eventually sends a reply back on both channels.
The problem:
โข Replies on both Telegram and WhatsApp are extremely delayed.
โข I see the typing indicator for a long time (or it times out after 2 minutes) before any text arrives.
โข Sometimes a reply shows up other times not
โข Sometimes I see log lines like:
Removed orphaned user message to prevent consecutive user turns
which suggests previous user turns are still considered โin progressโ and new ones get discarded.
During the same period, Ollama itself is fast:
Running this inside the container:
bash
time ollama run qwen2.5:14b-instruct-q4_K_M --verbose "Say hello in one short sentence."
-> returns in about 2.8 seconds total with:
-> load_duration โ 146 ms (warm, model already in RAM)
-> prompt_eval_rate โ 17.7 tokens/s
-> eval_rate โ 13.4 tokens/s
So the model server is not the bottleneck when warm and it is set to be constantly alive.
Channel configs (simplified):
WhatsApp:
dmPolicy: "allowlist"
selfChatMode: true
allowFrom: ["+1XXXXXXXXXX"] (the number I send from)
Linked via QR, inbound messages logged with correct from/to.
Telegram:
enabled: true
dmPolicy: "open"
allowFrom: ["*"]
groupPolicy: "open"
streamMode: "off"
Bot token matches the bot Iโm DMโing; provider starts with no errors.
Symptoms in logs when I send โHiโ:
Inbound line from WhatsApp or Telegram with my message.
embedded run start with the Ollama provider and messageChannel set appropriately.
No immediate โreply sentโ line; sometimes a typing TTL reached log after 2 minutes.
Much later, the actual reply finally appears in the chat.
What Iโm looking for:
Is there a known issue or configuration gotcha with embedded runs + local Ollama where runs stall or donโt signal completion promptly, causing long delays on chat channels?
Are there recommended settings (agent/session/typing config) or a way to forcefully reset a stuck session so new messages donโt get treated as โorphaned user messagesโ?
Any extra debug flags I can enable to see why the embedded run is hanging between โstartโ and โreply sentโ, given that direct ollama run calls are consistently fast?
What works:
clawdbot channels status reports:
โข Telegram: enabled, configured, running, mode: polling, token from config.
โข WhatsApp: enabled, configured, linked, running, connected, inbound seen recently.
When I DM the Telegram bot or selfโchat the WhatsApp number, Clawdbot:
- Receives the message (I see inbound logs for both channels).
- Starts an embedded run with provider=ollama and model=qwen2.5:14b-instruct-q4_K_M.
- Eventually sends a reply back on both channels.
The problem:
โข Replies on both Telegram and WhatsApp are extremely delayed.
โข I see the typing indicator for a long time (or it times out after 2 minutes) before any text arrives.
โข Sometimes a reply shows up other times not
โข Sometimes I see log lines like:
Removed orphaned user message to prevent consecutive user turns
which suggests previous user turns are still considered โin progressโ and new ones get discarded.
During the same period, Ollama itself is fast:
Running this inside the container:
bash
time ollama run qwen2.5:14b-instruct-q4_K_M --verbose "Say hello in one short sentence."
-> returns in about 2.8 seconds total with:
-> load_duration โ 146 ms (warm, model already in RAM)
-> prompt_eval_rate โ 17.7 tokens/s
-> eval_rate โ 13.4 tokens/s
So the model server is not the bottleneck when warm and it is set to be constantly alive.
Channel configs (simplified):
WhatsApp:
dmPolicy: "allowlist"
selfChatMode: true
allowFrom: ["+1XXXXXXXXXX"] (the number I send from)
Linked via QR, inbound messages logged with correct from/to.
Telegram:
enabled: true
dmPolicy: "open"
allowFrom: ["*"]
groupPolicy: "open"
streamMode: "off"
Bot token matches the bot Iโm DMโing; provider starts with no errors.
Symptoms in logs when I send โHiโ:
Inbound line from WhatsApp or Telegram with my message.
embedded run start with the Ollama provider and messageChannel set appropriately.
No immediate โreply sentโ line; sometimes a typing TTL reached log after 2 minutes.
Much later, the actual reply finally appears in the chat.
What Iโm looking for:
Is there a known issue or configuration gotcha with embedded runs + local Ollama where runs stall or donโt signal completion promptly, causing long delays on chat channels?
Are there recommended settings (agent/session/typing config) or a way to forcefully reset a stuck session so new messages donโt get treated as โorphaned user messagesโ?
Any extra debug flags I can enable to see why the embedded run is hanging between โstartโ and โreply sentโ, given that direct ollama run calls are consistently fast?
The AI that actually does things. Emails, calendar, home automation - all from your favorite chat app. New shell, ready to help. The lobster way.
134,426Members
Resources
Was this page helpful?
