#Extreme slowness after upgrade from v2026.4.22 to v2026.4.29

1 messages · Page 1 of 1 (latest)

lime stag
#

[Help] Extreme slowness after upgrade from v2026.4.22 to v2026.4.29 — embedded-run pipeline ~110s overhead per message
Hey team! After upgrading from v2026.4.22 to v2026.4.29 (Docker, self-hosted), every single Telegram message now takes 2-3 minutes instead of ~7 seconds. The actual LLM response (MiniMax via OpenAI-compatible proxy) only takes 5-12s — the problem is the internal embedded-run pipeline.

Trace for every message:

startup stages: totalMs=30135
model-resolution:3685ms, auth:12354ms, attempt-dispatch:14094ms
prep stages: totalMs=76303
core-plugin-tools:14339ms, bundle-tools:10142ms,
system-prompt:24817ms, stream-setup:24586ms
[fetch-timeout] fetch timeout reached; aborting operation (x2)
Setup:

Docker on VPS (4 vCPU, 8GB RAM)
Model via OpenAI-compatible proxy (OmniRoute → MiniMax-M2.7)
30 tools, contextTokens: 50000, timeoutSeconds: 300
Plugins: telegram, discord, browser, diagnostics-otel
Questions:

Why does auth take 12s and system-prompt 25s on every message? This was instant on v2026.4.22
Can I disable the slug-generator embedded agent or point it to a lighter model?
Is there a known performance regression in the v2026.4.27+ embedded-run pipeline?
Any help appreciated! 🦞

sinful ingotBOT
#

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

fallow beacon
#

YUp. Same here. I counted 1 min 20s from when I send a chat in the WebUI til LMStudio registers anything from Openclaw. Some sort of huge latency issue.