Bug type
Regression (worked before, now fails)
Beta release blocker
No
Summary
After upgrading to OpenClaw 2026.4.14, response speed became noticeably worse in both console and Telegram sessions, even when queue depth was 0. Some runs ended with errorMessage: "terminated". Responsiveness improved dramatically when Active Memory was disabled.
Steps to reproduce
Run OpenClaw 2026.4.14 with Active Memory enabled.
Use the agent normally in console and/or Telegram.
Observe increased latency and, in some cases, runs ending with errorMessage: "terminated".
Disable plugins.entries.active-memory.enabled.
Repeat the same usage and observe that responsiveness improves materially.
Expected behavior
OpenClaw should remain responsive on 2026.4.14 with Active Memory enabled, without major latency or terminated runs during normal use.
Actual behavior
With Active Memory enabled, the agent became sticky/slow, Telegram replies were delayed, and some runs terminated. Disabling Active Memory made performance much better immediately. console interaction was very sluggish. some queries got queued. Some took 10-30 seconds. expected <5 second response.
OpenClaw version
OpenClaw 2026.4.14
Operating system
Ubuntu Linux 25.10
Install method
told openclaw to update to current version via telegram chat
Model
openai-codex/gpt-5.4 via OAuth
Provider / routing chain
openclaw on beelink SER5 to google provided wifi router to internet
Additional provider/model setup details
Primary model: openai-codex/gpt-5.4 via OAuth. Memory search enabled with local embeddings. Active Memory was enabled during the slowdown, then disabled for A/B comparison. No additional provider-specific setup seems necessary to reproduce.
Environment
OpenClaw 2026.4.14
Ubuntu Linux 25.10
Beelink SER5 mini PC
16 GB RAM
Primary model: openai-codex/gpt-5.4 via OAuth
Memory search enabled with local embeddings
Active Memory enabled before test, then disabled for A/B comparison
Logs, screenshots, and evidence
Observed tuning that helped somewhat before full disable: queryMode message -> recent, promptStyle strict -> balanced, timeoutMs 3000 -> 1500, maxSummaryChars 220 -> 120, logging true -> false, cacheTtlMs 60000 -> 120000. Also reduced memory search maxResults 8 -> 5 and candidateMultiplier 4 -> 2. Related issues: #65159 #65441 #66060 #63526. If helpful, I can provide screenshots/log snippets showing delayed responses and errorMessage: "terminated".
Impact and severity
Affected: local console usage and Telegram sessions on this install. Severity: High, because it makes normal interaction noticeably slower and some runs terminate. Frequency: frequent while Active Memory was enabled, with clear improvement after disabling it. Consequence: degraded responsiveness, delayed replies, and reduced confidence in session stability.
Additional information
openclaw-2026-4-14-active-memory-bug-report.md
Bug type
Regression (worked before, now fails)
Beta release blocker
No
Summary
After upgrading to OpenClaw 2026.4.14, response speed became noticeably worse in both console and Telegram sessions, even when queue depth was 0. Some runs ended with errorMessage: "terminated". Responsiveness improved dramatically when Active Memory was disabled.
Steps to reproduce
Run OpenClaw 2026.4.14 with Active Memory enabled.
Use the agent normally in console and/or Telegram.
Observe increased latency and, in some cases, runs ending with errorMessage: "terminated".
Disable plugins.entries.active-memory.enabled.
Repeat the same usage and observe that responsiveness improves materially.
Expected behavior
OpenClaw should remain responsive on 2026.4.14 with Active Memory enabled, without major latency or terminated runs during normal use.
Actual behavior
With Active Memory enabled, the agent became sticky/slow, Telegram replies were delayed, and some runs terminated. Disabling Active Memory made performance much better immediately. console interaction was very sluggish. some queries got queued. Some took 10-30 seconds. expected <5 second response.
OpenClaw version
OpenClaw 2026.4.14
Operating system
Ubuntu Linux 25.10
Install method
told openclaw to update to current version via telegram chat
Model
openai-codex/gpt-5.4 via OAuth
Provider / routing chain
openclaw on beelink SER5 to google provided wifi router to internet
Additional provider/model setup details
Primary model: openai-codex/gpt-5.4 via OAuth. Memory search enabled with local embeddings. Active Memory was enabled during the slowdown, then disabled for A/B comparison. No additional provider-specific setup seems necessary to reproduce.
Environment
Logs, screenshots, and evidence
Impact and severity
Affected: local console usage and Telegram sessions on this install. Severity: High, because it makes normal interaction noticeably slower and some runs terminate. Frequency: frequent while Active Memory was enabled, with clear improvement after disabling it. Consequence: degraded responsiveness, delayed replies, and reduced confidence in session stability.
Additional information
openclaw-2026-4-14-active-memory-bug-report.md