openclaw - 💡(How to fix) Fix Active Memory plugin: all embedded sub-agent runs timeout with 0 summaryChars across 5 different providers [2 comments, 2 participants]

Describe the Bug

Active Memory plugin (introduced in v2026.4.10) consistently fails with timeout on every single embedded sub-agent run. The memory recall sub-agent never completes — all calls hang until the configured timeout expires, producing 0 characters of summary.

The same provider/model combinations work perfectly in normal chat sessions. This is specific to the Active Memory embedded sub-agent execution path.

Environment

Steps to Reproduce

1. Enable Active Memory plugin in openclaw.json
2. Set model to any valid provider/model
3. Restart gateway
4. Send a message to an eligible session (webchat direct or Telegram DM)
5. Active Memory triggers but always times out

Expected Behavior

Active Memory sub-agent should:

1. Call memory_search to find relevant memory entries
2. Call memory_get on relevant results
3. Generate a summary (maxSummaryChars chars)
4. Inject it into the main assistant's prompt
5. Complete within timeoutMs (15000ms)

Actual Behavior

Every embedded sub-agent run fails with:

• stage: "assistant"
• failoverReason: "timeout"
• summaryChars: 0
• Elapsed time: 15s to 95s (config timeoutMs=15000 but actual wall time is much longer)
• No memory search or get tool calls are ever completed

Providers/Models Tested (All Fail)

All five providers/models work perfectly in normal chat sessions. This is specific to the Active Memory embedded sub-agent execution path.

Log Evidence

Sample log entry (siliconflow/DeepSeek-V3):

active-memory: agent=auto session=agent:auto:main activeProvider=siliconflow activeModel=deepseek-ai/DeepSeek-V3 done status=timeout elapsedMs=17550 summaryChars=0

Failover decision log:

{
  "subsystem": "agent/embedded",
  "event": "embedded_run_failover_decision",
  "runId": "active-memory-mo9nll9o-6d4da698",
  "stage": "assistant",
  "decision": "surface_error",
  "failoverReason": "timeout",
  "profileFailureReason": "timeout",
  "provider": "siliconflow",
  "model": "deepseek-ai/DeepSeek-V3",
  "sourceProvider": "siliconflow",
  "sourceModel": "deepseek-ai/DeepSeek-V3",
  "fallbackConfigured": true,
  "timedOut": true,
  "aborted": true
}

Config change + restart log:

config change detected; evaluating reload (plugins.entries.active-memory.config.model)
config change requires gateway restart (plugins.entries.active-memory.config.model)
received SIGUSR1; restarting
ready (9 plugins: acpx, active-memory, browser, device-pair, memory-core, memory-wiki, phone-control, talk-voice, telegram; 21.7s)

Additional Context

• memorySearch (vector embedding via siliconflow BAAI/bge-m3) works fine — 16 agents indexed, ~107 MB
• Normal chat with all tested providers works fine
• Dreaming (memory-core plugin) works fine on schedule
• Only Active Memory's embedded sub-agent fails
• The actual wall-clock timeout far exceeds the configured timeoutMs: 15000 (up to 95 seconds observed), suggesting the timeout enforcement may also be buggy
• No HTTP error codes (4xx/5xx) are logged — the calls simply hang
• This occurs on both webchat (direct) and Telegram DM sessions
• The timeoutMs config value is set to 15000 but actual elapsed times range from 15821ms to 95436ms