openclaw - 💡(How to fix) Fix [Bug]: Active-memory on multi-agent Telegram gateway is inconsistent (ok / empty / timeout) and not safe to enable globally [1 comments, 2 participants]

Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

Summary

On a multi-agent OpenClaw gateway using Telegram direct chats, the active-memory plugin appears partially functional but unreliable when enabled: across real agent sessions it produces a mix of status=ok, status=empty, and status=timeout, making it difficult to trust for production use.

Steps to reproduce

Steps to reproduce

1. Run OpenClaw 2026.4.26 on a multi-agent gateway.
2. Load the official active-memory plugin.
3. Configure it for multiple agents and direct chats.
4. Use direct Telegram conversations with those agents.
5. Observe gateway logs for active-memory activity across several sessions/agents.

Expected behavior

Expected behavior

When active-memory is enabled for eligible sessions, it should behave consistently:

• either reliably return useful summaries when enough context exists • or fail in a predictable and clearly explained way

It should not frequently alternate between:

• successful summaries • empty summaries • timeouts

under the same deployment.

Actual behavior

Actual behavior

Observed results across multiple agents/sessions include:

• status=ok with non-zero summaryChars • status=empty with summaryChars=0 • status=timeout

This makes it unclear whether active-memory is healthy enough to enable across all agents in production.

OpenClaw version

2026.4.26 (be8c246)

Operating system

macOS arm64

Install method

npm global

Model

anthropic/claude-sonnet-4-6

Provider / routing chain

openclaw -> anthropic

Additional provider/model setup details

{ "plugins": { "entries": { "active-memory": { "enabled": true, "config": { "enabled": false, "agents": [ "agent-a", "agent-b", "agent-c", "agent-d" ], "allowedChatTypes": ["direct"], "logging": true, "maxSummaryChars": 220, "modelFallback": "anthropic/claude-haiku-4-5", "persistTranscripts": false, "promptStyle": "balanced", "queryMode": "recent", "timeoutMs": 15000 } } } } }

Notes:

• The plugin is loaded at gateway level. • The current config appears to be in a mixed state (enabled: true at plugin entry level, config.enabled: false inside plugin config). • It is unclear whether this state is expected/supported or whether it contributes to inconsistent active-memory behavior. • Models were intentionally left unchanged during testing to avoid confounding variab

Logs, screenshots, and evidence

Representative log patterns:
[plugins] active-memory: agent=agent-a session=agent:agent-a:main activeProvider=anthropic activeModel=claude-sonnet-4-6 done status=ok elapsedMs=14078 summaryChars=217

[plugins] active-memory: agent=agent-a session=agent:agent-a:main activeProvider=anthropic activeModel=claude-sonnet-4-6 done status=empty elapsedMs=8431 summaryChars=0

[plugins] active-memory: agent=agent-b session=agent:agent-b:main activeProvider=anthropic activeModel=claude-sonnet-4-6 done status=empty elapsedMs=11685 summaryChars=0

[plugins] active-memory: agent=agent-b session=agent:agent-b:main activeProvider=anthropic activeModel=claude-sonnet-4-6 done status=ok elapsedMs=12255 summaryChars=218

[plugins] active-memory: agent=agent-b session=agent:agent-b:main activeProvider=anthropic activeModel=claude-sonnet-4-6 done status=timeout elapsedMs=15024 summaryChars=0

[plugins] active-memory: agent=agent-c session=agent:agent-c:main activeProvider=anthropic activeModel=claude-sonnet-4-6 done status=empty elapsedMs=10525 summaryChars=0



Additional observations:

• This does not appear to be a file corruption problem.
• Manual memory/search functionality appears healthier than active-memory.
• The gateway itself remains otherwise healthy when active-memory is disabled.

Impact and severity

Impact and severity

Affected: operators using multi-agent gateways with Telegram/direct-chat flows Severity: Medium to High Frequency: Intermittent but repeatedly observed across multiple agents/sessions Consequence:

• unreliable memory enrichment • extra latency before replies • difficulty trusting active-memory in production • operational uncertainty around whether the feature is actually usable at scale

Additional information

Additional information

This seems related to, but not necessarily fully covered by, the following existing issues:

• #72015 — Reliability: active-memory blocks replies and QMD boot initialization can overload multi-agent gateways • #73306 — Active Memory plugin times out on every run • #74586 — AM embedded run aborts memory_search tool calls; classifies as timeout despite model completion • #65309 — Active Memory blocks direct-chat turns and times out • #70061 — Telegram DM session classification edge case (possibly adjacent, but likely not the main cause here)

Questions for maintainers:

1. Is the mixed state below expected/supported? • plugin entry: enabled: true • plugin config: config.enabled: false
2. Are mixed outcomes (ok, empty, timeout) currently expected for active-memory on multi-agent gateways, or does this indicate a known bug?
3. Do any of the in-flight fixes/PRs already address this exact mixed-behavior case?
4. What is the recommended safe validation path before enabling active-memory across all agents in production?