"memory-lancedb-pro" does not support dreaming settings

[delphiduan]

Plugin Version

1.1.0-beta.10

OpenClaw Version

2026.4.8

Bug Description

Selected memory plugin "memory-lancedb-pro" does not support dreaming settings.

Expected Behavior

When I turn on the dreaming setting.

Steps to Reproduce

1.open the config web page.
2.turn on the “Dreaming” setting.
3.an error appears:
“Selected memory plugin 'memory-lancedb-pro' does not support dreaming settings.”

Error Logs / Screenshots

Embedding Provider

None

OS / Platform

No response

[jaliwey]

Feature Request: Native Dreaming Support in memory-lancedb-pro

Adding a more detailed feature request to complement the bug report above.

Background

memory-core ships a dreaming subsystem that runs scheduled memory consolidation sweeps across three phases:

Phase	Purpose	Key defaults
Light	Deduplication of recent short-term memories	lookback 2d, limit 100, dedupe similarity threshold
REM	Pattern detection across consolidated memories	lookback 7d, limit 10, minPatternStrength
Deep	Promotion of high-value memories to durable long-term store	minScore, minRecallCount ≥ 3, minUniqueQueries ≥ 3

Deep phase also includes a recovery sub-pass that rescues high-confidence memories that were previously filtered out.

The Problem

When memory-lancedb-pro is set as the active memory slot (plugins.slots.memory = "memory-lancedb-pro"), memory-core is automatically disabled. This means:

1. The dreaming subsystem never registers and never runs
2. Any plugins.entries.memory-core.config.dreaming config is silently ignored
3. OpenClaw doctor reports a warning: "plugin disabled but config is present"
4. Users who want both advanced hybrid retrieval (lancedb-pro) and scheduled memory consolidation (dreaming) have no path forward

Why This Matters for memory-lancedb-pro Specifically

memory-lancedb-pro already has a sophisticated retrieval pipeline (hybrid BM25 + vector, cross-encoder reranking, noise filtering, adaptive retrieval). Dreaming would complement this well:

• Light phase could deduplicate overlapping LanceDB entries using the existing vector similarity infrastructure
• Deep phase promotion logic could leverage lancedb-pro's existing scoring pipeline (minScore, recencyHalfLifeDays, etc.) — the config fields already overlap
• REM phase pattern detection could use BM25 + embedding search already built into the retriever

Requested Config Schema

Suggest mirroring memory-core's existing schema so users can migrate without relearning:

// plugins.entries.memory-lancedb-pro.config
{
  "dreaming": {
    "enabled": true,
    "frequency": "0 3 * * *",   // cron, default: daily at 3am
    "timezone": "Asia/Shanghai",
    "phases": {
      "light": {
        "enabled": true,
        "lookbackDays": 2,
        "limit": 100,
        "dedupeSimilarity": 0.92
      },
      "rem": {
        "enabled": true,
        "lookbackDays": 7,
        "limit": 10,
        "minPatternStrength": 0.6
      },
      "deep": {
        "enabled": true,
        "limit": 10,
        "minScore": 0.6,
        "minRecallCount": 3,
        "minUniqueQueries": 3,
        "recencyHalfLifeDays": 14,
        "maxAgeDays": 90
      }
    }
  }
}

And the /dreaming status / /dreaming on|off slash commands would be great to have as well.

Current Workaround

None. Users must choose between lancedb-pro's retrieval quality and memory-core's dreaming consolidation — they cannot have both simultaneously.

---

Happy to help test or provide more config details if useful. Thanks for the great plugin!

[vivimvi]

同样请求支持梦境配置，如果能组合使用就太酷了

[Xinslive]

memory_compact --dry-run=false --limit=1000

As far as I know, LanceDB-Pro employs a similar mechanism; you can try running this command to perform deduplication.

[Xinslive]

memory_compact is an MCP Tool (not a CLI command) within memory-lancedb-pro, designed to organize and compress the memory library.

Functionality: It archives duplicate, low-value memories and establishes links to a single "canonical memory," thereby reducing redundancy and keeping the memory library tidy.

Invocation: It is invoked directly via the AI ​​Agent tool table; the exec command is not required.

Parameters:

• dryRun (boolean, default: true) — Preview mode; does not execute actual changes.
• limit (number, default: 200) — The maximum number of memories to scan per run.
• scope (string, optional) — Specifies the scope of the operation.

Usage Example:

memory_compact(dryRun=true, limit=200)

First, perform a dry run to preview the results; once you have confirmed that everything looks correct, proceed with:

memory_compact(dryRun=false, limit=200)

Note: Although this tool is not explicitly listed in the "MCP Tools" table within the SKILL.md documentation, it is indeed included in the actual list of available tools. If you are using an older version of memory-lancedb-pro, you may need to upgrade the plugin before you can utilize this feature.