Feature request: Memory retention policy and table optimization

[slj130]

Problem

memory-lancedb-pro's LanceDB storage grows indefinitely with no built-in way to manage data lifecycle. Currently at 127MB after 20 days of use with 4 agents and autoCapture enabled. While manageable now, there's no mechanism to prevent unbounded growth over months of operation.

Missing capabilities:

1. No retention policy — No way to auto-expire old, low-importance memories after N days
2. No bulk prune — memory_forget works on individual memories (by ID or search query), but there's no way to batch-delete by age, importance threshold, or scope
3. No table.optimize() exposure — LanceDB's native table.optimize({ cleanupOlderThan }) is never called, so deleted data versions accumulate on disk (Lance MVCC). After many memory_forget calls, disk usage won't decrease without manual optimization.

Proposed solution

1. memory_prune tool

A new agent-facing tool for batch cleanup:

memory_prune({
  olderThanDays: 90,           // Delete memories created before N days ago
  maxImportance: 0.3,          // Only prune low-importance memories
  scope: "global",             // Target scope
  dryRun: true                 // Preview before executing (default: true)
})

2. memory_optimize tool (or automatic)

Expose LanceDB's table.optimize() to reclaim disk space after deletions:

memory_optimize({
  cleanupOlderThan: "7d"      // Clean up Lance versions older than 7 days
})

Alternatively, auto-run table.optimize() after a threshold number of deletes.

3. Optional: config-based retention policy

{
  "retention": {
    "enabled": true,
    "maxAgeDays": 180,
    "minImportance": 0.3,
    "optimizeAfterPrune": true
  }
}

Memories below minImportance and older than maxAgeDays would be automatically pruned on a configurable schedule (e.g. weekly).

Context

• The Weibull decay model already tracks memory aging via access_count and last_accessed_at, so the infrastructure for importance-based expiry partially exists
• LanceDB JS SDK exposes table.optimize({ cleanupOlderThan: Date }) which handles compact + cleanup in one call
• The plugin already imports @lancedb/lancedb and has full table access in store.ts

Environment

• memory-lancedb-pro v1.1.0-beta.10
• OpenClaw 2026.3.23-2
• LanceDB @lancedb/lancedb (bundled)
• Jina embeddings + reranker
• 4 agents, autoCapture + smartExtraction enabled
• macOS arm64

[AliceLJY]

Well-structured feature request. The three proposed capabilities are all valid:

1. memory_prune — batch cleanup by age + importance threshold
2. memory_optimize — expose LanceDB's table.optimize() to reclaim disk after deletes
3. Config-based retention — automatic lifecycle management

table.optimize() exposure is the lowest-hanging fruit — LanceDB MVCC cleanup is genuinely needed after bulk memory_forget calls. We'll track this as an enhancement. PRs welcome if you'd like to tackle any of the three.