Skip to content

Commit 0e9d13d

Browse files
committed
Update to 2026.4.5: Add Part 22 Built-In Dreaming (memory-core), cross-reference Part 16 custom autoDream as prototype
1 parent cb4e21e commit 0e9d13d

2 files changed

Lines changed: 139 additions & 1 deletion

File tree

README.md

Lines changed: 137 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,6 @@
11
# OpenClaw Optimization Guide
22

3-
> **Tested on OpenClaw 2026.4.2 — April 3, 2026** · 21 parts · Battle-tested on a 14+ agent production deployment
3+
> **Tested on OpenClaw 2026.4.5 — April 7, 2026** · 22 parts · Battle-tested on a 14+ agent production deployment
44
55
### Make Your OpenClaw AI Agent Faster, Smarter, and Actually Useful
66
#### Speed optimization, memory architecture, context management, model selection, graph RAG, codebase intelligence, and agent observability
@@ -32,6 +32,7 @@
3232
19. [Repowise — Codebase Intelligence](./part19-repowise-codebase-intelligence.md) - 60% fewer tokens, 4x faster coding agents. Dependency graphs, git analytics, dead code detection, architectural decisions.
3333
20. [Agent Observability](./part20-observability-and-services.md) - LangFuse for tracing all agent calls. n8n for workflow automation. Reranker for better search quality.
3434
21. [Real-Time Knowledge Sync](./part21-realtime-knowledge-sync.md) - Event-driven file watcher that syncs vault changes to LightRAG in under 6 seconds. No cron. Self-healing offline queue. The final piece that makes the knowledge graph always current.
35+
22. [Built-In Dreaming (memory-core)](#part-22-built-in-dreaming) - OpenClaw's official memory consolidation system. 3-phase sweep: Light, Deep, REM. Automatic promotion to MEMORY.md. The native replacement for our custom autoDream.
3536

3637
**📊 [Benchmarks](./benchmarks/)** — Real numbers from a production system (context savings, search latency, reindex results, SWE-bench rankings)
3738

@@ -1391,6 +1392,141 @@ Start with the basic vault system (Parts 4, 9). It works well up to ~500 files.
13911392

13921393
---
13931394

1395+
## Part 22: Built-In Dreaming
1396+
1397+
> **OpenClaw 2026.4+** — The official memory consolidation system built into memory-core. This is the native version of what our custom autoDream (Part 16) prototyped.
1398+
1399+
### What Changed
1400+
1401+
Our [autoDream implementation (Part 16)](./part16-autodream-memory-consolidation.md) was a custom AGENTS.md-driven workaround we built by reverse-engineering Claude Code's memory consolidation pattern. It worked, but it required you to set up triggers, state tracking, and consolidation logic manually.
1402+
1403+
OpenClaw 2026.4+ ships a **built-in dreaming system** inside memory-core that does all of this natively. If you're on 2026.4+, use the built-in version instead.
1404+
1405+
### How It Works
1406+
1407+
Dreaming runs a 3-phase sweep on a cron schedule (default: 3am daily):
1408+
1409+
| Phase | What It Does | Writes to MEMORY.md? |
1410+
|-------|-------------|---------------------|
1411+
| **Light** | Sorts and stages recent short-term material. Dedupes daily signals. | No |
1412+
| **Deep** | Scores candidates and promotes durable entries to MEMORY.md using weighted ranking. | Yes |
1413+
| **REM** | Extracts themes and reflective patterns. Builds reflection summaries. | No |
1414+
1415+
Plus a **Dream Diary** — a human-readable narrative entry in DREAMS.md written by a background subagent turn after each sweep.
1416+
1417+
### Deep Ranking Signals
1418+
1419+
Deep phase decides what becomes long-term memory using 6 weighted signals:
1420+
1421+
| Signal | Weight | What It Measures |
1422+
|--------|--------|-----------------|
1423+
| Frequency | 0.24 | How many short-term signals the entry accumulated |
1424+
| Relevance | 0.30 | Average retrieval quality for the entry |
1425+
| Query diversity | 0.15 | Distinct query/day contexts that surfaced it |
1426+
| Recency | 0.15 | Time-decayed freshness score |
1427+
| Consolidation | 0.10 | Multi-day recurrence strength |
1428+
| Conceptual richness | 0.06 | Concept-tag density from snippet/path |
1429+
1430+
Entries must pass `minScore`, `minRecallCount`, and `minUniqueQueries` thresholds to be promoted.
1431+
1432+
### Quick Start
1433+
1434+
```json
1435+
{
1436+
"plugins": {
1437+
"entries": {
1438+
"memory-core": {
1439+
"config": {
1440+
"dreaming": {
1441+
"enabled": true
1442+
}
1443+
}
1444+
}
1445+
}
1446+
}
1447+
}
1448+
```
1449+
1450+
### Custom Cadence
1451+
1452+
```json
1453+
{
1454+
"plugins": {
1455+
"entries": {
1456+
"memory-core": {
1457+
"config": {
1458+
"dreaming": {
1459+
"enabled": true,
1460+
"timezone": "America/New_York",
1461+
"frequency": "0 */6 * * *"
1462+
}
1463+
}
1464+
}
1465+
}
1466+
}
1467+
}
1468+
```
1469+
1470+
### Slash Commands
1471+
1472+
```
1473+
/dreaming status — Check current state, last run, next scheduled run
1474+
/dreaming on — Enable dreaming
1475+
/dreaming off — Disable dreaming
1476+
/dreaming help — Show usage info
1477+
```
1478+
1479+
### CLI Commands
1480+
1481+
```bash
1482+
# Preview what would be promoted (dry run)
1483+
openclaw memory promote
1484+
1485+
# Apply promotions to MEMORY.md
1486+
openclaw memory promote --apply
1487+
1488+
# Limit to top N candidates
1489+
openclaw memory promote --limit 5
1490+
1491+
# Explain why a specific candidate would/wouldn't promote
1492+
openclaw memory promote-explain "router vlan"
1493+
1494+
# Preview REM reflections without writing
1495+
openclaw memory rem-harness
1496+
1497+
# JSON output for scripting
1498+
openclaw memory promote --json
1499+
openclaw memory rem-harness --json
1500+
```
1501+
1502+
### What Gets Written
1503+
1504+
| Location | Content |
1505+
|----------|---------|
1506+
| `memory/.dreams/` | Machine state — recall store, phase signals, ingestion checkpoints, locks |
1507+
| `DREAMS.md` | Human-readable Dream Diary entries + phase summaries |
1508+
| `memory/dreaming/<phase>/YYYY-MM-DD.md` | Optional phase reports |
1509+
| `MEMORY.md` | Long-term promoted entries (Deep phase only) |
1510+
1511+
### Dreams UI
1512+
1513+
The Gateway Dreams tab shows:
1514+
- Current enabled/disabled state
1515+
- Phase-level status and managed-sweep presence
1516+
- Short-term, long-term, and promoted-today counts
1517+
- Next scheduled run timing
1518+
- Expandable Dream Diary reader
1519+
1520+
### Should You Use This or Part 16?
1521+
1522+
**Use Part 22 (built-in dreaming)** if you're on OpenClaw 2026.4+. It's the official, supported implementation with proper scoring, phase management, and a UI.
1523+
1524+
**Use Part 16 (custom autoDream)** only if you're on an older version or need custom consolidation logic that the built-in system doesn't support.
1525+
1526+
*Source: [OpenClaw Dreaming Documentation](https://docs.openclaw.ai/concepts/dreaming)*
1527+
1528+
---
1529+
13941530
## About
13951531

13961532
*Built by [Terp - Terp AI Labs](https://x.com/OnlyTerp)*

part16-autodream-memory-consolidation.md

Lines changed: 2 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,5 +1,7 @@
11
# Part 16: autoDream — Automatic Memory Consolidation
22

3+
> **Note:** OpenClaw 2026.4+ has a [built-in dreaming system (Part 22)](./README.md#part-22-built-in-dreaming) inside memory-core. If you're on 2026.4+, use that instead — it's the official, supported version with proper scoring, phase management, and a UI. This part is kept for older versions and custom implementations.
4+
35
*Your agent's memory grows forever. Session files pile up. MEMORY.md gets stale. Knowledge gets buried. autoDream fixes this by teaching your agent to consolidate its own memory — automatically, on every session start.*
46

57
---

0 commit comments

Comments
 (0)