DAG Lossless Memory

Added in SoulClaw v2026.3.18

The DAG (Directed Acyclic Graph) Memory provides lossless conversation persistence for Soul-powered agents. Every message exchanged between the user and agent is stored permanently in a local SQLite database, ensuring no information is ever lost to context window pruning or compaction.

The Problem

AI agents forget. Context windows are finite, and when conversations exceed the limit, older messages are pruned — silently, permanently. Summarization-based pruning loses nuance. Hard truncation loses everything.

How It Works

The DAG store hooks into the agent lifecycle and runs automatically after each conversation turn:

Level 0: Raw messages (never deleted)
   ↓ Every 10 turns
Level 1: Chunk summaries (LLM-generated)
   ↓ Every 10 summaries  
Level 2+: Higher-level summaries (recursive)

1. Raw storage (Level 0): Every user and assistant message is stored as-is
2. Chunk summarization (Level 1): After 10 accumulated turns, the LLM generates a concise summary
3. Higher-level summaries (Level 2+): Summaries are recursively rolled up for long-running sessions

The DAG structure means you can always trace back from a summary to the original messages that produced it. Nothing is lost.

3-Tier Memory Architecture

DAG Memory is part of SoulClaw's 3-tier memory system:

Tier	Component	Purpose	Storage
1	DAG Store	Lossless conversation persistence + FTS5 search	.dag-memory.sqlite
2	Vector Search	Semantic similarity search over memory files	memory-index.db
3	Passive Memory	Automatic extraction of important facts	memory/*.md files

All three tiers work together:

• DAG ensures nothing is lost
• Passive Memory extracts signal from noise
• Vector Search provides semantic retrieval

Key Design Decisions

• SQLite — Zero-dependency, single-file database. No external services required.
• FTS5 full-text search — Instant keyword search across all conversation history.
• Incremental storage — Only new messages are written per turn; no duplicates.
• Level-based hierarchy — Raw messages at level 0, progressive summarization at higher levels.

Storage

The database is stored at:

<workspace>/.dag-memory.sqlite

Typical storage: ~50 conversation turns ≈ 300KB.

Activation

The DAG store activates automatically when memorySearch is configured in SoulClaw. No additional setup required:

{
  "agents": {
    "defaults": {
      "memorySearch": {
        "provider": "local"
      }
    }
  }
}

If memorySearch is not configured, the DAG store remains inactive and no SQLite file is created.

Retrieval

When memory_search is called, results come from all three tiers:

1. FTS5 searches the DAG for exact keyword matches across all history
2. Semantic search finds conceptually related memories from indexed files
3. Passive memories provide curated, high-signal facts

The combination gives you both precision (exact matches from FTS5) and recall (conceptual matches from vectors).

Inspecting the DAG

You can query the DAG database directly:

# Count stored nodes by level
sqlite3 .dag-memory.sqlite \
  "SELECT level, COUNT(*) FROM dag_nodes GROUP BY level"

# Search conversation history
sqlite3 .dag-memory.sqlite \
  "SELECT content FROM dag_fts WHERE dag_fts MATCH 'deployment' LIMIT 5"

# Recent messages
sqlite3 .dag-memory.sqlite \
  "SELECT role, substr(content,1,80) FROM dag_nodes \
   WHERE level=0 ORDER BY created_at DESC LIMIT 10"

Getting Started

npm install -g soulclaw
soulclaw onboard

The 3-tier memory system activates automatically once you configure memory search during onboarding.