memory-bank

Memory Bank

Conversations → Knowledge Graph. Claude Code conversations become searchable, structured knowledge.

What's New in v1.2.0

• 🆕 analyzing-all-conversations skill — new plugin skill that analyzes and organizes your ENTIRE indexed conversation history into one report, and kicks off backfill for unanalyzed sessions
• 🆕 memory-bank analyze command — the deterministic engine behind the skill: coverage, per-project rollups, fact/domain breakdowns, monthly timeline, backfill recommendations
• 🆕 Transparent .jsonl.zst archive support — archives compressed out-of-band keep working across every read path (no extra dependency; requires Node >= 22.15 for built-in zstd)
• ⚡ FTS5 text search — BM25-ranked full-text search replaces the LIKE full scan (recall@10 0.93 → 1.00, FTS index 2.9GB → 407MB)
• ⚡ Search-path performance — cached DB connection, int8 vector quantization (dual-dtype), query-embedding memoization
• 🧹 Fact extraction quality/cost controls — trivial-exchange filtering, in-session dedup, strict confidence gating, per-session LLM call budget

Features

• Knowledge Graph -- Ontology classification (Domain → Category) + typed relations (INFLUENCES, SUPPORTS, SUPERSEDES, CONTRADICTS)
• RAG Search -- Search results auto-enriched with related facts and ontology context
• Conversation Search -- Semantic vector search + FTS5 (BM25) full-text search across all past conversations
• Full-History Analysis -- memory-bank analyze + analyzing-all-conversations skill: coverage-checked report over the entire conversation index (projects, facts, domains, timeline, backfill gaps) — run memory-bank sync first so new conversations are indexed
• Fact Extraction -- Automatic extraction of decisions, preferences, patterns from conversations (trivial-exchange filtering, in-session dedup, confidence gating, LLM call budgeting)
• Fact Consolidation -- Duplicate detection, contradiction handling, evolution tracking
• Graph Traversal -- Multi-hop exploration (up to 3 hops) to trace decision chains
• Cross-Project Insights -- Find similar decisions from other projects
• Fact Provenance -- Trace any fact back to its source conversation
• Scope Isolation -- Project facts stay in their project, global facts are shared
• Compressed Archive Support -- Transparent .jsonl.zst reads across every path (parser, read tool, search, sync, stats, verify) using Node's built-in zstd (Node >= 22.15) — archives compressed out-of-band keep working
• MCP Integration -- 9 tools: search, read, search_facts, search_ontology, ask_avatar, trace_fact, explore_graph, cross_project_insights, graph_stats
• 3D Visualization -- Interactive neon-style knowledge graph with data flow animation
• Web UI -- Dark-theme web interface for browsing and searching conversations

How It Works

Data Pipeline

▲ Prompt Input       User messages, tool calls, assistant responses
│                    → JSONL archiving → embedding (384-dim)
│
◎ User Scope         LLM extracts facts (decisions, preferences, patterns)
│                    → global facts shared across all projects
│
● Project Scope      Facts scoped per project, clustered by domain
│                    → cross-project insights available
│
◇ Ontology           Auto-classified into domains & categories
                     → typed relations: INFLUENCES / SUPPORTS / SUPERSEDES / CONTRADICTS
                     → multi-hop graph traversal (up to 3 hops)

graph LR
    A[Session End] -->|sync| B[Parse JSONL]
    B --> C[Generate Embeddings]
    C --> D[SQLite + sqlite-vec]

    A -->|extract| E[Haiku LLM]
    E --> F[Facts]
    F -->|classify| G[Ontology]
    G -->|detect| H[Relations]

    I[Session Start] -->|consolidate| J[Vector Search]
    J --> K[Haiku LLM]
    K -->|merge/replace/evolve| F

    L[Query] -->|search| D
    L -->|RAG| F
    L -->|traverse| H

Install

In Claude Code:

/plugin marketplace add https://github.com/Kalamari0227/our_room
/plugin install memory-bank

Update

/plugin update memory-bank

Quick Start

memory-bank sync      # Sync & index conversations
memory-bank search "React auth"  # Semantic search
memory-bank stats     # Index statistics
memory-bank analyze   # Full-history analysis report (coverage, projects, facts)

Full-History Analysis

memory-bank analyze aggregates the entire conversation index into one report — deterministic, read-only, no LLM calls. It reports on what has been indexed into SQLite: archive files that were never synced are not visible to it. The SessionStart hook syncs in the background, but that runs asynchronously — when freshness matters, run memory-bank sync (foreground) and let it finish before analyzing.