v6.0.0

Release v6.0.0 — ...

v5.0.0

32 MCP tools, Knowledge Graph, Episodic Memory, Reflection System. See
README for details.

v4.0.0 — Privacy & Observations

Claude Total Memory v4.0.0

Privacy-first update with lightweight tracking and token optimization.

New: Privacy Stripping

• Automatic redaction of API keys, JWTs, passwords, credit cards, emails, IPs
• <private> tag support for explicit content exclusion
• All redaction happens before storage — sensitive data never persists

New: Observations

• memory_observe — lightweight file change tracking (no embeddings, no dedup)
• 6 types: bugfix, feature, refactor, change, discovery, decision
• Auto-cleanup after 30 days

New: Progressive Disclosure

• 3 detail levels: compact (~50 tokens), summary (150 chars), full
• Up to 10x token savings on search results
• Token cost estimation in every response

New: Live Feed

• Real-time SSE stream in dashboard
• Knowledge, errors, and observations in real-time

Also Added

• Branch-aware context filtering
• Codex CLI support with shared memory database
• 20 tools total

v3.0.0 — Self-Improving Agent

Claude Total Memory v3.0.0

Major update: Claude now learns from its own mistakes.

New: Self-Improvement Pipeline

• self_error_log — automatic error logging with categories and severity
• self_insight — ExpeL-style insight extraction with voting (importance + confidence)
• self_rules — behavioral rules (SOUL) promoted from high-confidence insights
• self_patterns — error pattern analysis and improvement trends
• self_reflect — session reflections for meta-observations
• self_rules_context — load rules at session start

How It Works

Error → (3+ same category) → Insight → (importance≥5, confidence≥0.8) → Rule (SOUL)

Also Added

• 6 new tools (13 → 19 total)
• Pattern detection: 3+ errors in 30 days triggers suggestion
• Auto-suspend: rules with <20% success rate after 10+ fires
• Dashboard: Self-Improvement tab + Rules/SOUL tab

v1.0.0 — Initial Release

Claude Total Memory v1.0.0

First public release.

Features

• MCP server with 13 tools for persistent memory
• 4-tier search pipeline: FTS5 (BM25) → semantic (ChromaDB) → fuzzy → graph expansion
• 5 knowledge types: decision, solution, lesson, fact, convention
• Automatic deduplication (Jaccard + fuzzy)
• Decay scoring with spaced repetition
• Web dashboard at localhost:37737
• Session timeline and knowledge graph
• SQLite + ChromaDB local storage
• One-command installer for macOS/Linux/Windows