MemPalace is an 🏷️#open-source AI memory system that stores conversation history verbatim and makes it semantically searchable. Built in Python on @ChromaDB, it uses the ancient Greek Method of Loci as its organizational metaphor — Wings (people/projects), Rooms (topics), Halls (memory types), and Drawers (verbatim content). Created by 📝Milla Jovovich.

Architecture — local-first. ChromaDB for vector storage, SQLite for a temporal knowledge graph with valid_from/valid_to on every fact. No cloud dependencies. 96.6% retrieval on LongMemEval with zero API calls
Key differentiator — stores everything verbatim rather than using LLMs to extract and summarize. This beats LLM-extraction approaches (Mem0 at 85%, Zep at 85%) because no information is lost during ingestion
Context efficiency — 4-layer memory stack (L0 identity, L1 essential story, L2 room recall, L3 deep search) loads at ~170 tokens on wake-up, leaving 99%+ of the context window free
MCP server — 19 tools for reading, writing, knowledge graph queries, navigation, and agent diaries. Compatible with 📝Claude Code, ChatGPT, 📝Cursor, and Gemini
Ingestion — mines 5 chat formats (Claude Code JSONL, Claude.ai JSON, ChatGPT, Slack, plain text) plus project files. Auto-save hooks capture sessions every 15 messages

MemPalace answers "how do I make an AI remember everything?" — verbatim storage with semantic search. 📝MythOS answers "how do I make thinking visible, structured, and sovereign while also making the AI remember everything?" MemPalace captures raw conversation transcripts that no one will ever read directly; MythOS produces curated artifacts that humans navigate and build upon. The approaches are complementary — MemPalace's strengths in tiered context loading, temporal knowledge graphs, and conversation capture point to real gaps in MythOS that are worth closing.

Contexts