What is MemPalace?
MemPalace is a local knowledge management system co-founded by actress Milla Jovovich and developer Ben Sigman that stores, organizes, and retrieves everything you discuss with AI tools across every session. It addresses a core problem with AI assistants: every conversation ends and the context disappears. MemPalace keeps a full verbatim record of your AI conversations, code sessions, project files, and decisions, then makes that record searchable through semantic queries. It works as an MCP server with Claude, ChatGPT, Cursor, and Gemini, and installs natively as a Claude Code plugin. It also supports local models like Llama and Mistral through a Python API and CLI. The entire system runs on your machine with no cloud dependency and no API key required after install.
Features & Benefits
- Verbatim Storage — store complete, unaltered conversation and project content in a local ChromaDB vector database without summarization or data loss.
- Hierarchical Palace Structure — organize knowledge into wings (people and projects), rooms (specific topics), halls (memory type corridors), closets (summary pointers), and drawers (original verbatim files) for structured, navigable knowledge management.
- Semantic Search — query your full knowledge base using natural language to retrieve relevant results across all stored content.
- Wing and Room Filtering — narrow semantic search to a specific wing, room, or combination to improve retrieval precision.
- Tunnel Navigation — connect the same topic room across different wings automatically, cross-referencing related knowledge from separate people or projects.
- Three Mining Modes — ingest knowledge from project files, conversation exports, or general content with automatic classification into decisions, preferences, milestones, problems, and emotional context.
- Temporal Knowledge Graph — store entity-relationship triples with validity windows in a local SQLite database, track changes over time, and query the state of any entity at any historical point.
- Four-Layer Memory Stack — load a minimal identity and critical facts context on every session wake-up, then retrieve deeper knowledge on demand through room recall and full semantic search.
- Specialist Agent Diaries — create focused agents for roles like code review or architecture, each maintaining its own persistent diary and expertise across sessions.
- Auto-Save Hooks — trigger structured memory saves automatically at set message intervals and before context compression in Claude Code.
- AAAK Compression (Experimental) — apply a lossy abbreviation dialect to pack repeated entities and relationships into fewer tokens for context loading at scale.
- Contradiction Detection (Experimental) — check new assertions against stored entity facts to flag attribution conflicts, stale dates, and tenure mismatches.
- Transcript Splitting — split concatenated multi-session chat export files into individual per-session files before mining.
- 19 MCP Tools — access the full knowledge management system through a structured set of read, write, knowledge graph, navigation, and agent diary tools via MCP.
- CLI Access — run init, mine, search, wake-up, compress, split, and status commands directly from the terminal.
- Local Model Support — generate a wake-up context file or run CLI searches to feed results into any offline LLM without a cloud connection.
- Integrations — connect to MCP-compatible AI tools and coding environments through a standard MCP server interface.
- Open Source — access, modify, and self-host the full codebase under an MIT license with no subscription required.
Real-World Applications
Six months of daily AI-assisted development produces millions of tokens of decisions, debugging sessions, and architecture debates. A knowledge management system like MemPalace can give a solo developer instant recall of why they chose a specific database, how they resolved a recurring bug, or what tradeoffs they weighed before picking an authentication provider. A search query returns the verbatim exchange from months earlier, not a summary.
Product and engineering teams that work across multiple projects may use MemPalace to preserve institutional knowledge that would otherwise live only in chat windows. When a new team member needs context on a past decision, or when a lead wants to trace who recommended a specific approach and when, the knowledge graph can surface that history with time-accurate results.
Developers who work with local models and need persistent memory across sessions can use the wake-up command to load a compact context file into any offline LLM before starting work. The CLI search function lets them pull targeted results into a prompt without sending data to any external service, keeping the entire knowledge management workflow air-gapped.
Teams that build and manage specialist AI agents may use the agent diary system to give each agent its own persistent knowledge store. A code review agent can accumulate and query its own history of patterns, findings, and past recommendations. An architecture agent can maintain a running record of every design decision it has weighed, separate from the general project knowledge base.
