Every conversation you have with an AI — every decision, every debugging session, every architecture debate — disappears when the session ends. Six months of work, gone. You start over every time.
Other memory systems try to fix this by letting AI decide what's worth remembering. It extracts "user prefers Postgres" and throws away the conversation where you explained why. MemPalace takes a different approach: store everything, then make it findable.
The Palace — Ancient Greek orators memorized entire speeches by placing ideas in rooms of an imaginary building. Walk through the building, find the idea. MemPalace applies the same principle to AI memory: your conversations are organized into wings (people and projects), halls (types of memory), and rooms (specific ideas). No AI decides what matters — you keep every word, and the structure gives you a navigable map instead of a flat search index.
Raw verbatim storage — MemPalace stores your actual exchanges in ChromaDB without summarization or extraction. The 96.6% LongMemEval result comes from this raw mode. We don't burn an LLM to decide what's "worth remembering" — we keep everything and let semantic search find it.
AAAK (experimental) — A lossy abbreviation dialect for packing repeated entities into fewer tokens at scale. Readable by any LLM that reads text — Claude, GPT, Gemini, Llama, Mistral — no decoder needed. AAAK is a separate compression layer, not the storage default, and on the LongMemEval benchmark it currently regresses vs raw mode (84.2% vs 96.6%). We're iterating. See the note above for the honest status.
Local, open, adaptable — MemPalace runs entirely on your machine, on any data you have locally, without using any external API or services. It has been tested on conversations — but it can be adapted for different types of datastores. This is why we're open-sourcing it.
Quick Start · The Palace · AAAK Dialect · Benchmarks · MCP Tools
| 96.6% LongMemEval R@5 raw mode, zero API calls |
500/500 questions tested independently reproduced |
$0 No subscription No cloud. Local only. |
Reproducible — runners in benchmarks/. Full results. The 96.6% is from raw verbatim mode, not AAAK or rooms mode (those score lower — see note above).
The community caught real problems in this README within hours of launch and we want to address them directly.
What we got wrong:
The AAAK token example was incorrect. We used a rough heuristic (
len(text)//3) for token counts instead of an actual tokenizer. Real counts via OpenAI's tokenizer: the English example is 66 tokens, the AAAK example is 73. AAAK does not save tokens at small scales — it's designed for repeated entities at scale, and the README example was a bad demonstration of that. We're rewriting it."30x lossless compression" was overstated. AAAK is a lossy abbreviation system (entity codes, sentence truncation). Independent benchmarks show AAAK mode scores 84.2% R@5 vs raw mode's 96.6% on LongMemEval — a 12.4 point regression. The honest framing is: AAAK is an experimental compression layer that trades fidelity for token density, and the 96.6% headline number is from RAW mode, not AAAK.
"+34% palace boost" was misleading. That number compares unfiltered search to wing+room metadata filtering. Metadata filtering is a standard ChromaDB feature, not a novel retrieval mechanism. Real and useful, but not a moat.
"Contradiction detection" exists as a separate utility (
fact_checker.py) but is not currently wired into the knowledge graph operations as the README implied."100% with Haiku rerank" is real (we have the result files) but the rerank pipeline is not in the public benchmark scripts. We're adding it.
What's still true and reproducible:
- 96.6% R@5 on LongMemEval in raw mode, on 500 questions, zero API calls — independently reproduced on M2 Ultra in under 5 minutes by @gizmax.
- Local, free, no subscription, no cloud, no data leaving your machine.
- The architecture (wings, rooms, closets, drawers) is real and useful, even if it's not a magical retrieval boost.
What we're doing:
- Rewriting the AAAK example with real tokenizer counts and a scenario where AAAK actually demonstrates compression
- Adding
mode raw / aaak / roomsclearly to the benchmark documentation so the trade-offs are visible- Wiring
fact_checker.pyinto the KG ops so the contradiction detection claim becomes true- Pinning ChromaDB to a tested range (Issue #100), fixing the shell injection in hooks (#110), and addressing the macOS ARM64 segfault (#74)
Thank you to everyone who poked holes in this. Brutal honest criticism is exactly what makes open source work, and it's what we asked for. Special thanks to @panuhorsmalahti, @lhl, @gizmax, and everyone who filed an issue or a PR in the first 48 hours. We're listening, we're fixing, and we'd rather be right than impressive.
— Milla Jovovich & Ben Sigman
pip install mempalace
# Set up your world — who you work with, what your projects are
mempalace init ~/projects/myapp
# Mine your data
mempalace mine ~/projects/myapp # projects — code, docs, notes
mempalace mine ~/chats/ --mode convos # convos — Claude, ChatGPT, Slack exports
mempalace mine ~/chats/ --mode convos --extract general # general — classifies into decisions, milestones, problems
# Search anything you've ever discussed
mempalace search "why did we switch to GraphQL"
# Your AI remembers
mempalace statusThree mining modes: projects (code and docs), convos (conversation exports), and general (auto-classifies into decisions, preferences, milestones, problems, and emotional context). Everything stays on your machine.
After the one-time setup (install → init → mine), you don't run MemPalace commands manually. Your AI uses it for you. There are two ways, depending on which AI you use.
# Connect MemPalace once
claude mcp add mempalace -- python -m mempalace.mcp_serverNow your AI has 19 tools available through MCP. Ask it anything:
"What did we decide about auth last month?"
Claude calls mempalace_search automatically, gets verbatim results, and answers you. You never type mempalace search again. The AI handles it.
MemPalace also works natively with Gemini CLI (which handles the server and save hooks automatically) — see the Gemini CLI Integration Guide.
Local models generally don't speak MCP yet. Two approaches:
1. Wake-up command — load your world into the model's context:
mempalace wake-up > context.txt
# Paste context.txt into your local model's system promptThis gives your local model ~170 tokens of critical facts (in AAAK if you prefer) before you ask a single question.
2. CLI search — query on demand, feed results into your prompt:
mempalace search "auth decisions" > results.txt
# Include results.txt in your promptOr use the Python API:
from mempalace.searcher import search_memories
results = search_memories("auth decisions", palace_path="~/.mempalace/palace")
# Inject into your local model's contextEither way — your entire memory stack runs offline. ChromaDB on your machine, Llama on your machine, AAAK for compression, zero cloud calls.
Decisions happen in conversations now. Not in docs. Not in Jira. In conversations with Claude, ChatGPT, Copilot. The reasoning, the tradeoffs, the "we tried X and it failed because Y" — all trapped in chat windows that evaporate when the session ends.
Six months of daily AI use = 19.5 million tokens. That's every decision, every debugging session, every architecture debate. Gone.
| Approach | Tokens loaded | Annual cost |
|---|---|---|
| Paste everything | 19.5M — doesn't fit any context window | Impossible |
| LLM summaries | ~650K | ~$507/yr |
| MemPalace wake-up | ~170 tokens | ~$0.70/yr |
| MemPalace + 5 searches | ~13,500 tokens | ~$10/yr |
MemPalace loads 170 tokens of critical facts on wake-up — your team, your projects, your preferences. Then searches only when needed. $10/year to remember everything vs $507/year for summaries that lose context.
The layout is fairly simple, though it took a long time to get there.
It starts with a wing. Every project, person, or topic you're filing gets its own wing in the palace.
Each wing has rooms connected to it, where information is divided into subjects that relate to that wing — so every room is a different element of what your project contains. Project ideas could be one room, employees could be another, financial statements another. There can be an endless number of rooms that split the wing into sections. The MemPalace install detects these for you automatically, and of course you can personalize it any way you feel is right.
Every room has a closet connected to it, and here's where things get interesting. We've developed an AI language called AAAK. Don't ask — it's a whole story of its own. Your agent learns the AAAK shorthand every time it wakes up. Because AAAK is essentially English, but a very truncated version, your agent understands how to use it in seconds. It comes as part of the install, built into the MemPalace code. In our next update, we'll add AAAK directly to the closets, which will be a real game changer — the amount of info in the closets will be much bigger, but it will take up far less space and far less reading time for your agent.
Inside those closets are drawers, and those drawers are where your original files live. In this first version, we haven't used AAAK as a closet tool, but even so, the summaries have shown 96.6% recall in all the benchmarks we've done across multiple benchmarking platforms. Once the closets use AAAK, searches will be even faster while keeping every word exact. But even now, the closet approach has been a huge boon to how much info is stored in a small space — it's used to easily point your AI agent to the drawer where your original file lives. You never lose anything, and all this happens in seconds.
There are also halls, which connect rooms within a wing, and tunnels, which connect rooms from different wings to one another. So finding things becomes truly effortless — we've given the AI a clean and organized way to know where to start searching, without having to look through every keyword in huge folders.
You say what you're looking for and boom, it already knows which wing to go to. Just that in itself would have made a big difference. But this is beautiful, elegant, organic, and most importantly, efficient.
┌─────────────────────────────────────────────────────────────┐
│ WING: Person │
│ │
│ ┌──────────┐ ──hall── ┌──────────┐ │
│ │ Room A │ │ Room B │ │
│ └────┬─────┘ └──────────┘ │
│ │ │
│ ▼ │
│ ┌──────────┐ ┌──────────┐ │
│ │ Closet │ ───▶ │ Drawer │ │
│ └──────────┘ └──────────┘ │
└─────────┼──────────────────────────────────────────────────┘
│
tunnel
│
┌─────────┼──────────────────────────────────────────────────┐
│ WING: Project │
│ │ │
│ ┌────┴─────┐ ──hall── ┌──────────┐ │
│ │ Room A │ │ Room C │ │
│ └────┬─────┘ └──────────┘ │
│ │ │
│ ▼ │
│ ┌──────────┐ ┌──────────┐ │
│ │ Closet │ ───▶ │ Drawer │ │
│ └──────────┘ └──────────┘ │
└─────────────────────────────────────────────────────────────┘
Wings — a person or project. As many as you need. Rooms — specific topics within a wing. Auth, billing, deploy — endless rooms. Halls — connections between related rooms within the same wing. If Room A (auth) and Room B (security) are related, a hall links them. Tunnels — connections between wings. When Person A and a Project both have a room about "auth," a tunnel cross-references them automatically. Closets — summaries that point to the original content. (In v3.0.0 these are plain-text summaries; AAAK-encoded closets are coming in a future update — see Task #30.) Drawers — the original verbatim files. The exact words, never summarized.
Halls are memory types — the same in every wing, acting as corridors:
hall_facts— decisions made, choices locked inhall_events— sessions, milestones, debugginghall_discoveries— breakthroughs, new insightshall_preferences— habits, likes, opinionshall_advice— recommendations and solutions
Rooms are named ideas — auth-migration, graphql-switch, ci-pipeline. When the same room appears in different wings, it creates a tunnel — connecting the same topic across domains:
wing_kai / hall_events / auth-migration → "Kai debugged the OAuth token refresh"
wing_driftwood / hall_facts / auth-migration → "team decided to migrate auth to Clerk"
wing_priya / hall_advice / auth-migration → "Priya approved Clerk over Auth0"
Same room. Three wings. The tunnel connects them.
Tested on 22,000+ real conversation memories:
Search all closets: 60.9% R@10
Search within wing: 73.1% (+12%)
Search wing + hall: 84.8% (+24%)
Search wing + room: 94.8% (+34%)
Wings and rooms aren't cosmetic. They're a 34% retrieval improvement. The palace structure is the product.
| Layer | What | Size | When |
|---|---|---|---|
| L0 | Identity — who is this AI? | ~50 tokens | Always loaded |
| L1 | Critical facts — team, projects, preferences | ~120 tokens (AAAK) | Always loaded |
| L2 | Room recall — recent sessions, current project | On demand | When topic comes up |
| L3 | Deep search — semantic query across all closets | On demand | When explicitly asked |
Your AI wakes up with L0 + L1 (~170 tokens) and knows your world. Searches only fire when needed.
AAAK is a lossy abbreviation system — entity codes, structural markers, and sentence truncation — designed to pack repeated entities and relationships into fewer tokens at scale. It is readable by any LLM that reads text (Claude, GPT, Gemini, Llama, Mistral) without a decoder, so a local model can use it without any cloud dependency.
Honest status (April 2026):
- AAAK is lossy, not lossless. It uses regex-based abbreviation, not reversible compression.
- It does not save tokens at small scales. Short text already tokenizes efficiently. AAAK overhead (codes, separators) costs more than it saves on a few sentences.
- It can save tokens at scale — in scenarios with many repeated entities (a team mentioned hundreds of times, the same project across thousands of sessions), the entity codes amortize.
- AAAK currently regresses LongMemEval vs raw verbatim retrieval (84.2% R@5 vs 96.6%). The 96.6% headline number is from raw mode, not AAAK mode.
- The MemPalace storage default is raw verbatim text in ChromaDB — that's where the benchmark wins come from. AAAK is a separate compression layer for context loading, not the storage format.
We're iterating on the dialect spec, adding a real tokenizer for stats, and exploring better break points for when to use it. Track progress in Issue #43 and #27.
A separate utility (fact_checker.py) can check assertions against entity facts. It's not currently called automatically by the knowledge graph operations — this is being fixed (track in Issue #27). When enabled it catches things like:
Input: "Soren finished the auth migration"
Output: 🔴 AUTH-MIGRATION: attribution conflict — Maya was assigned, not Soren
Input: "Kai has been here 2 years"
Output: 🟡 KAI: wrong_tenure — records show 3 years (started 2023-04)
Input: "The sprint ends Friday"
Output: 🟡 SPRINT: stale_date — current sprint ends Thursday (updated 2 days ago)
Facts checked against the knowledge graph. Ages, dates, and tenures calculated dynamically — not hardcoded.
# Mine each project's conversations
mempalace mine ~/chats/orion/ --mode convos --wing orion
mempalace mine ~/chats/nova/ --mode convos --wing nova
mempalace mine ~/chats/helios/ --mode convos --wing helios
# Six months later: "why did I use Postgres here?"
mempalace search "database decision" --wing orion
# → "Chose Postgres over SQLite because Orion needs concurrent writes
# and the dataset will exceed 10GB. Decided 2025-11-03."
# Cross-project search
mempalace search "rate limiting approach"
# → finds your approach in Orion AND Nova, shows the differences# Mine Slack exports and AI conversations
mempalace mine ~/exports/slack/ --mode convos --wing driftwood
mempalace mine ~/.claude/projects/ --mode convos
# "What did Soren work on last sprint?"
mempalace search "Soren sprint" --wing driftwood
# → 14 closets: OAuth refactor, dark mode, component library migration
# "Who decided to use Clerk?"
mempalace search "Clerk decision" --wing driftwood
# → "Kai recommended Clerk over Auth0 — pricing + developer experience.
# Team agreed 2026-01-15. Maya handling the migration."Some transcript exports concatenate multiple sessions into one huge file:
mempalace split ~/chats/ # split into per-session files
mempalace split ~/chats/ --dry-run # preview first
mempalace split ~/chats/ --min-sessions 3 # only split files with 3+ sessionsTemporal entity-relationship triples — like Zep's Graphiti, but SQLite instead of Neo4j. Local and free.
from mempalace.knowledge_graph import KnowledgeGraph
kg = KnowledgeGraph()
kg.add_triple("Kai", "works_on", "Orion", valid_from="2025-06-01")
kg.add_triple("Maya", "assigned_to", "auth-migration", valid_from="2026-01-15")
kg.add_triple("Maya", "completed", "auth-migration", valid_from="2026-02-01")
# What's Kai working on?
kg.query_entity("Kai")
# → [Kai → works_on → Orion (current), Kai → recommended → Clerk (2026-01)]
# What was true in January?
kg.query_entity("Maya", as_of="2026-01-20")
# → [Maya → assigned_to → auth-migration (active)]
# Timeline
kg.timeline("Orion")
# → chronological story of the projectFacts have validity windows. When something stops being true, invalidate it:
kg.invalidate("Kai", "works_on", "Orion", ended="2026-03-01")Now queries for Kai's current work won't return Orion. Historical queries still will.
| Feature | MemPalace | Zep (Graphiti) |
|---|---|---|
| Storage | SQLite (local) | Neo4j (cloud) |
| Cost | Free | $25/mo+ |
| Temporal validity | Yes | Yes |
| Self-hosted | Always | Enterprise only |
| Privacy | Everything local | SOC 2, HIPAA |
Create agents that focus on specific areas. Each agent gets its own wing and diary in the palace — not in your CLAUDE.md. Add 50 agents, your config stays the same size.
~/.mempalace/agents/
├── reviewer.json # code quality, patterns, bugs
├── architect.json # design decisions, tradeoffs
└── ops.json # deploys, incidents, infra
Your CLAUDE.md just needs one line:
You have MemPalace agents. Run mempalace_list_agents to see them.
The AI discovers its agents from the palace at runtime. Each agent:
- Has a focus — what it pays attention to
- Keeps a diary — written in AAAK, persists across sessions
- Builds expertise — reads its own history to stay sharp in its domain
# Agent writes to its diary after a code review
mempalace_diary_write("reviewer",
"PR#42|auth.bypass.found|missing.middleware.check|pattern:3rd.time.this.quarter|★★★★")
# Agent reads back its history
mempalace_diary_read("reviewer", last_n=10)
# → last 10 findings, compressed in AAAK
Each agent is a specialist lens on your data. The reviewer remembers every bug pattern it's seen. The architect remembers every design decision. The ops agent remembers every incident. They don't share a scratchpad — they each maintain their own memory.
Letta charges $20–200/mo for agent-managed memory. MemPalace does it with a wing.
claude mcp add mempalace -- python -m mempalace.mcp_serverPalace (read)
| Tool | What |
|---|---|
mempalace_status |
Palace overview + AAAK spec + memory protocol |
mempalace_list_wings |
Wings with counts |
mempalace_list_rooms |
Rooms within a wing |
mempalace_get_taxonomy |
Full wing → room → count tree |
mempalace_search |
Semantic search with wing/room filters |
mempalace_check_duplicate |
Check before filing |
mempalace_get_aaak_spec |
AAAK dialect reference |
Palace (write)
| Tool | What |
|---|---|
mempalace_add_drawer |
File verbatim content |
mempalace_delete_drawer |
Remove by ID |
Knowledge Graph
| Tool | What |
|---|---|
mempalace_kg_query |
Entity relationships with time filtering |
mempalace_kg_add |
Add facts |
mempalace_kg_invalidate |
Mark facts as ended |
mempalace_kg_timeline |
Chronological entity story |
mempalace_kg_stats |
Graph overview |
Navigation
| Tool | What |
|---|---|
mempalace_traverse |
Walk the graph from a room across wings |
mempalace_find_tunnels |
Find rooms bridging two wings |
mempalace_graph_stats |
Graph connectivity overview |
Agent Diary
| Tool | What |
|---|---|
mempalace_diary_write |
Write AAAK diary entry |
mempalace_diary_read |
Read recent diary entries |
The AI learns AAAK and the memory protocol automatically from the mempalace_status response. No manual configuration.
Two hooks for Claude Code that automatically save memories during work:
Save Hook — every 15 messages, triggers a structured save. Topics, decisions, quotes, code changes. Also regenerates the critical facts layer.
PreCompact Hook — fires before context compression. Emergency save before the window shrinks.
{
"hooks": {
"Stop": [{"matcher": "", "hooks": [{"type": "command", "command": "/path/to/mempalace/hooks/mempal_save_hook.sh"}]}],
"PreCompact": [{"matcher": "", "hooks": [{"type": "command", "command": "/path/to/mempalace/hooks/mempal_precompact_hook.sh"}]}]
}
}Tested on standard academic benchmarks — reproducible, published datasets.
| Benchmark | Mode | Score | API Calls |
|---|---|---|---|
| LongMemEval R@5 | Raw (ChromaDB only) | 96.6% | Zero |
| LongMemEval R@5 | Hybrid + Haiku rerank | 100% (500/500) | ~500 |
| LoCoMo R@10 | Raw, session level | 60.3% | Zero |
| Personal palace R@10 | Heuristic bench | 85% | Zero |
| Palace structure impact | Wing+room filtering | +34% R@10 | Zero |
The 96.6% raw score is the highest published LongMemEval result requiring no API key, no cloud, and no LLM at any stage.
| System | LongMemEval R@5 | API Required | Cost |
|---|---|---|---|
| MemPalace (hybrid) | 100% | Optional | Free |
| Supermemory ASMR | ~99% | Yes | — |
| MemPalace (raw) | 96.6% | None | Free |
| Mastra | 94.87% | Yes (GPT) | API costs |
| Mem0 | ~85% | Yes | $19–249/mo |
| Zep | ~85% | Yes | $25/mo+ |
# Setup
mempalace init <dir> # guided onboarding + AAAK bootstrap
# Mining
mempalace mine <dir> # mine project files
mempalace mine <dir> --mode convos # mine conversation exports
mempalace mine <dir> --mode convos --wing myapp # tag with a wing name
# Splitting
mempalace split <dir> # split concatenated transcripts
mempalace split <dir> --dry-run # preview
# Search
mempalace search "query" # search everything
mempalace search "query" --wing myapp # within a wing
mempalace search "query" --room auth-migration # within a room
# Memory stack
mempalace wake-up # load L0 + L1 context
mempalace wake-up --wing driftwood # project-specific
# Compression
mempalace compress --wing myapp # AAAK compress
# Status
mempalace status # palace overviewAll commands accept --palace <path> to override the default location.
{
"palace_path": "/custom/path/to/palace",
"collection_name": "mempalace_drawers",
"people_map": {"Kai": "KAI", "Priya": "PRI"}
}Generated by mempalace init. Maps your people and projects to wings:
{
"default_wing": "wing_general",
"wings": {
"wing_kai": {"type": "person", "keywords": ["kai", "kai's"]},
"wing_driftwood": {"type": "project", "keywords": ["driftwood", "analytics", "saas"]}
}
}Plain text. Becomes Layer 0 — loaded every session.
| File | What |
|---|---|
cli.py |
CLI entry point |
config.py |
Configuration loading and defaults |
normalize.py |
Converts 5 chat formats to standard transcript |
mcp_server.py |
MCP server — 19 tools, AAAK auto-teach, memory protocol |
miner.py |
Project file ingest |
convo_miner.py |
Conversation ingest — chunks by exchange pair |
searcher.py |
Semantic search via ChromaDB |
layers.py |
4-layer memory stack |
dialect.py |
AAAK compression — 30x lossless |
knowledge_graph.py |
Temporal entity-relationship graph (SQLite) |
palace_graph.py |
Room-based navigation graph |
onboarding.py |
Guided setup — generates AAAK bootstrap + wing config |
entity_registry.py |
Entity code registry |
entity_detector.py |
Auto-detect people and projects from content |
split_mega_files.py |
Split concatenated transcripts into per-session files |
hooks/mempal_save_hook.sh |
Auto-save every N messages |
hooks/mempal_precompact_hook.sh |
Emergency save before compaction |
mempalace/
├── README.md ← you are here
├── mempalace/ ← core package (README)
│ ├── cli.py ← CLI entry point
│ ├── mcp_server.py ← MCP server (19 tools)
│ ├── knowledge_graph.py ← temporal entity graph
│ ├── palace_graph.py ← room navigation graph
│ ├── dialect.py ← AAAK compression
│ ├── miner.py ← project file ingest
│ ├── convo_miner.py ← conversation ingest
│ ├── searcher.py ← semantic search
│ ├── onboarding.py ← guided setup
│ └── ... ← see mempalace/README.md
├── benchmarks/ ← reproducible benchmark runners
│ ├── README.md ← reproduction guide
│ ├── BENCHMARKS.md ← full results + methodology
│ ├── longmemeval_bench.py ← LongMemEval runner
│ ├── locomo_bench.py ← LoCoMo runner
│ └── membench_bench.py ← MemBench runner
├── hooks/ ← Claude Code auto-save hooks
│ ├── README.md ← hook setup guide
│ ├── mempal_save_hook.sh ← save every N messages
│ └── mempal_precompact_hook.sh ← emergency save
├── examples/ ← usage examples
│ ├── basic_mining.py
│ ├── convo_import.py
│ └── mcp_setup.md
├── tests/ ← test suite (README)
├── assets/ ← logo + brand assets
└── pyproject.toml ← package config (v3.0.0)
- Python 3.9+
chromadb>=0.4.0pyyaml>=6.0
No API key. No internet after install. Everything local.
pip install mempalacePRs welcome. See CONTRIBUTING.md for setup and guidelines.
MIT — see LICENSE.