addyosmani/mempalace — explained in plain English
Analysis updated 2026-07-09 · repo last pushed 2026-04-07
Search past AI conversations to recall why a specific technical decision was made.
Import Slack exports and Claude chats to build a searchable team knowledge base.
Detect contradictions when someone cites outdated facts or wrong task ownership.
Load months of project context into a local AI model without cloud API calls.
| addyosmani/mempalace | bennybar/lulireddit | dantiicu/wine-nx | |
|---|---|---|---|
| Stars | 48 | 48 | 48 |
| Language | — | Dart | C |
| Last pushed | 2026-04-07 | — | — |
| Maintenance | Maintained | — | — |
| Setup difficulty | moderate | moderate | hard |
| Complexity | 3/5 | 2/5 | 5/5 |
| Audience | developer | general | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires importing existing conversations and optionally configuring a local AI model like Llama or Mistral for offline use.
MemPalace solves a problem anyone who uses AI tools regularly will recognize: every conversation you have with ChatGPT, Claude, or Cursor disappears when the session ends. Six months of decisions, debugging sessions, and architecture debates just vanish. You start over each time. Instead of letting an AI decide what's worth remembering (which usually means it extracts "user prefers Postgres" and throws away the explanation of why), this project stores everything and makes it findable. The system organizes your conversations into a spatial structure inspired by the ancient Greek memory palace technique. Your data gets sorted into wings (people or projects), rooms (specific topics like auth-migration or graphql-switch), and closets (compressed summaries that point to the original files). Cross-references between related topics in different wings are connected automatically through what it calls tunnels. This structure alone improves retrieval by 34% compared to searching through unorganized data. It also uses a custom shorthand language called AAAK that compresses text about 30x with no information loss. It's not meant for humans to read, it's designed so any AI model can load months of context in roughly 120 tokens. This means you can run it with local models like Llama or Mistral and keep your entire memory stack completely offline, with no cloud API calls. The tool is built for developers, team leads, and anyone working across multiple AI-assisted projects. A solo developer could mine six months of conversations across three projects and later search "why did I choose Postgres here?" to get the exact decision and date. A team lead could import Slack exports and Claude conversations, then ask "who decided to use Clerk for auth?" and get the recommendation, the reasoning, and who agreed. It also catches contradictions, if someone says the wrong person handled a task or cites an outdated date, it flags the conflict against stored records. Everything runs locally on your machine using a SQLite-based knowledge graph and local storage. The project is free and open-source, and the benchmarks are reproducible.
MemPalace saves and organizes your past AI chat conversations locally so you can search them later. It compresses months of context so any AI model can load it instantly without cloud calls.
Maintained — commit in last 6 months (last push 2026-04-07).
Free and open-source software that you can use without cloud dependencies.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.