🧠 Mnemos

Personal thinking environment. Structure thoughts, not just words.

What

Mnemos stores atomic notes as typed units of knowledge. Each note is:

• One concept per file (.md or .m.md)
• Typed: fact, question, todo, pitch, related
• Tagged and linked to other notes
• Versioned through Git
• Organized into four main sections:
  • p/ - Projects and concepts by namespace (e.g., p/moul/, p/amigos/)
  • ml/ - moul.life public content (blog posts, about pages)
  • kb/ - Freeform personal wiki for general knowledge
  • b/ - Brainstorm/braindump with dated random ideas

Why

• Atomic thinking → ideas become reusable building blocks
• Typed notes → structure emerges from classification
• Git-native → version your mind’s evolution
• Terminal-first → no friction between thought and capture

Install

go install moul.io/mnemos@latest

Usage

Create notes

mnemos new fact "Go channels are typed conduits"
mnemos new question "How does context propagation affect goroutine lifecycles?"
mnemos new todo "Implement retry logic for flaky tests"

Search and connect

mnemos find --type question --tag golang
mnemos graph --from "channels.md" --depth 2
mnemos related "concurrency patterns"

Daily workflow

mnemos sync              # Push/pull with Git
mnemos summary --today   # What changed today
mnemos gc                # Find orphaned notes

Note format

---
type: fact
tags: [golang, concurrency]
created: 2024-01-15
---

- Go channels are typed conduits for communication
- Send/receive operations block by default
- Closed channels can't be reopened

## See also

- <a href="/wiki/p/moul/golang" class="wiki-link">p/moul/golang</a>
- <a href="/wiki/p/amigos/technical-vision" class="wiki-link">p/amigos/technical-vision</a>

Configuration

# ~/.config/mnemos/config.toml
repo = "~/brain"
editor = "nvim"
ai_provider = "anthropic"
default_type = "fact"

Architecture

Single Go binary handling:

• Note CRUD operations
• Git synchronization
• Full-text + semantic search
• Graph traversal and visualization
• Garbage collection for orphaned notes
• AI-powered clarification dialogues

Philosophy

Mnemos treats knowledge as:

1. Atomic - one idea per note
2. Typed - structure through classification
3. Connected - meaning through relationships
4. Versioned - evolution through Git

Unlike traditional note-taking, Mnemos encourages refactoring thoughts like code.

Roadmap

Core Features

• [ ] Interactive capture - mnemos new with guided prompts for better note creation
• [ ] Smart search - mnemos search "keyword" with semantic understanding, not just grep
• [ ] Q&A mode - mnemos ask "question" uses knowledge base to synthesize answers
• [ ] Chat mode - mnemos chat for interactive clarification dialogues

Telegram Bot (mnemos tg-bot)

• [ ] Private bot - Personal Telegram interface for on-the-go capture
• [ ] Interactive note creation - Bot asks clarifying questions, creates structured notes
• [ ] Voice notes - Transcribe audio messages into typed notes
• [ ] Git sync - Auto-pull updates, notify on changes
• [ ] Daily summaries - “Today: 5 new notes, 3 commits, trending: golang, architecture”
• [ ] Smart notifications - Alert on orphaned notes, suggest connections

Advanced Features

• [ ] Semantic embeddings - Vector search for conceptual similarity
• [ ] Auto-linking - Suggest connections between related notes
• [ ] Mermaid diagrams - Auto-generate concept maps from links
• [ ] Export formats - Obsidian, Roam, Notion compatibility
• [ ] Multi-vault - Separate work/personal knowledge bases

Integration Goals

• [ ] Mimir bridge - Reference datasets in notes
• [ ] Claude deep integration - Continuous learning from interactions
• [ ] GitHub Actions - Auto-organize PRs into notes
• [ ] Calendar integration - Time-based note reviews

AI Enhancement

• [ ] Fine-tuned model - Train custom LLM on your knowledge base for personalized responses
• [ ] RAG system - Retrieval-augmented generation using your notes as context
• [ ] Knowledge distillation - Create specialized AI assistant that thinks like you
• [ ] Embedding index - Pre-compute semantic embeddings for instant similarity search
• [ ] Personal AI agent - Export knowledge base to create domain-specific chatbot
• [ ] Claude Projects integration - Export notes for persistent Claude knowledge
• [ ] Local model support - Fine-tune Llama/Mistral models for privacy
• [ ] Training data export - Generate JSONL/Parquet for various platforms
• [ ] Continuous learning - Update models as knowledge base grows

See also

• Mimir - Dataset archival companion
• CLAUDE.md - AI agent instructions

Sources

• Mimir Repository: https://github.com/moul/mimir - Dataset archival tool companion to Mnemos