Skip to content
BranchesTags

Repository files navigation

AetherMind

Private AI that remembers your life. Locally. Forever.
Local-first  ·  Privacy-first  ·  Zero cloud  ·  Zero subscriptions  ·  Open source

AetherMind is a personal cognitive infrastructure: it collects data from your life (git commits, notes, calendar, location history), creates semantic memories using local AI, reflects on your days, and answers natural language questions about your past.

AetherMind Timeline View

Features

  • Automatic data collection - git commits, notes, Google Calendar, Google Takeout Timeline
  • Incremental sync - only imports new data; full deduplication via content hashing
  • Semantic memory search - ask "when was I most productive?" and get answers from your actual history
  • Daily AI reflection - local LLM generates daily summaries, wins, risks, and pattern detection
  • 100% local - runs on your machine; no data leaves your computer
  • CUDA-accelerated - embeddings run on your GPU for speed
  • Streamlit UI - Timeline, Ask Memory, Reflections, Stats views
  • One-command setup - python setup.py handles everything

Architecture

Data Sources                Pipeline                   Storage
────────────                ────────                   ───────
📝 Notes (.md/.txt)  ─┐    collect.py                 SQLite (events)
💻 Git commits       ─┤ →  normalize.py   →   embed → Qdrant (vectors)
📅 Google Calendar   ─┤    index.py                   JSONL (log)
🗺  Google Timeline  ─┘    reflect.py                 JSON (reflections)
                            │
                            ▼
                     Ollama (local LLM)
                     BAAI/bge-small embeddings
                            │
                            ▼
                     ask.py / app.py (Streamlit UI)

Quick Start

Prerequisites

  • Python 3.11+
  • Ollama installed
  • NVIDIA GPU recommended (RTX series with CUDA 12.x)

1. Clone & setup

git clone https://github.com/YOUR_USERNAME/aethermind.git
cd aethermind
python setup.py

The setup wizard will:

  • Check system requirements
  • Install all Python dependencies
  • Optionally connect your Google account
  • Configure daily automation
  • Run the first import

2. Pull the AI model

ollama pull qwen2.5:7b

3. Add your data

input/
├── notes/                    ← Drop .txt or .md files here
├── calendar.csv              ← Export from Google/Outlook/Apple Calendar
└── Semantic_Location_History/ ← Google Takeout → Location History → JSON files

4. Run the pipeline

python run_pipeline.py

5. Open the UI

streamlit run app.py

6. Ask your memory

python ask.py "When was I most productive?"
python ask.py "What projects did I work on in March?"
python ask.py "How many times did I go to the gym this month?" --type health
python ask.py --interactive

Google Calendar Setup

AetherMind can automatically sync your Google Calendar events.

1. Create Google Cloud credentials

  1. Go to Google Cloud Console
  2. Create a new project: AetherMind
  3. Enable the Google Calendar API: APIs & Services → Library → Search "Calendar API" → Enable
  4. Create OAuth credentials: APIs & Services → Credentials → Create Credentials → OAuth client ID
    • Application type: Desktop app
    • Name: AetherMind
  5. Click Download JSON
  6. Save as: credentials/google_client_secret.json

2. Authenticate

python setup.py    # Step 4 runs OAuth (opens browser)

or directly:

python -c "from collect.google_calendar import get_credentials; get_credentials()"

3. How it works

  • First sync: fetches all events from the past year
  • Subsequent syncs: uses Google's syncToken - only fetches changes since last sync (fast)
  • Token storage: saved in credentials/google_token.json (auto-refreshes, never expires)
  • All calendars: syncs every calendar on your account (configurable in config.yaml)

Note: credentials/ is gitignored. Your tokens never leave your machine.

Google Timeline (Location History)

Google does not provide an API for Maps Timeline. Options:

Option A - Google Takeout (recommended)

  1. Go to takeout.google.com
  2. Select only "Location History (Timeline)"
  3. Export as JSON
  4. Extract the ZIP → copy Semantic_Location_History/ to input/
  5. Run python run_pipeline.py

AetherMind handles both old format (timelineObjects) and new format (semanticSegments).

Option B - OwnTracks (continuous tracking) Set up OwnTracks on your phone → export GPX → import as notes.

Daily Automation

After running python setup.py, two Windows Task Scheduler tasks are created:

Task Time Action
AetherMind-Pipeline 20:00 Collect + normalize + index
AetherMind-Reflect 21:00 AI daily reflection

Manual management:

# View tasks
schtasks /query /tn "AetherMind-Pipeline"

# Delete tasks
schtasks /delete /tn "AetherMind-Pipeline" /f
schtasks /delete /tn "AetherMind-Reflect" /f

# Re-run setup
python setup.py

CLI Reference

# Setup
python setup.py                                    # Full setup wizard

# Pipeline
python run_pipeline.py                             # Full pipeline (collect+normalize+index)
python run_pipeline.py --stages collect,normalize  # Specific stages
python run_pipeline.py --reflect                   # Reflection only
python run_pipeline.py --source git                # One source only

# Individual modules
python normalize.py --dry-run                      # Preview events without saving
python index.py --stats                            # Show vector index stats
python index.py --rebuild                          # Rebuild Qdrant from scratch
python reflect.py --date 2026-04-14               # Reflect on specific date
python reflect.py --force                          # Overwrite existing reflection

# Q&A
python ask.py "question"                           # One-shot question
python ask.py "question" --type health             # Filter by event type
python ask.py "question" --since 2026-01-01        # Only events after date
python ask.py --interactive                        # REPL mode

# UI
streamlit run app.py                              # Web interface (localhost:8501)

Configuration

All settings in config.yaml:

embedding:
  model_name: "BAAI/bge-small-en-v1.5"  # 384-dim, MIT license
  device: "cuda"                          # "cuda" or "cpu"
  batch_size: 128                         # Higher = faster on good GPU

ollama:
  model: "qwen2.5:7b"                    # Any Ollama model
  temperature: 0.3                        # Lower = more factual

collect:
  google_calendar:
    lookback_days: 365                    # How far back on first sync
    calendars: []                         # [] = all calendars
  git:
    max_days_back: 365
  google_timeline:
    min_duration_minutes: 5               # Skip very short visits

ask:
  top_k: 8                               # Events to retrieve per query
  rerank: true                           # Hybrid semantic+importance reranking

Data Sources

Source Format Collection
Notes .txt, .md Drop in input/notes/
Git commits Auto-detected Automatic (all local repos)
Google Calendar API (OAuth) Automatic (incremental sync)
Calendar CSV .csv Drop input/calendar.csv
Google Timeline JSON (Takeout) Drop in input/Semantic_Location_History/

Project Structure

aethermind/
├── setup.py               ← First-run wizard
├── run_pipeline.py        ← Daily orchestrator
├── config.yaml            ← All configuration
├── storage.py             ← SQLite + Qdrant layer (immutable core)
├── normalize.py           ← Raw data → canonical events
├── index.py               ← GPU embedding → Qdrant
├── reflect.py             ← Daily AI reflection
├── ask.py                 ← CLI RAG Q&A
├── app.py                 ← Streamlit web UI
├── collect/
│   ├── notes.py           ← .txt/.md collector
│   ├── git_collector.py   ← Git commit collector
│   ├── google_calendar.py ← Google Calendar API (OAuth + incremental)
│   ├── calendar_collector.py ← CSV calendar fallback
│   └── google_timeline.py ← Google Takeout JSON parser
├── credentials/           ← OAuth tokens (gitignored)
├── input/                 ← Drop your data files here (gitignored)
├── data/                  ← Canonical events + reflections (gitignored)
├── db/                    ← SQLite + Qdrant storage (gitignored)
└── logs/                  ← Pipeline logs (gitignored)

Privacy

  • All data stays local - nothing is sent to external servers
  • No API keys required - LLM runs via Ollama on your machine
  • Embeddings run locally - BAAI/bge-small-en-v1.5 runs on your GPU
  • Google OAuth tokens are stored in credentials/ which is gitignored
  • Personal data directories (input/, data/, db/) are gitignored

Recovery

If something breaks:

# Rebuild vector index (SQLite is source of truth)
python index.py --rebuild

# Re-run normalization only
python normalize.py

# Check what's in the database
python index.py --stats

# View pipeline logs
cat logs/pipeline.log

Roadmap

  • OwnTracks/GPX location import
  • Voice notes via Whisper (local)
  • GitHub API integration (PRs, issues)
  • Weekly summary report (PDF/email)
  • Pattern detection engine ("you abandon projects after 12 days")
  • Mobile companion app (Flutter)

License

MIT - use freely, keep private, share improvements.

Built for people who want AI to know them better than Google does - but only on their own terms.

About

Local-first Personal AI Memory OS - RAG over your entire life. Git, notes, calendar, location. 100% offline. No cloud.

tomaszwi66.github.io/

Topics

aillmlocal-ai-llmollamaopen-sourcepersonal-productivityprivacypythonqdrant-vector-databaserag-chatbotsecond-brainself-hostedstreamlitvector-database

Resources

Readme
MIT license
Activity

Stars

8 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages