A private, offline AI companion that becomes the long-term memory of one grandparent.
Built for a real person. Their stories, recipes, routines, and people live on your laptop — not on someone else's server. Local Gemma 3 (via Ollama) does the talking. Tiger Data (Postgres + pgvector + full-text + Timescale) does the remembering.
Closed chatbots forget. Cloud notebooks feel wrong when the subject is family history. Keepsake keeps the archive at home: hybrid search over growing episodic memory, so every conversation makes the next one more personal.
Stories go in → embeddings land in Tiger Data → hybrid search (vector + keyword) builds the prompt → Gemma 3 answers locally → the exchange is stored so memory grows. Sentry wraps embed, retrieve, and generate.
# 1. Ollama with models already pulled
ollama pull gemma3:4b
ollama pull nomic-embed-text
# 2. Configure secrets (never commit .env)
cp .env.example .env
# set DATABASE_URL (Tiger Cloud) and optional SENTRY_DSN
# 3. Install
python -m venv .venv
# Windows: .venv\Scripts\activate
pip install -r requirements.txt
pip install -e .
# 4. Init schema + seed demo memories
python -m keepsake init
python -m keepsake seed
# 5. Ask something real
python -m keepsake chat "What dal does Dadi make on Sundays?"
python -m keepsake onthisday- Ingest stories / recipes / people notes as markdown (optional YAML header for kind, people, tags).
- Embed locally with
nomic-embed-text. - Store in Tiger Data:
vector+tsvector+ Timescale hypertable oncreated_at. - Retrieve with reciprocal-rank fusion of cosine similarity and full-text rank.
- Answer with Gemma 3 grounded only in retrieved memories.
- Grow — each chat turn is written back into memory.
Optional: point Cursor at .cursor/mcp.json so Tiger MCP can query the same database the agent uses.
| Layer | Choice |
|---|---|
| Chat | Gemma 3 (gemma3:4b) via Ollama |
| Embeddings | nomic-embed-text via Ollama |
| Memory | Tiger Data — pgvector, GIN full-text, Timescale hypertable |
| Observability | Sentry transactions/spans around embed → retrieve → generate |
| UI | Typer + Rich CLI |
python -m keepsake init
python -m keepsake seed
python -m keepsake ingest path/to/memory.md
python -m keepsake chat "Who is Uncle Ravi?"
python -m keepsake onthisday
python -m keepsake status
No cloud LLM calls. Chat and embeddings hit localhost Ollama. The only remote piece is your Tiger database (or a local Postgres with pgvector if you swap the URL). Put real family text in .env-gated deployments you control.
MIT