A local AI companion with persistent memory. Designed for old hardware.
No cloud. No telemetry. No GPU required. Just you, an Ollama model, and a companion that remembers.
- Linux x86-64
- Python 3.11+
- Ollama (with at least one model pulled)
- 4GB RAM minimum (8GB recommended)
- 2GB free disk
git clone https://github.com/your-username/gwen.git
cd gwen
./setup.sh
gwenpython3 -m venv venv
source venv/bin/activate
pip install -e .
gwen init
gwengwen Start interactive chat
gwen ask "question" Ask one question, get one answer
gwen remember "fact" Store a fact to memory
gwen search "query" Search past memories
gwen status System health check
gwen sessions List recent conversations
gwen init Re-initialize config and database
/help Show commands
/exit Quit
/stats Memory and session stats
/remember Store a fact mid-conversation
Edit ~/.config/gwen/personality.md to change how Gwen speaks.
The default is based on the companion personality from personality.example.md.
Config file: ~/.config/gwen/config.toml
| Key | Default | Description |
|---|---|---|
ollama_host |
http://localhost:11434 |
Ollama API endpoint |
model |
llama3.2:1b |
Chat model |
embed_model |
all-minilm:22m |
Embedding model |
temperature |
0.1 |
Response creativity |
max_tokens |
150 |
Max response length |
num_ctx |
1024 |
Context window size |
All data lives under ~/.local/share/gwen/:
memory.db— SQLite database with all memories, embeddings, and conversationshistory/— JSON conversation logs
MIT