Your personal wise companion that grows with you.
· AGPL-3.0 · built on Bun
A self-hosted AI companion that runs on your own box, remembers what matters in a Markdown vault, and grows a new skill — a new head — for every procedure you teach it. Local model by default; your data stays yours.
→ balaur.org · Design
Current architecture is deliberately small:
balaur — branded Pi terminal session.pi/ — local Pi package: extensions, native skills, theme, model config.llama-apu — local model service on 127.0.0.1:8082.$VAULT_PATH — private Markdown vault outside the repo.There is intentionally no core HTTP gateway / always-on balaur-agent service now. Future
Signal/web/HTTP bridges should be Pi extensions attached to a long-lived terminal/tmux session, so
there is one live agent session and no split-brain model state.
git clone https://github.com/alexradunet/balaur.git ~/balaur
cd ~/balaur
./install.sh
balaur
Health check:
make doctor
llama-apu, Qwen3.6 by default)./login + /model.pi/skills/.bun test → commit → push → /reload.bin/balaur # terminal launcher
pi/ # Pi package: extensions, skills, themes, model config
pi/extensions/ # memory, vault, Balaur terminal UI
pi/skills/ # Pi-native Markdown skills
pi/themes/ # Balaur theme
docs/ # install, troubleshooting, maintenance docs
scripts/seed-pi-config.sh # registers the local Pi package in ~/.pi/agent
scripts/setup-local-llm.sh # installs/configures llama-apu
scripts/doctor.sh # make doctor
make doctor # tooling/deps/tests/local-model/Pi-config health check
make test # bun test
make smoke # Pi SDK smoke check
make index # rebuild vault/memory FTS index
make cli # open the terminal via bin/balaur
Inside the terminal:
/reload # reload extensions/skills/theme/prompts
/model # switch model
/login # optional frontier login
/skill:doctor # run the doctor playbook
Personal data stays in the vault by default. The local model is the default. Frontier models are manual and opt-in; do not move personal memory to them automatically.