Skillget

nazar-studio

agent
Security Audit
Warn
Health Pass
  • License — License: AGPL-3.0
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 10 GitHub stars
Code Warn
  • process.env — Environment variable access in extensions/brand.test.ts
  • process.env — Environment variable access in extensions/local-llm.test.ts
  • process.env — Environment variable access in extensions/local-llm.ts
  • network request — Outbound network request in extensions/local-llm.ts
  • process.env — Environment variable access in extensions/memory.test.ts
  • process.env — Environment variable access in extensions/memory.ts
  • process.env — Environment variable access in extensions/personality.test.ts
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Local-first personal AI environment built as a Pi-native TypeScript extension product with Markdown memory, local models, tools, and terminal UI.

README.md

Nazar

Your personal wise companion that grows with you.

License: AGPL-3.0
Node
Pi package

Nazar is a sovereign, local-first personal agent that lives on your own box. It remembers
the facts of your life in durable Markdown memory, learns skills that act on them, and keeps
everything in open files you own. It ships as a single Pi
package — install it and its extensions, native skills, theme, persona, and local model load
themselves.

It is named for the nazar — the watchful blue eye of old Balkan fairy tales that wards off
the evil eye. Each new skill it learns is one more eye watching one more corner of your life.

Your life is not a product. The record of your life — what you ate, where you walked, what
you felt, who you love — should answer to you alone.

What makes it different

Most personal AI is rented memory in someone else's cloud. Nazar is the opposite: a companion
whose source of truth is a vault you own.

  • Yours by construction — journal, diet, sport, and durable facts live as plain Markdown
    in a vault you can read, grep, back up, and move. Your data never trains anyone's model.
  • Local model first — the terminal defaults to a local model served by
    llamafile. Personal context stays on the box
    unless you deliberately switch to a frontier model.
  • Two rails, one ethos — facts ride a disposable SQLite FTS5 index over Markdown;
    procedures ride Pi's own native skill system. Each on the simplest rail that fits.
  • Grows by approval — when a recurring need emerges, Nazar suggests a new skill; you
    approve; git records the change. No silent self-modification.
  • Terminal-first — one live Pi session with a hand-built, old-school RPG terminal UI
    (the Basm pixel-art design language). No gateway service, no container stack, no cloud.

Quick start

Requires Node 23.4+ (Node 24 LTS recommended) — Nazar's memory index uses the built-in
node:sqlite FTS5, which is flag-free from 23.4.

# 1. Install Pi (the terminal agent runtime)
npm install -g --ignore-scripts @earendil-works/pi-coding-agent

# 2. Install Nazar as a Pi package
pi install npm:pi-nazar-studio

# 3. Run
pi

On first run the local-llm extension lazily downloads llamafile and the default GGUF model
and serves it on 127.0.0.1:8082. Your private vault is created at $VAULT_PATH
(default ~/.local/share/nazar). Full guide: docs/INSTALL.md.

How it works

Layer What it is
Extensions (extensions/) Pi registration + orchestration: brand/UI, terminal-font, persona, memory, vault, local-llm provider, channel gateway.
Memory (facts) Plain Markdown under vault/memory/, organized by the Johnny Decimal memory structure, indexed by a disposable FTS5 accelerator over title, body, tags, and whenToUse. Relevant notes are auto-recalled into each turn.
Skills (procedures) Pi-native skills under skills/, discovered and invoked with /skill:name. The agent can propose new ones for your approval.
Local model Registered in-process from models.json by extensions/local-llm.ts; nothing is hand-seeded into your Pi config.
Persona + rules SYSTEM.md (who Nazar is) and AGENTS.md (how the code is built), injected at before_agent_start.

Useful commands inside Pi

/reload                     reload extensions, skills, prompts, theme
/model                      switch model; frontier models are manual/opt-in
/local-llm                  manage the local llamafile + whisperfile runtime
/nazar-ui low|medium|high   switch terminal avatar rendering quality
/nazar-terminal-font        check/configure a terminal for high-fidelity avatars
/skill:doctor               run the health-check playbook

Project layout

extensions/   Pi extensions — registration + orchestration only
lib/          Shared, unit-testable logic (memory, sqlite, paths, provider, ui/, gateways/)
skills/       Pi-native skills (doctor, open-websearch, terminal-font)
themes/       Generated Pi theme (nazar.json)
assets/       Avatar PNG masters + bundled fallback fonts
design/       The "Basm" design system: tokens, palette, avatar specs
docs/         Install, troubleshooting, persona, governance, security
scripts/      Dev tooling (token build, sprite generation, backup/restore)
models.json   Local provider + model registration
SYSTEM.md     Nazar's persona (injected into the system prompt)
AGENTS.md     Engineering conventions (injected into the system prompt)

Documentation

Develop

git clone https://github.com/alexradunet/nazar-studio.git
cd nazar-studio
npm install
npm run typecheck                 # tsc --noEmit
npm test                          # vitest
npm run build:tokens -- --check   # verify generated design tokens are in sync
pi -e .                           # load the working tree into Pi for one run

Engineering conventions live in AGENTS.md. The design system is documented
in design/README.md.

License

AGPL-3.0-or-later. Your vault (personal data) is not in this repo — code
stays public, data stays yours. See NOTICE for attributions.

Built in the open, in Brașov.

Reviews (0)

No results found