DEV Community

Cover image for Introducing OpenPawz: A Native Desktop AI Platform That Runs Fully Offline
openpawz
openpawz

Posted on

Introducing OpenPawz: A Native Desktop AI Platform That Runs Fully Offline

#ai #rust #opensource #tauri

Every AI tool I tried had the same trade-off: send your data to the cloud to get anything useful. I wanted something different — a native desktop app where I own my data, pick my providers, and the AI still has real power.

So I built OpenPawz.

What is OpenPawz?

OpenPawz is an open-source, native desktop AI platform built with Tauri v2 (Rust backend, TypeScript frontend). It ships as a ~5 MB binary — no Electron, no Node.js backend, no open network ports.

It runs fully offline with Ollama, connects to any OpenAI-compatible provider (OpenAI, Anthropic, Gemini, DeepSeek, Grok, Mistral, OpenRouter, and more), and gives you multi-agent orchestration, 75+ built-in tools, and 11 channel bridges — all controlled from your desktop.

GitHub: github.com/OpenPawz/openpawz

Why Another AI Tool?

Three things bothered me about existing solutions:

  1. Privacy — Most AI tools phone home. Your prompts, your data, your credentials — all flowing to someone else's servers. OpenPawz has zero telemetry, zero open ports, and encrypts all credentials with AES-256-GCM in your OS keychain.

  2. Lock-in — Most tools chain you to one provider. OpenPawz supports 10 providers out of the box plus any custom OpenAI-compatible endpoint. Switch models mid-conversation. Run local and cloud simultaneously.

  3. Control — AI agents doing real things on your machine is scary. OpenPawz puts 7 security layers between the agent and your OS. Every tool call flows through the Rust engine where it's classified, risk-scored, and — for anything with side effects — shown to you for approval before it runs.

Architecture at a Glance 

Frontend (TypeScript)                  Rust Engine
┌──────────────────────┐              ┌────────────────────────────────┐
│ Vanilla DOM · 20+ views │◄── IPC ──►│ Tauri commands                  │
│ Feature modules         │   (typed) │ Channel bridges                 │
│ Material Icons          │           │ AI providers                    │
│                         │           │ Tool executor + HIL approval    │
└──────────────────────┘              │ AES-256-GCM encrypted SQLite    │
                                      │ OS keychain · Docker sandbox    │
                                      └────────────────────────────────┘
Enter fullscreen mode Exit fullscreen mode

No Node.js backend. No gateway process. No open ports. Everything flows through Tauri IPC.

The codebase is ~86k LOC (39k Rust + 35k TypeScript + 12k CSS), with 600+ automated tests across a 3-job CI pipeline: Rust (check + test + clippy with -D warnings), TypeScript (tsc + eslint + vitest + prettier), and Security (cargo audit + npm audit).

Key Features

Multi-Agent System

Create unlimited agents with custom personalities, models, and tool policies. Agents can:

  • Delegate tasks via boss/worker orchestration
  • Spawn sub-agents at runtime
  • Communicate through direct messages, broadcasts, and squad channels
  • Form squads with coordinator roles for collaborative tasks

7-Layer Security Model

The agent never touches the OS directly. Every tool call goes through the Rust engine:

  1. Prompt injection scanner — Dual TypeScript + Rust detection, 30+ patterns
  2. Command risk classifier — 30+ danger patterns across 5 risk levels (Critical auto-denies by default)
  3. Human-in-the-Loop approval — side-effect tools require explicit user approval
  4. Per-agent tool policies — allowlist, denylist, or unrestricted per agent
  5. Container sandboxing — Docker isolation with CAP_DROP ALL, memory/CPU limits, network disabled
  6. Browser network policy — domain allowlist/blocklist prevents data exfiltration
  7. Credential vault — OS keychain + AES-256-GCM; keys never appear in prompts

11 Channel Bridges

Telegram · Discord · IRC · Slack · Matrix · Mattermost · Nextcloud Talk · Nostr · Twitch · WebChat · WhatsApp

Each bridge includes user approval flows, per-agent routing, and uniform start/stop/config. The same agent brain, memory, and tools work across every platform.

75+ Built-in Tools

File operations, shell execution, web fetch, browser automation, email (IMAP/SMTP), GitHub integration, DeFi trading (7 EVM chains + Solana), research workflows, task management, inter-agent messaging, and more — all with encrypted credential injection and risk classification.

Hybrid Memory

BM25 + vector similarity search with Ollama embeddings, MMR re-ranking for diversity, temporal decay (30-day half-life), plus a Memory Palace visualization UI.

Quick Start 

git clone https://github.com/OpenPawz/openpawz.git
cd paw
npm install
npm run tauri dev
Enter fullscreen mode Exit fullscreen mode

Prerequisites: Node.js 18+, Rust (latest stable), and Tauri prerequisites.

That's it. No cloud account, no API key required for local models. Add a provider API key under Settings if you want to use cloud models.

The Stack

Layer Technology
Framework Tauri v2
Backend Rust (async with Tokio)
Frontend TypeScript (vanilla DOM — no React, no Vue)
Database SQLite (21 tables, AES-256-GCM encrypted fields)
Bundler Vite
Testing vitest (TypeScript) + cargo test (Rust)
CI GitHub Actions (3 parallel jobs)

What's Next

  • v0.2 — Stable release binaries, Homebrew/AUR/Snap/Flatpak packages
  • v0.3 — Plugin API, PawzHub marketplace launch
  • v0.4 — Mobile companion (iOS/Android), cloud sync opt-in
  • v1.0 — Production-ready with full enterprise hardening

Get Involved

OpenPawz is MIT-licensed and contributions are welcome:

If you've been looking for an AI platform that respects your privacy, doesn't lock you in, and gives you real control — give OpenPawz a try and let me know what you think.

Top comments (0)