One intelligent shell. Every machine. Verified execution.
Wazir is a cross-platform, AI-native execution environment where traditional operating-system commands and natural-language instructions coexist in one universal terminal—backed by local Qwen models, distributed workers, and evidence-gated verification.
Shell + AI Coexistence
Run ls -la, git status, pipes, and SSH natively in a real PTY—or type natural-language requests. Use ! to force shell, ? to force AI, or automatic deterministic classification.
Local-First Compute
Target 90%+ local inference using Qwen models on Apple Silicon (MLX/Ollama) and NVIDIA DGX Spark / Linux GPUs—keeping source code and credentials off external APIs.
Verified Execution, Never Self-Report
Every AI task passes through PolicyEngine, lease-fenced worker dispatch, and VerificationEngine. Completion requires observed tool/test evidence—never an LLM's claim alone.
Interactive Component Explorer
Click any of the 9 connected Wazir AI OS subsystems to inspect its purpose, capabilities, source location, and interactions.
Interactive Fleet Visualization
Public Demo Mode — Sanitized Sample Topology (Private fleet data requires operator authentication)Downloads & Verified Installation
All artifacts below are generated from verified builds with SHA-256 checksums. No unverified curl | sh pipelines.
Verify Checksum & Install
shasum -a 256 -c dist/SHA256SUMS.txt # Source build (Node 22+ & Rust): git clone https://github.com/whassan007/wazir.git && cd wazir && npm install && npm run build
Interactive 8-Step Onboarding Walkthrough
Simulated environment walkthrough—no real shell commands are executed from the public portal.
Documentation
Complete platform reference for Wazir AI OS.
Quickstart & Platform Install
Install via verified .dmg, .deb, .rpm, Homebrew, or Termux ARM64. Run wa doctor to verify local runtimes.
Configuration & Commands
Use wa shell, wa exec, wa fleet, wa about, wa license, and wa diagnostics. Configure endpoints in ~/.wazir/config.json.
Distributed Execution & Workers
Register workers with WAZIR_REGISTRATION_TOKEN over SSE + lease-fenced execution claims (POST /computers/:id/executions/:id/claim).
Model Provider Setup
Supports Ollama, MLX, LM Studio, vLLM, and authorized remote providers with local-first routing and hardware capability matching.
Security Model & Sandboxing
Realpath workspace containment, PolicyEngine command classification, bearer-token & device-scoped auth, and zero unauthenticated remote shell exposure.
Troubleshooting & Releases
Run wa doctor --detail to inspect runtime health, or review release notes and known limitations in the Downloads section.
Telemetry & Diagnostics Policy
Affirmative Consent Only: Wazir never sends telemetry before explicit acceptance on first launch. Declining disables diagnostics completely without blocking your native OS shell.
Minimum Schema: App version, OS family/arch, error code, failure category, component ID, sanitized exception category, task phase, worker capability, safe model category, timestamp, and random installation ID. Raw shell histories, source files, prompts, env vars, API keys, tokens, and private paths are redacted and never collected by default.
Licensing, Attribution & Commercial Authorization
Licensing Boundaries
Existing Core, CLI, API, Desktop & Mobile: Licensed under AGPL-3.0-only. Prior AGPL rights are preserved.
Selective Qwen Code Transplant: Apache-2.0 components adapted from QwenLM/qwen-code (Alibaba Group / QwenLM).
New Standalone Modules: Source-available, noncommercial license (PolyForm Noncommercial 1.0.0 candidate, pending legal review) paired with separate commercial authorization. Not claimed as OSI-approved open source.
Request Commercial Authorization
Submitting this form records an inquiry for rights-holder review. It does not automatically grant a commercial license.