owlette
ai-powered fleet management for Windows applications
live app • documentation • download agent
owlette is a cloud-connected system for monitoring, managing, and deploying software across fleets of Windows machines — from anywhere. a lightweight Python agent runs on each machine as a Windows service (hosted by a small Rust supervisor, owlette-host), reporting metrics and executing commands. a modern web dashboard gives you real-time visibility and control over your entire fleet, backed by Firebase and Cloud Firestore.
built for teams running digital signage, media servers, kiosks, TouchDesigner installations, and any Windows application that needs to stay running.
features
real-time monitoring
live CPU, memory, disk, GPU, and network metrics. process health tracking with crash detection. remote screenshots and live view. historical metrics with sparkline charts.
remote deployment
push software silently to any number of machines. supports NSIS, InnoSetup, MSI, and custom installers. save deployment templates, track progress in real-time, and cancel mid-install.
hoot ai
LLM-powered fleet management with natural language. AI executes commands on agents via tool-calling across 3 tiers — from read-only diagnostics to privileged operations. screenshot analysis, autonomous crash investigation, and multi-provider support (Anthropic + OpenAI).
multi-site management
organize machines by location, department, or project. three-tier role-based access control (member / admin / superadmin) with site-level permissions. multi-user accounts, per-site admin delegation, and platform-wide administrator roles.
roost — project distribution
drag a project folder onto the dashboard and it syncs to your fleet. files are split into 4 MiB content-addressed chunks, hashed in the browser, and uploaded once — re-deploys and machines that already hold a chunk skip it entirely. every deploy is an immutable version you can roll back to in one click, and agents verify every chunk by SHA-256 before assembling files atomically into an allowlisted destination.
roost is the only distribution system — the legacy by-URL path (point the fleet at a hosted ZIP and let the agent download and extract it) has been removed.
alerts & notifications
configurable threshold alerts for system metrics. email notifications with branded templates. webhook integrations for external platforms. activity logging across your entire fleet.
full documentation →
architecture
control and state flow through Cloud Firestore — there is no direct connection between agents and the dashboard. Firestore acts as the message bus, with Next.js API routes handling authenticated control actions.
bulk file bytes do not go through Firestore. roost chunks and version bodies live in object storage (Cloudflare R2); browsers and agents read and write them directly using short-lived, single-object signed URLs minted by the API.
agents cloud dashboard
+--------------+ +----------------+ +-----------------+
| machine a | --> | | <-- | |
| machine b | --> | Firestore | --> | Next.js web |
| machine c | <-- | | | application |
+--------------+ +----------------+ +-----------------+
|
+----------------+
| Firebase Auth |
+----------------+
- agent — Python Windows service hosted by
owlette-host. monitors processes every 5s, sends heartbeats + metrics on an adaptive interval (5s with the desktop window open, 30s while processes run, 120s idle), executes commands, works offline.
- dashboard — Next.js 16 web app. real-time Firestore listeners, a REST API with OpenAPI documentation.
- firestore — real-time NoSQL database. state sync, command relay, and roost version pointers.
- object storage — Cloudflare R2. content-addressed roost chunks and version bodies under a per-site prefix, reachable only through signed URLs.
- hoot ai — LLM chat with tool-calling capabilities relayed through Firestore to agents.
quick start
hosted (fastest)
- create an account at owlette.app
- create a site to organize your machines
- download the agent installer from the dashboard
- run the installer on your target Windows machine
- a 3-word pairing phrase appears — authorize it from the dashboard or your phone
- your machine appears in the dashboard within 30 seconds
full setup guide →
self-host / develop from source
setting up the repo — toolchain, the agent's Python venv, local env files, and running the agent, desktop app, and dashboard from source — is covered in the maintainer quickstart.
screenshots
tech stack
| component | technology |
|---|
| dashboard | Next.js 16, React 19, TypeScript, Tailwind CSS 4, shadcn/ui |
| agent | Python 3.11, Windows service hosted by owlette-host (Rust), psutil, pywin32 |
| desktop app | Tauri 2 (Rust), React 19, TypeScript, Tailwind CSS 4 |
| database | Cloud Firestore (real-time NoSQL) |
| auth | Firebase Auth, WebAuthn/Passkeys, TOTP 2FA, device code pairing |
| ai | Anthropic + OpenAI via AI SDK, tool-calling |
| email | Resend (branded dark-theme templates) |
| hosting | Railway (web, with a Vercel standby behind Cloudflare load balancing), Firebase (Auth, Firestore, Cloud Functions), Cloudflare R2 (roost storage) |
documentation
full documentation is available at owlette.app/docs.
contributing
contributions are welcome! please open an issue or submit a pull request.
guidelines:
- fork the repo and create a feature branch from
dev
- use conventional commits (
feat:, fix:, docs:, etc.)
- run the tests for what you changed before submitting — see the maintainer quickstart
- use
node scripts/sync-versions.js X.Y.Z for version bumps
- all PRs merge to
dev first, then dev → main for production
open an issue →
license
this project is licensed under the Functional Source License, Version 1.1, Apache 2.0 Future License (FSL-1.1-Apache-2.0). you may freely use, modify, and self-host owlette for any purpose other than a competing commercial product or service. two years after each release, that version automatically converts to Apache License 2.0.
