One companion, everywhere you are. Every tile shows exactly what ships today, so nothing you click is a dead end.
Signed native installers for Windows, macOS and Linux (.deb and AppImage). Prefer one binary that installs Nunba or writes HART OS to a USB? The universal installer is in the HART OS section below.
Nunba is the app. HART OS is the operating system it was built for: the agent stack, model runtime and hive networking, ready to go. Three clear ways in, so you never have to guess which file to grab.
Make it your OS
Writes the installed image (raw) to a USB, boots as your OS with a writable root, and updates over the air. Flash once.
Installer for WindowsDouble-click, no Python. exe, macOS and Linux, all signed.Try before you commit
Boot the LiveOS ISO from a USB to try HART OS with nothing installed, or install it to disk from there. Works on older BIOS machines too.
Get the live ISOThe installer can flash this for you, or write it yourself.On your existing OS
One command on any machine that has Docker. The full agent runtime in a container, no reboot, no USB.
docker run -d -p 6777:6777 ghcr.io/hertz-ai/hartos:latestEach of the three ways above comes in three editions. Same OS, sized for the machine.
A workstation you sit at. Full desktop, the agent stack, GPU drivers.
The desktop image ships in four parts because GitHub caps assets at 2 GB. Grab all four plus the reassemble script, or just use the installer or a torrent above. Prefer a script over a binary? The raw hart_usb_flasher.py (needs Python 3) is here too. Nobody needs any of this for the paths above.
The same intelligence you install is an endpoint you can build on. Drop the widget into any page, call the API from your stack, or help build the hive itself.
A full agent in any website, one tag. It runs a rate-limited guest trial, so visitors can talk to it with no signup.
<iframe src="https://hevolve.ai/agents/your-agent?plugin=1" style="width:380px;height:560px;border:0"> </iframe>
An intelligence endpoint with a free tier (100 requests a day) and an SDK per surface: chat, analyze, generate, keys. The full SDK, API reference and widget guide are in the docs.
curl https://hevolve.ai/api/v1/intelligence/chat \
-H "Authorization: Bearer $KEY" \
-d '{"prompt":"summarize this"}'Installing is the start. Live paths are linked; the rest are in build and marked honestly, so nothing here is a dead end.
contribute
Lend compute or run a node. Help build HART OS and Nunba in the open.
earn
Seed a Hevolve endpoint or lend compute for Spark. Real potential for students.
jobs
Put agents to work on real tasks, or take on jobs other people need done.
refer
Share Nunba with a tagged link. Every install traces back to you, credited.
Talk, type, share images. Local Qwen3-4B + 0.8B draft model gives sub-second first-token; speculative decoding handles speed.
F5, Kokoro, Indic Parler and CosyVoice give GPU-tiered TTS, with Piper as the CPU fallback. Whisper STT for input.
MiniCPM-V multimodal vision. Toggle anytime; nothing leaves your machine until you ask.
31 adapters: Discord, Slack, Telegram, WhatsApp, Teams, Matrix, Reddit and more. Your AI joins your rooms with your consent.
Your agent gets better from cumulative experience across the network, without sending your data to the cloud.
No subscription. No metered tokens. Source code on GitHub.
OS
Windows 10 / 11 (64-bit) native. macOS and Linux via the universal installer. Android and WearOS on Google Play. iOS nightly.
CPU
4-core x86-64 (Intel / AMD)
RAM
8 GB minimum, 16 GB recommended
Disk
20 GB free (models + cache)
GPU
Optional. NVIDIA 6+ GB VRAM enables the draft model and faster TTS. CPU-only mode works.
Internet
Required for the first-run model download (~6 GB). Runs offline after.
Your AI lives on your machine. You own it. It gets better with you, not for someone else's quarterly numbers.
Download for Windows