🌸 21st Century Counterculture

Run AI in a herd,
not a data center

Local by default. The network when you need more.

Download the app and run open models on your own machine — free and private. Plug in the engine you already use — Ollama, vLLM, LM Studio, llama.cpp, Exo — and reach a global peer-to-peer network when you need a bigger model. One OpenAI-compatible endpoint. No gatekeeper, no rate limits.

Download for your OS
v0.2.0 · local-first · macOS, Windows & Linux

Builds aren’t code-signed yet, so your OS may warn on first launch. On macOS: right-click the app → Open (or run xattr -dr com.apple.quarantine /Applications/OpenHydra.app). On Windows: More info → Run anyway. Signed builds are coming. Prefer the terminal? curl -fsSL https://openhydra.co/install.sh | sh

Live now · v0.2.0

Cross-continent inference over a direct, hole-punched peer link

A pure-Rust agent, an OpenAI-compatible endpoint, and a libp2p network that punches through CGNAT to connect peers directly — falling back to an encrypted relay when it can’t. Proven between machines on different continents.

🍏

Got a Mac Mini running local models?

Apple Silicon is a quiet, efficient inference machine. Point OpenHydra at the model you already run in Ollama or LM Studio, leave it on in the background, and serve the herd while you sleep — earning priority for when you need a model bigger than your own.

Why OpenHydra

The network for everyone’s local AI

Peer-to-peer inference — BitTorrent, for AI. Bring your own engine, reach the whole herd.

🔌
Plug in what you already run

Ollama, vLLM, LM Studio, llama.cpp, Exo — any engine, any OS, any GPU. A thin Rust agent joins your existing stack to the network. Keep the tools you know; reach the whole herd.

🦙
Reach models your machine can’t

When you ask for a model you don’t have, the request routes to a peer who already has it loaded — it runs there at full local speed and streams straight back to you.

🤝
Free, by reciprocity

Give-to-get: serve inference while you’re idle to earn priority when you consume. Contribute when you can, draw on the herd when you need it.

The five-year-old version

OK but what actually is this?

Glad you asked. It’s honestly not that complicated once you stop calling it “AI infrastructure”.

🗺️
Someone already has the model

Somewhere in the herd, the model you want is loaded and running on a real machine right now. The network keeps a live map of who’s serving what.

🌊
Route the question to them

Like BitTorrent finds a peer who has the file, OpenHydra finds a peer who has the model. Your prompt goes there, the answer streams back — no central server in the middle.

🔍
And bad answers get caught

Responses are spot-checked and cross-verified across peers. Good llamas build reputation and get routed more work; cheaters get downranked until the swarm stops trusting them.

Small models run on a single laptop. Bigger ones live on whoever has the hardware. You install once and you’re in — no VRAM math, no “which model fits” anxiety.

How it works

Three steps. Zero guesswork.

Download the desktop app, run an open model locally, and flip a toggle to join the herd when you want more. One switch, zero restart.

01
Download and run — local by default

Install the app for macOS, Windows, or Linux. It runs an open model on your own machine — free, private, offline-capable. No account, no config.

02
Join the herd when you want more

Flip to network mode to reach bigger models other peers are serving — or reach your own rig at home from anywhere. It’s all behind the same OpenAI-compatible endpoint your tools already speak.

03
Serve while you idle

Leave it running and serve tokens to the herd in the background — that’s how you earn priority, and how the network stays alive. Routing, verification, and encryption are handled for you.

What you get

Features, listed professionally

We also have proper docs, but here’s the version where we’re allowed to be slightly smug.

⚡
Drop-in OpenAI API

Change one URL and your existing code works. /v1/chat/completions with SSE streaming, plus /v1/embeddings and /v1/models — the OpenAI SDK talks to the whole network unchanged.

🧠
Bring your own engine

Ollama, vLLM, LM Studio, llama.cpp, Exo — OpenHydra wraps them with a thin adapter. No new runtime to learn, no model re-hosting. Keep the exact stack you already have.

🔗
Native libp2p networking

A Rust libp2p stack: Kademlia DHT discovery across three continents, QUIC + TCP, DCUtR hole-punching, and Circuit Relay v2 fallback. No central broker. If one bootstrap goes down, the llamas find another way.

🖥
Desktop app for Mac, Windows & Linux

A Tauri v2 desktop app you can download today: chat with a local model in one click, or flip a toggle to join the global network — no restart, no terminal required.

🛡
Encrypted, authenticated transport

Ed25519 peer identities; every connection encrypted in transit via libp2p Noise and QUIC. Even over a relay the stream stays end-to-end encrypted — the relay just forwards bytes, it never sees your prompt.

🏅
Reputation & verification built in

Redundant execution and content hashing catch bad actors; reputation decides who gets routed work. Economic security without a token — your standing is the bond.

Three ways to run it

Your machine, your rig, or the herd

Same app, same API. Start fully local and reach further only when you want to.

💻
On your machine

Run an open model locally. Free, private, works offline. Nothing leaves your laptop — the perfect default for everyday chat and code.

🏠
On your own remote rig

Left a beefy box at home? Reach it from your laptop anywhere — OpenHydra punches through CGNAT so your GPU is one endpoint away, no port-forwarding, no VPN to babysit.

🌐
On the herd

Need something bigger than anything you own? Route to a peer who’s serving it. Reciprocity keeps it free; reputation keeps it honest.

The herd’s catalog is simply the union of what everyone is serving — pull a model in Ollama or point at a vLLM server and it’s instantly reachable on the network.

Standing on the shoulders of giants

We didn’t invent this. We just added llamas.

OpenHydra builds directly on two brilliant ideas. We want to be upfront about our inspirations, because intellectual honesty is cool (and mandatory if you don’t want to get ratio’d on HackerNews).

Academic inspiration
🌸 Petals

“Run large language models at home, BitTorrent‑style.” Petals proved volunteer compute can serve real LLM inference across the internet. We took that spirit and added a desktop app and a very strong llama motif.

petals.dev →
Protocol inspiration
🌊 BitTorrent

Since 2001, BitTorrent has proved you can distribute enormous things to billions of people without a central server — and that reciprocity keeps a network alive. Same energy, applied to tokens instead of files.

bittorrent.com →

🦙 Fun llama fact #1: Real llamas are pack animals because they share the load across the herd. The weakest llama doesn’t carry the whole tent. This is also the core architectural principle of OpenHydra.

🦙 Fun llama fact #2: A group of llamas is called a herd. OpenHydra’s network of peers is also called a herd. We are very consistent in our metaphors and proud of this.

🦙 Fun llama fact #3: The Hydra in Greek mythology had multiple heads — cut one off and two grow back. Our bootstrap nodes work the same way. (Please don’t cut our bootstrap nodes.)

🦙 Fun llama fact #4: Llamas can spit up to 10 feet when stressed. Our nodes politely return HTTP 503 instead. Both are valid responses to being overwhelmed.

Your laptop is a supercomputer waiting to happen.

It’s sitting there doing nothing useful right now. Download the app and it runs AI locally in one click — then joins a global herd of volunteers when you need more. No cloud. No lock-in. Just llamas.

Download for your OS