Hive Inference P2P logo Hive Inference P2P
Open source · Peer-to-peer AI

A swarm of AI workers,
powered by people

Hive is a free, peer-to-peer network for collaborative AI inference. Lend your idle compute, a local model or spare cloud quota, to a project you believe in, and let a swarm of workers think together.

MIT licensed Free forever No accounts Encrypted Local-first

Why Hive exists

“We contribute code. We contribute money. What about compute?”

Right now there is a quiet abundance going to waste. Powerful local models sit idle between prompts. Daily Claude, ChatGPT and Gemini quotas reset unused.

Hive turns that idle capacity into something you can give. You do not need to understand the codebase. You do not need to write a single line of code. You just need to care about a project enough to turn a client on. It is a new kind of open-source contribution.

How it works

An owner, and a swarm of bees

Machines talk directly to each other. One sends a task, the others send results back. No middle-man server, no platform, no account.

Owner
Defines the work
  • Describe a task
  • Pick how many workers
  • Review, merge, accept
Bee
Runs the model
  • Local: Ollama, LM Studio
  • Cloud: 12+ providers
  • Sends the result back
1

The owner describes a task in the desktop app and chooses how many workers tackle it.

2

Connected bees pick it up and run it on their own LLM, local or cloud, with their own key.

3

Results return over the encrypted link to compare across versions, merge, and accept.

The headline feature

Make small models smarter, together

Send one task to several bees, then let Hive merge their answers across rounds. The swarm refines its own work, often landing close to frontier quality while running on everyday hardware.

Round 15 bees, 5 drafts
Round 2merge the best parts
Round 3one polished answer

Built right

Private by design, yours to control

End-to-end encrypted

Every connection uses libp2p with the Noise protocol: ChaCha20-Poly1305, mutual ed25519 auth, forward secrecy. Peers verify each other directly.

Your code never leaves the circle

Working on something closed-source? Keep it off third-party clouds. Run the swarm on local models so nothing leaves the people you trust.

Native desktop apps

Friendly Tauri apps for both owners and bees, plus headless CLIs. Turn it on, connect, contribute. No terminal required.

12+ LLM backends

Local Ollama and LM Studio, or cloud APIs from Anthropic, OpenAI, DeepSeek, Mistral, Groq, Google and more. Each bee brings its own.

Connects through any NAT

Bees make outbound connections that punch through firewalls naturally. Run them anywhere, on any network, in any country. Pairs neatly with Tailscale.

Open and auditable

MIT-licensed Rust, every line on GitLab. No telemetry, no lock-in, nothing phoning home. Fork it, run it, change it.

No catch

This is not a product. It is an idea worth existing.

Hive does not sell anything, host anything, or take a cut of anything. There is no company behind a paywall. Just a prototype built because the concept deserved to be real, and shared in case it is useful to you.

Try it

Clone it and bring a bee online

It is an early prototype, built with Claude under my direction. Honest, hackable, and looking for people who want to tinker.

# grab the source
git clone https://gitlab.com/lexy.callemeyn/hive-inference-p2p.git
cd hive-inference-p2p && cargo build --release

Hive is an experimental prototype shared as-is under the MIT licence. Bring your own LLM and API keys. You stay in control of what your machine runs.