Part of the H.A.N.S.A. Institute — an independent effort to keep AI inference cheap, sovereign, and accessible.
MVP Demo · Weeks Out

Your hardware. Your AI.
Connect when you're ready.

Hanseatic Alliance for Networked Sovereign Architecture

The easiest way to run an LLM on your own hardware — a private, OpenAI-compatible endpoint, fully offline. When you choose to, share spare capacity with the community swarm and route your own requests across the network, geo-scoped to your rules.

Request early access Learn more
Default
Self-host
Private local endpoint. No network. No accounts. Fully offline.
Optional
Private group
Your team or home lab. Allowlist only.
Opt-in
Community swarm
Share capacity. Pick your geo scope: country, EU, or world.
Who it's for

One network, three kinds of participant

The same software serves radically different needs — because self-sovereignty, research freedom, and regulatory compliance are not in conflict.

For Users & Builders
Your hardware. A drop-in API.

Install H.A.N.S.A. on a gaming PC or Apple Silicon machine. You get a private, OpenAI-compatible endpoint — no API keys, no usage bills, no data leaving your device. Opt into the community swarm when you want more capacity.

  • OpenAI-compatible endpoint — drop-in replacement
  • Runs Mistral, Llama, Phi-3, Hermes out of the box
  • Agentic batch splitting across peers (opt-in)
  • Contribute idle compute, consume from the network
  • Planned: GPU temp-aware routing — long tasks fan out to peers when your silicon runs hot, lowering sustained temperature, cutting electricity draw, and extending hardware life
🔬
For Researchers
A substrate for distributed inference research.

The long-term thesis: combining smaller models can outperform single large ones. Mixture of Agents, self-consistency, speculative ensembling — this research exists but lacks a real-world substrate. H.A.N.S.A. aims to provide it.

  • Fan a single prompt to N peers, aggregate results (MoA)
  • Agentic workloads distributed across specialized small models
  • Opt-in anonymized telemetry for research analysis
  • Moving inference out of big datacenters — studied at scale
🛡
For Regulators & Enterprises
Geo enforcement at the infrastructure layer.

Data residency is designed to be enforced by the routing layer, not left to a policy checkbox. The coordinator filters peers so a node is prevented from sending or receiving inference traffic outside its configured boundary — and is designed so the coordinator itself never sees inference data.

  • Geo-scope: country / EU / world — enforced at routing, by design
  • Coordinator filters peer lists; designed never to touch inference traffic
  • Signed updates via Sigstore with public Rekor audit trail (planned)
  • JSONL audit log — policy decisions included, grant-friendly
Architecture

How it works

Built on proven open-source primitives. The novel parts are the policy engine, the batch splitter, and the routing layer.

01 // INSTALL

Hardware → running model

The installer detects your GPU (CUDA / Metal / ROCm) and suggests the right model tier. One command. Works offline. No accounts required.

02 // ENDPOINT

OpenAI-compatible API

LiteLLM Proxy exposes a standard OpenAI endpoint on your machine. Any app that calls OpenAI works with H.A.N.S.A. — no code changes.

03 // CONNECT

Opt into the swarm

Enable sharing. Set your geo scope. The coordinator matches you with compatible peers — applying your policy before any peer sees your node.

04 // SPLIT

Agentic batch dispatch

The batch splitter detects parallel agentic calls and fans them across peers automatically. The same primitive enables multi-model ensemble research.

Built on open standards

H.A.N.S.A. builds on Ollama for local inference, LiteLLM Proxy for OpenAI compatibility, and Iroh for encrypted peer transport with NAT traversal. The coordinator handles discovery and geo-policy — and is designed to stay completely off the inference data path.

The novel layer is what we build on top: geo-policy enforcement, latency-aware peer ranking, the agentic batch splitter, signed update distribution, and a public network dashboard.

ollama litellm-proxy iroh-p2p fastapi sigstore maxmind-geo ed25519 prometheus
// hansa-router v0.1 [H.A.N.S.A.]
──────────────────────────────────
request → litellm_proxy
model: "mistral:7b"
type: agentic_batch[4]

splitter → fan_out(
peers: policy_filtered
scope: "eu"
)

✓ peer hns-waw-04 (PL)
✓ peer hns-ber-11 (DE)
✓ local ollama:mistral

coordinator: policy only
inference data: off-path
audit log → writing...
Timeline

From self-hosted MVP to network

MVP demo in weeks. Community alpha by Q3. Research program by Q4.

Done
Foundation
  • Concept & thesis
  • Angel pre-seed secured
  • Architecture design
Now
MVP Demo
  • install.sh → offline node
  • 2 EU nodes forwarding
  • Geo policy rejects out-of-scope
  • Hermes batch fan-out
  • Public network map
  • Signed update verified
Q3 2026
Community Alpha
  • Open node enrollment
  • Contribution credits (earn by sharing)
  • GDPR attestation layer
Q4 2026
Research Program
  • Ensemble inference (MoA)
  • Academic partnerships
  • Seed raise
2027
Production
  • Thermal-aware task routing
  • Self-sustaining contribution economy
  • Open ecosystem
  • Research publications
Demo soon

Get early access

Join the builders and researchers shaping a community-run AI network.

No spam. Demo invites and major milestones only.