AntSeed
Self-hosted peer-to-peer routing layer that routes AI inference requests across an open market of independent providers instead of a central aggregator.
Features
Properties
- Privacy focused
- Decentralized
Features
- Peer-To-Peer
AntSeed News & Activities
Recent activities
AntSeed information
What is AntSeed?
AntSeed is a decentralized peer-to-peer network for AI inference. It moves AI requests the way BitTorrent moves files: no central server, no company in the middle, and no single point that can go down, lock your account, or read your prompts.
Every centralized AI aggregator occupies the same position. It decides which models get listed, routes every request through its own servers, marks up the price, and holds provider payouts until withdrawal. AntSeed removes that position from the architecture entirely.
The Virtual Private Router (VPR) is self-hosted software you run on your own computer. It exposes OpenAI and Anthropic compatible APIs on a local port and routes each request across an open provider market by price, latency, reputation, capability, or privacy. Requests never pass through anyone else's servers. There is no configuration step and no migration: existing tools and agent frameworks are pointed at one local address and keep working unchanged, with automatic fallback when a provider is slow or unavailable.
Because providers set their own prices on an open market rather than accepting a single company's rate card, the same models cost substantially less than official API rates, with savings ranging from roughly 38 to 69 percent depending on the model. A number of models are available at no cost. You pay per request, only for what you use, by card or crypto, with direct non-custodial settlement.
Privacy is structural rather than a policy promise. There is no account and no email, so nothing ties requests back to a user identity. For sensitive work, TEE-verified providers run requests end-to-end encrypted inside a sealed hardware enclave that the provider itself cannot read into.
On the supply side, anyone can serve inference. There is no application and no approval queue. Providers can serve open-source models on their own GPU, resell API capacity, run a custom router, or offer specialised inference, and they build a portable public reputation score that belongs to them rather than to a platform.
The protocol, the router, verification, network data, and payments are all open source.
