Every gesture verification APTOGON runs depends on an AI classification step — deciding whether the motion pattern we captured looks human. That inference has to run somewhere. We chose to run it on Gonka, a decentralized network of independently-operated GPUs, instead of routing every request through a single centralized AI vendor. This wasn't an incidental infrastructure choice; it's directly tied to why the product is built the way it is.
The Name Is Not a Coincidence
APTOGON is a portmanteau: Aptos — the blockchain where every HumanCredential is anchored — plus Gonka — the decentralized AI network that classifies every gesture. Two decentralized layers, one product name. If either layer were a single company's black box, the "no single point of control" story we tell about identity would have a hole in it.
Why Decentralized Compute, Not Big Tech
Today's AI inference runs almost entirely through four companies — OpenAI, Google, Anthropic, Meta. That concentration is a real risk for any product that depends on it: a single company can raise prices unilaterally, restrict access by region, or go down entirely and take every dependent service with it. Export restrictions on advanced GPUs also mean entire countries can be physically locked out of AI compute, regardless of ability to pay. Gonka routes inference across thousands of independently-operated GPUs instead of one company's data centers, which means no single entity can price us out, cut us off, or shut us down.
How the Network Actually Works
- ML-nodes — GPU-equipped servers that run inference for whichever request the network assigns them.
- Transfer Agents — intermediary nodes that route each request to available compute, wherever it physically is.
- Sprint Consensus — the network confirms, in real time, which computation actually happened while serving live requests.
- Proof of Computation — a slice of every request is independently re-computed by a different node specifically to catch dishonest results.
The security model backing this isn't just "trust the network": cryptographic signatures verify each computation in under 10 milliseconds, a node caught cheating loses a share of its staked collateral — a real financial cost for dishonesty — and the entire codebase is open-source and has been independently audited by CertiK.
Why 'Cheap' Matters — Not Just for Us
Gonka's inference cost runs at a small fraction of a centralized provider's list price for comparable models, because no single company's margin is baked into every request. That matters concretely for APTOGON: gesture verification has to run at scale, for free-tier users, without our costs scaling in lockstep with a Big Tech price list we don't control. Cheap, distributed compute is what makes a genuinely free verification tier possible in the first place.
This is the same principle behind APTOGON's own architecture: zero-PII, on-chain proof, no single company holding the keys to whether you can prove you're human. We picked Gonka because it's built on that principle too.
Respect to the Gonka team and community for building infrastructure that makes this possible — we're glad to be running on it.
Read more about Gonka at gonka.ai →