About ICME Labs
ICME Labs builds zero-knowledge proof infrastructure for verifiable AI.
Jolt Atlas is our state-of-the-art zkML engine — the fastest open-source zero-knowledge machine learning implementation available, using folding schemes to make ZKP fast enough and small enough to run in browsers, consumer hardware, and DePin devices. 3-7x faster than competing implementations.
ICME PreFlight is our verifiable agentic guardrails API. It converts natural language policies into formal logic, checks AI agent actions with mathematical certainty, and wraps the entire process in a succinct zero-knowledge proof — verifiable by any machine in under a second. Don't trust black-box AI security. Verify it.
Under the hood, we have built NovaNet — a peer-to-peer non-uniform incremental verifiable compute (NIVC) prover network. Ultra-fast proving via parallelization, a DeSci layer for specialized provers (zkML, zkFHE), privacy, and aggregation. It works for verifiable AI, scaling blockchains, DePin, and privacy-critical applications.
On the consumer side, we incubated Kinic — a "Plaid for AI memory." Users own their AI memory, prove attributes about it with ZKP, and verify additions and updates cryptographically. Kinic's zero-knowledge memory work is the first specialized prover in NovaNet and runs entirely on ICME's zkML stack.
Math receipts beat black-box trust every time.
Interested in learning more about our tech?
Reach out on Twitter. Or visit our main site at https://icme.io/