Seed Round: $0.25/FRAC — Listing at $1.00 (300% ROI) — Buy Now
Fractal AI Neural Engine

AI you can prove.

FANE isn't the biggest model. It's the one you can verify, self-host, and seal with a post-quantum attestation. A whole engine in ~80 MB — no data center required.

The demo talks to the live public node at api.fractalai.net.co. It degrades gracefully if the node is busy.

Verifiable

Every answer can be sealed with a Dilithium-2 (NIST FIPS 204) signature you re-verify yourself — resistant to known classical & quantum attacks, not 'unbreakable'.

Sovereign

Phi-compressed weights run on-chain in ~80 MB RAM. No cloud API, no GPU farm — any node can reconstruct and run the model.

Learning

ATHENA turns each conversation into a training pair and persists weights on-chain. It gets sharper over time without renting datacenter GPUs.

Honest scope: FANE is a lightweight engine, not a frontier LLM. Its edge is trust and sovereignty, not parameter count. For open-domain depth it can be enriched by a larger model while keeping the on-chain, verifiable core.

Ask FANE — then notarise the answer

Ask anything about FractalAI. Then seal that exact answer with a post-quantum signature and re-verify it — the thing no chatbot can do.

Under the hood

A phi-cascade of on-chain systems, each verifiable and self-hostable.

ATHENA

Transformer core with phi-sigmoid activation. Learns from every conversation; weights persist on-chain.

PHIRAG

Blockchain-native retrieval. Knowledge scored by a golden-ratio relevance function, not a black box.

ALETHEIA

Real-time truth scoring — extracts claims and cross-references the on-chain knowledge store.

FOCI

On-chain intelligence proofs — each inference can emit a verifiable proof hash.

Phi-compression

Weights quantized with PHI=1.618 + a compile-time lookup table. The whole model fits in ~80 MB RAM.

Post-quantum

CRYSTALS-Dilithium signatures + Kyber encryption (NIST-standardized) protect every operation.

Bring verifiable AI into your stack

Use the engine directly, or build on the products above. Every answer can carry a proof your auditors and users can check themselves.