Company Brief
A concise operating document covering the company's origin, product architecture, AI compute, protocol vision, risk boundaries and team formation. All product stages are stated explicitly. This brief is not an offering document.
Brief item
Detail
Discipline
Position
Ztakora is an early-stage protocol and product studio working on two connected problems: making AI runs verifiable, and making the data and hardware underneath them permanent and sovereign.
Company
Origin - verifiable-compute thesis
The company was formed in January 2026 around a verifiable-compute thesis: a network where models can be published, executed and audited without trusting a single provider.
Research
Origin - on-chain runtime and Grid contract surface
Ztakora committed to an on-chain AI track and began building the ErisAI runtime alongside the Ztakora Grid contract surface.
Engineering
Origin - product direction, protocol architecture and compute PoC
Runtime work sharpened the company's approach to model provenance, compute receipts, hardware specification and evidence-led release.
Product
Origin - selective technical hiring through paid trials
Senior engineers join through technical discussion, a working session and a paid real-task trial rather than an open applicant pool.
Operations
Portfolio - ErisAI
Decentralized AI engineering: train, evaluate and run models as on-chain contracts with verifiable receipts.
Runtime
Portfolio - Ztakora Grid
An AI-native Layer 1 that anchors model provenance, dataset references, compute receipts and access policy.
Protocol
Portfolio - Ztakora Node
Standardized node machines and GPU subnets run by independent providers across data centers.
Hardware
Portfolio - Ztakora Drive
The data layer that keeps datasets, weights and artifacts retrievable behind content hashes and access policy.
Storage
AI compute
ErisAI runs training and inference as jobs that produce a compact receipt recording model version, dataset slice, runtime build, hardware class, cost and outcome.
Compute
Protocol thesis
Publishing is open, certification is earned: every job defines its operating limits, telemetry, success conditions and rollback path before it earns wider hardware scope.
Protocol
Later-stage capabilities
GPU-subnet training, permissionless scheduling and broader model coverage follow registries, receipts, access policy and evaluation - they do not precede them.
Roadmap
Boundary - not an offering document
This website provides product and research information. It is not financial, legal or investment advice and does not solicit an investment.
Disclosure
Boundary - design targets, not measurements
Model, latency, cost and throughput figures are design targets, not measured production results, until published with a verifiable compute receipt.
Evidence
Boundary - bounded experiments only
Experimental subnets and models use bounded compute, permissions, budgets, review dates and rollback paths.
Operations
Boundary - registration asserts nothing
Registering a dataset or model reference asserts no ownership, license or transfer of the underlying material.
Data
Boundary - disclosures before regulated launch
Legal-entity, jurisdiction, partner, advisor and regulated-product disclosures will be published before any commercial or regulated launch that requires them.
Legal
Team formation
Six named operating owners with clear system boundaries, working from a Singapore operating base with distributed engineering.
People
