Tesserax: a post-quantum proof-of-work chain for autonomous agents

Version 1.0 · October 2026 · Tesserax developers

Abstract

Two shifts are coming to public blockchains at once. The first is cryptographic: a large enough quantum computer breaks the elliptic-curve signatures that protect nearly every account in existence, and every public key ever revealed on chain stays exposed until then. The second is economic: software agents built on language models are starting to buy, sell and pay for work on their own, and they need money that a program can hold under limits a human can trust.

Tesserax is a layer-1 blockchain designed for both. Every account is an ML-DSA-87 key (FIPS 204) and every transaction an ML-DSA signature; there is no elliptic-curve account, no classical fallback and no migration to plan. Nodes and the public RPC exchange keys with ML-KEM-768 (FIPS 203). Private transfers are proven with hash-based zero-knowledge STARKs and their notes are encrypted with ML-KEM, so the privacy layer carries no elliptic-curve or pairing assumption either. Smart contracts run on PolkaVM, with the elliptic-curve precompiles removed and a native precompile that verifies ML-DSA signatures.

Blocks are produced by proof of work every 18 seconds on average, with RandomTESS, a configuration of RandomX v2 that keeps mining on general-purpose CPUs. The supply is capped at 14,000,000 TESS: 85% is only ever created as block rewards, halving every 7,000,000 blocks (about four years), and the rest is held by the team and a foundation under vesting schedules enforced by the chain.

On top of this base, the runtime gives AI agents native economic primitives: delegations that let an agent key spend an owner's funds within a budget, a per-call cap, allowed recipients and an expiry; a registry of agents whose reputation can only be earned through payments; payment channels whose vouchers pay HTTP services per request with status 402; and a task market with escrow and arbitration. An MCP server exposes all of it to any agent that speaks the Model Context Protocol. Atomic swaps trade TESS against BTC, ETH, USDC and USDT without bridges or custodians, and the protocol evolves by on-chain referenda, without any privileged key.

This paper describes the design, its parameters and its security, as implemented in the open-source reference implementation running on the public testnet.

Contents

1. Introduction: the quantum threat, the agent economy, design goals.
2. Architecture: the node, the runtime and the clients.
3. Cryptography: ML-DSA accounts, keys, post-quantum transport, hashing.
4. Consensus: RandomTESS, the seal, difficulty, fork choice and finality, mining.
5. Economics: fees, throughput, emission, allocations.
6. The agent economy: delegation, registry, channels and HTTP 402, tasks, MCP.
7. Privacy: the shielded pool and its STARK.
8. Smart contracts: PolkaVM, precompiles, addresses.
9. Atomic swaps: offers, hash time-locked contracts on Bitcoin and EVM chains.
10. Governance: referenda, conviction voting, the technical committee, upgrades.
11. Security analysis: threat model, attacks and limits.
12. Roadmap: from testnet to mainnet.
13. References.

Notation

Symbol	Meaning
TESS	The native token. 1 TESS = $10^{12}$ planck.
$T$	Target block time, 18 s.
$D$	Difficulty: the expected number of hashes to find a block.
$\mathcal{H}(x)$	BLAKE2b-256 unless stated otherwise.
$x \,\Vert\, y$	Concatenation of byte strings.
$H(a, b, d)$	The Poseidon2 hash of the shielded pool (chapter 7).

Parameters are given as configured in the runtime (spec version 105). Time expressed in hours or days is converted to blocks at the 18-second target: 200 blocks an hour, 4,800 a day. Actual durations vary with the time between blocks.

note

This document describes software. It is not an offer of any financial product, and the testnet token has no value. Parameters may change by referendum; the code is the authority.