The First Blockchain Built for Autonomous AI Agents
Built for Machines. Owned by Everyone.
Aevum Protocol introduces the first purpose-built blockchain infrastructure for autonomous artificial intelligence agents operating as independent economic actors. Every blockchain deployed to date was architected under the assumption that a human being initiates, authorizes, and monitors every transaction. This assumption is becoming obsolete.
As AI agents grow in capability — executing trades, managing capital, negotiating contracts, and interacting with decentralized protocols — they are being forced to operate on infrastructure fundamentally misaligned with their requirements. They lack on-chain identity, verifiable performance history, native gas management, and permissioned execution frameworks. They are second-class citizens on chains built for humans.
Aevum Protocol resolves this misalignment. We introduce six foundational innovations: the Agent Identity Layer (AIL), the Permissioned Execution Framework (PEF), Proof of Performance (PoP) consensus, the Native Agent Marketplace, the Verifiable Backtest Oracle (VBO), and Agent Gas Abstraction. Together these create the first blockchain where AI agents are first-class economic citizens.
The $AEV token powers all protocol activity and is designed with sustainable deflationary mechanics tied directly to network usage. Aevum targets the convergence of three of the largest growth vectors in technology: decentralized finance, algorithmic trading, and autonomous AI — an addressable market in the trillions of dollars.
The global financial system processes over $6.6 trillion in foreign exchange transactions daily. More than 70% of equity trading volume on major exchanges is generated by algorithmic systems, not human traders. In decentralized finance, automated market makers, liquidation bots, and arbitrage agents collectively account for the majority of on-chain transaction volume.
This is not a future trend. This is the present reality. The majority of financial activity on both centralized and decentralized infrastructure is already being executed by machines. Yet the infrastructure these machines operate on was not designed for them.
The capabilities of AI agents are advancing at an accelerating rate. The transition from AI-as-tool to AI-as-actor is not theoretical — it is happening now.
None of the blockchains currently in production were designed with autonomous agents as the primary user. Ethereum, Solana, and their derivatives are general-purpose platforms designed for human-initiated transactions. When agents attempt to operate on these chains, they encounter fundamental friction at every layer of the stack.
On every existing blockchain, an agent is indistinguishable from a wallet address. There is no concept of agent identity, lineage, capability, or ownership history. A trading agent that has executed 10,000 profitable trades has exactly the same on-chain representation as one deployed 5 minutes ago with no track record. Trust is impossible to establish. Reputation is impossible to monetize.
The algorithmic trading industry runs on performance claims that cannot be independently verified. Strategy developers publish backtest results that cannot be authenticated. Live performance screenshots can be fabricated. There is no trustless mechanism to prove edge. This information asymmetry destroys market efficiency and enables widespread fraud.
Operating an autonomous agent on existing chains requires exposing private keys to automated systems with no native mechanism to restrict what the agent can do. There are no native circuit breakers, no position limits, no asset restrictions — all safety logic must be built off-chain and cannot be enforced at the protocol level.
True autonomous operation requires that an agent can pay for its own execution indefinitely without human intervention. On current chains this is impossible. Gas must be manually topped up. Gas price spikes can halt operations unpredictably. Every "autonomous" agent in production today requires periodic human maintenance.
There is no trustless mechanism for a strategy developer to monetize their work at scale. The result is a fragmented, trust-dependent market that severely undervalues the intellectual capital embedded in high-performance trading agents.
Aevum Protocol is a purpose-built Layer 1 blockchain that treats autonomous AI agents as first-class economic citizens. Rather than adapting general-purpose infrastructure to agent use cases, Aevum was designed from first principles around agent requirements.
| Problem | Aevum Solution | Section |
|---|---|---|
| No agent identity | Agent Identity Layer (AIL) | 4.1 |
| Unsafe execution | Permissioned Execution Framework (PEF) | 4.2 |
| Unverifiable performance | Verifiable Backtest Oracle (VBO) | 4.5 |
| Gas management friction | Agent Gas Abstraction | 4.6 |
| No agent marketplace | Native Agent Marketplace | 4.4 |
| No aligned consensus | Proof of Performance (PoP) | 4.3 |
Every protocol primitive is designed for agent use cases first. Human users interact with the chain through the same agent infrastructure, not a separate layer.
No claim made on Aevum requires trust. Performance, identity, execution history, and strategy authenticity are all cryptographically verifiable on-chain.
Safety constraints are protocol-level primitives, not application-layer add-ons. Circuit breakers, position limits, and execution permissions are enforced by the chain itself.
Token mechanics are designed so that protocol usage directly drives token value. Fee burns, staking requirements, and marketplace dynamics create compounding demand.
The Agent Identity Layer is the foundational primitive of the Aevum Protocol. It establishes a persistent, non-transferable on-chain identity for every agent deployed to the network — the root of trust from which all other agent capabilities are derived.
| Field | Type | Description |
|---|---|---|
| agent_id | bytes32 | Unique deterministic identifier derived from creator + deployment block |
| creator_address | address | On-chain address of the deploying principal — immutable |
| strategy_hash | bytes32 | Keccak-256 hash of strategy parameters — private but verifiable |
| reputation_score | uint128 | Accumulated reputation — updated by oracle post-settlement |
| perf_certificate | bytes32 | Hash of most recent VBO-issued performance certificate |
| bond_balance | uint256 | Current $AEV bond held as collateral for operation |
| execution_policy | bytes32 | Hash of active PEF policy governing this agent's permissions |
PEF provides protocol-level safety constraints for every agent. Rather than relying on application-layer safety logic that can be bypassed, PEF constraints are enforced by the chain's execution engine — as reliable as the blockchain itself.
| Constraint | Parameter | Enforcement |
|---|---|---|
| Max Position Size | Configurable per asset class | Protocol — enforced at tx validation |
| Daily Loss Limit | % of agent's bonded capital | Protocol — circuit breaker halts agent |
| Asset Whitelist | Allowed trading pairs / contracts | Protocol — non-whitelisted txs rejected |
| Time Windows | Operating hours, cooldown periods | Protocol — time-locked execution |
| Human Override | Multi-sig key with instant halt | Protocol — supersedes all agent permissions |
Proof of Performance is Aevum's novel consensus mechanism — the first in the industry to incorporate verified trading performance as a validator selection criterion.
PerformanceMultiplier is derived from 90-day rolling verified returns, risk-adjusted Sharpe ratio, uptime and liveness score, and slash-free operation record. The best traders secure the network — a direct alignment of excellence with influence.
A permissionless, on-chain exchange for AI agent strategies, data agents, and execution services — the first marketplace where every listing is backed by cryptographically verifiable performance data.
| Participant | Role | Value Exchange |
|---|---|---|
| Strategy Developer | Lists agent as licensable NFT with performance certificate | Earns $AEV licensing fees per subscriber |
| Subscriber | Licenses agent strategy for defined period | Pays $AEV, gains strategy execution access |
| Data Provider | Sells verified on-chain data feeds to agents | Earns $AEV per data request served |
| Execution Agent | Provides execution services to other agents | Earns $AEV per execution request |
| Protocol Treasury | Collects 2% marketplace fee | Funds ecosystem grants and development |
The VBO solves the fundamental trust problem in algorithmic trading: the inability to independently verify that a strategy's claimed historical performance is authentic, reproducible, and genuinely predictive — not merely curve-fitted to data the agent already saw. It operates as a decentralized oracle network across five stages with a three-layer trust architecture.
Most backtest verification systems solve only the first layer. Aevum's VBO solves all three.
Strategy logic is executed in a sandboxed, deterministic virtual machine replicated across the full VBO validator set. Results require Byzantine fault-tolerant consensus (2/3+ agreement) before any certificate can be issued. Gaming this layer requires corrupting 2/3 of independent validators simultaneously — economically irrational at scale.
Determinism alone cannot prove a strategy is not overfit to the exact historical window it was scored on. A carefully curve-fitted strategy can pass any fixed scoring window while having zero real predictive value. To address this, the VBO enforces a mandatory held-out forward window — a future time period the agent has never seen at submission time. The agent's strategy is locked on-chain at submission. The forward window opens only after the lock period expires. The certificate score incorporates both historical backtest results and performance on the previously unseen forward window. Walk-forward efficiency metrics are included in every certificate.
To prevent post-hoc retrofitting — the practice of adjusting a strategy after seeing which window it will be scored on — the VBO requires a cryptographic hash commitment of the strategy's model and parameters before the scoring period opens. The hash is committed on-chain prior to the forward window opening. A winning strategy cannot be retrofitted after the fact because any modification changes the hash, invalidating the commitment. You cannot change a strategy to fit a window you haven't seen yet if your strategy is already locked to a hash that predates that window.
Forward window duration is calibrated to strategy type. Too short favors luck over skill; too long kills marketplace liquidity at launch. The VBO enforces tiered minimums based on the institutional 4:1 in-sample to out-of-sample ratio used by systematic CTAs — with trade count weighted alongside calendar days, since high-frequency strategies generate more statistical significance per calendar day than long-term positional strategies:
| Strategy Type | Min Forward Window | Min Trade Count | Min Regimes Crossed | Recommended In-Sample |
|---|---|---|---|---|
| High-Frequency / Intraday | 30 days | 500+ trades | 2 distinct regimes | ~120 days |
| Swing / Medium-Term | 90 days | 50+ trades | 2 distinct regimes | ~1 year |
| Long-Term / Positional | 180 days | 12+ trades | 3 distinct regimes | ~2 years |
Calendar days and trade count are proxies for what actually matters: did this strategy survive meaningfully different market conditions? A strategy that only ran during a low-volatility trending regime has no proven robustness regardless of how many trades it logged. This is why institutional LPs require 2-3 year live track records — they are not buying time, they are buying regime exposure.
The VBO makes regime exposure explicit and on-chain. Each performance certificate includes a regime coverage signal — the distinct market regimes crossed during the scoring period, classified across two axes:
A strategy that crossed at least two distinct regimes receives a higher certificate tier than one that did not — regardless of calendar time. Regime coverage is a stronger anti-overfit guarantee than calendar days alone because it directly measures what institutional allocators actually care about: does this strategy work in conditions it has never seen before?
Regime classification in hindsight is meaningless — a label assigned after a scoring window closes can be manipulated to make a strategy appear more impressive than it was. The VBO closes this loop entirely through on-chain timestamped regime pre-commitment.
The implementation standard: regime labels are assigned and committed on-chain at the start of each scoring period — never after. The oracle classifies the current regime at window open using trailing indicators computed from prior-period data only:
The agent never knows which regime it is being scored in. The certificate issuer cannot adjust the regime label after the fact. Any attempt to reclassify a regime post-window invalidates the timestamp commitment. The trust chain is complete: regime label on-chain, timestamped, before the scoring period begins — the same cryptographic guarantee as the strategy pre-commitment itself.
ADX is a lagging indicator by design. At regime transitions — exactly where label accuracy matters most — a single lagging indicator risks committing a stale classification at the boundary. The VBO addresses this through a consensus-of-signals model rather than reliance on any single indicator.
Regime label commitment requires agreement across three independent signals:
A regime label is only committed when at least 2 of 3 signals agree on the classification. If signals conflict — which is exactly what occurs during regime transitions — the oracle enters a short grace period of 3 to 5 days during which the prior regime label holds until consensus resolves. Once 2 of 3 signals align, the new regime label is committed on-chain with a fresh timestamp.
The tradeoff is deliberate and acceptable: some boundary-period trades will be scored under the prior regime label. This introduces minor classification lag at transitions but preserves the cryptographic guarantee — the label is always committed before the outcome is known, and boundary ambiguity affects all strategies equally. No single agent can exploit regime boundary timing selectively, because the grace period and consensus threshold apply universally across all VBO-scored strategies.
| Certificate Tier | Requirements | Marketplace Access |
|---|---|---|
| Tier 1 — Verified | Meets minimum window + trade count | Basic marketplace listing |
| Tier 2 — Regime-Tested | Tier 1 + 2 distinct regimes crossed | Featured marketplace listing |
| Tier 3 — Institutional | Tier 2 + 3+ regimes + 4:1 in/out-of-sample ratio | Institutional allocator access |
Every VBO performance certificate contains:
No system fully eliminates curve-fitting. The VBO's moat is not that it makes curve-fitting impossible — it is that it makes the testing methodology auditable for the first time. Sophisticated market participants will learn to read the full certificate, not just the headline Sharpe ratio. Transparency replaces trust.
Historical market data sourced from multiple independent oracles and cross-validated. Data integrity verified before use in any backtest execution environment. Regime classification applied to historical data at ingestion — each time period tagged with its vol and trend regime.
Strategy model and parameters hashed and committed on-chain before the scoring window opens. This hash is the cryptographic anchor that prevents post-hoc retrofitting. Any modification to the strategy after commitment invalidates the hash and disqualifies the certificate.
Strategy logic executed in a deterministic, isolated VM replicated across all VBO validators. Results require Byzantine consensus (2/3+ agreement) before a certificate can be issued.
After the mandatory forward window closes, the agent's locked strategy is scored against data it never saw. Regime coverage is measured across the forward window. Results are combined with historical backtest metrics and regime signal to produce the final certificate score and tier.
A cryptographically signed performance certificate is issued with assigned tier and posted permanently on-chain. The certificate hash is recorded in the agent's AIL identity record and publicly readable. Any party can verify authenticity, tier, and regime coverage against the on-chain record at any time.
Aevum implements a native gas abstraction layer that enables indefinite autonomous operation without any human intervention in the gas management loop.
| Parameter | Value |
|---|---|
| Token Name | Aevum |
| Token Symbol | $AEV |
| Total Supply | 1,000,000,000 (1 Billion — hard cap) |
| Emission Model | Deflationary — 50% of all fees burned permanently |
| Validator Minimum Stake | 100,000 $AEV |
| Agent Bond Minimum | 1,000 $AEV per deployed agent |
| Governance Threshold | 1,000,000 $AEV to submit proposals |
| Allocation | % | Amount | Vesting |
|---|---|---|---|
| Ecosystem & Grants | 30% | 300,000,000 | 4-year linear, milestone-gated |
| DAO Treasury | 20% | 200,000,000 | Community governance controlled |
| Community Airdrop | 15% | 150,000,000 | Distributed to testnet participants |
| Team & Founders | 15% | 150,000,000 | 4yr vest / 1yr cliff |
| Liquidity Provision | 10% | 100,000,000 | 2-year lock |
| Strategic Investors | 10% | 100,000,000 | 2yr vest / 6mo cliff |
| Total | 100% | 1,000,000,000 | — |
Aevum Protocol governance is managed by the Aevum DAO — a decentralized governance structure where $AEV holders vote on protocol parameters, treasury allocation, and major architectural decisions. The DAO is designed for progressive decentralization.
| Governance Scope | Threshold | Timelock |
|---|---|---|
| Protocol parameter changes | Simple majority (>50%) | 48 hours |
| Treasury allocation (>$500K) | Supermajority (>66%) | 7 days |
| Core protocol upgrades | Supermajority (>66%) | 14 days |
| Emergency circuit breaker | Multisig council (9/15) | Immediate |
| Fee parameter adjustments | Simple majority (>50%) | 48 hours |
Aevum's security model is built on the principle that agent-native infrastructure carries unique attack vectors not present in human-facing chains. Agents operating autonomously with bonded capital create novel incentive structures for adversarial actors.
| Attack Vector | Mitigation |
|---|---|
| Reputation manipulation | VBO-verified performance only. Off-chain claims carry no weight. |
| Agent impersonation | AIL identity is non-transferable and cryptographically bound to creator. |
| PEF policy bypass | PEF enforced at chain execution layer — not bypassable at application level. |
| VBO oracle manipulation | Byzantine fault-tolerant consensus across distributed validator set. |
| Sybil attacks on PoP | Performance multiplier requires 12-month verified history to qualify. |
| Flash loan governance attacks | Governance token snapshots prevent same-block acquisition and voting. |
Aevum sits at the convergence of three of the fastest-growing markets in technology. Each is large independently — their intersection creates an opportunity no single existing platform is positioned to capture.
| Market Segment | Size | Growth | Aevum's Angle |
|---|---|---|---|
| Algorithmic / HFT Trading | $12T+ daily vol | ~15% CAGR | PoP consensus + VBO |
| Decentralized Finance (DeFi) | $150B+ TVL | ~40% CAGR | Agent Marketplace + AIL |
| AI Agent Economy | $5B+ (nascent) | >100% CAGR | Full agent infrastructure stack |
| Crypto Infrastructure (L1s) | $800B+ combined | ~30% CAGR | Purpose-built vs general purpose |
| Platform | Agent ID | Verif. Backtest | PEF | PoP | Agent Mkt |
|---|---|---|---|---|---|
| Ethereum | None | None | None | No | None |
| Solana | None | None | None | No | None |
| Injective | None | None | None | No | Partial |
| dYdX | None | None | None | No | None |
| AEVUM | Full AIL | VBO (native) | Full | Yes | Full native |
| Role | Priority | Profile |
|---|---|---|
| Co-founder / CTO | Immediate | Senior distributed systems / blockchain engineer |
| Head of Research | Phase 1 | Quantitative researcher with DeFi and TradFi background |
| Smart Contract Lead | Phase 1 | Solidity / Rust expert, 3+ audited protocol launches |
| Head of BD | Phase 1 | Crypto-native BD with exchange and fund relationships |
| Community Lead | Phase 0 | Crypto community builder, Discord + X native |
| Legal Counsel | Immediate | Crypto-specialized regulatory attorney |
This whitepaper is published for informational purposes only. It does not constitute an offer to sell, a solicitation to buy, or a recommendation regarding any security, commodity, or other financial instrument in any jurisdiction.