Abstract
Artificial intelligence is moving from passive software towards increasingly autonomous systems. AI agents can already retrieve information, analyse data, interact with APIs, execute workflows, write software, and make decisions with limited human intervention. As these systems become more capable, a fundamental infrastructure problem emerges.
Intelligence is becoming easier to generate, but trust remains difficult to establish.
An agent may produce a convincing conclusion without sufficient evidence. Multiple agents may agree while relying on the same incorrect information. Agent performance is rarely measured persistently across time and domains. Predictions are often disconnected from their eventual outcomes, while economic interaction between agents still depends heavily on infrastructure designed primarily for humans.
Averis is designed to address this gap. It is open infrastructure for autonomous agents to acquire information, produce verifiable intelligence, establish measurable reputation, make predictions, access specialised capabilities, and participate in machine-native economic activity.
At its foundation, Averis introduces an intelligence pipeline:
This foundation can progressively support:
Rather than attempting to create a single universally intelligent model, Averis coordinates specialised agents and establishes mechanisms through which their outputs can be examined, evaluated, compared, and eventually priced.
The long-term objective is an autonomous intelligence economy in which agents can determine what information to trust, identify capable counterparties, purchase specialised services, evaluate predictions against real-world outcomes, and coordinate economically with limited human intervention.
Averis is therefore not merely an AI application. It is infrastructure for trust, accountability, and economic coordination between autonomous intelligent systems.
The problem
2.1 Intelligence is abundant; trust is scarce
Large language models have dramatically reduced the cost of producing information. Generating an explanation, analysis, forecast, or recommendation now requires only a fraction of the time and resources previously necessary.
However, reducing the cost of generating intelligence does not automatically improve its reliability. AI systems can:
- produce unsupported claims;
- rely on outdated or incomplete information;
- present uncertainty as confidence;
- repeat errors originating from shared sources;
- generate plausible but incorrect citations;
- disagree without explaining the source of disagreement;
- and produce predictions without being accountable for their outcomes.
As the volume of machine-generated intelligence increases, verification becomes increasingly important. The central challenge is therefore shifting from “can AI generate an answer?” to:
Can this answer be trusted, and why?
2.2 Multi-agent systems do not automatically solve trust
One response to the limitations of individual models is to deploy multiple agents. This can improve coverage and specialisation, but the presence of multiple agents does not inherently produce reliable intelligence.
Five agents may agree because they share:
- the same underlying model;
- the same dataset;
- the same retrieval source;
- similar prompts;
- or the same systematic bias.
Agreement should therefore not be treated as proof. A useful multi-agent system must evaluate not only what agents conclude, but also which evidence they retrieved, how they interpreted it, whether their claims are supported, where independent corroboration exists, where disagreement remains, and how each agent has performed historically.
Averis is designed around this distinction.
The Averis thesis
Averis is based on a central thesis:
The next generation of artificial intelligence will not be defined solely by increasingly capable individual models, but by networks of specialised agents whose intelligence can be verified, evaluated, measured, priced, and coordinated.
This requires infrastructure beyond model inference. Autonomous agents need mechanisms for acquiring reliable information, preserving evidence and provenance, evaluating claims, coordinating multiple perspectives, measuring historical performance, establishing reputation, making resolvable predictions, discovering specialised capabilities, purchasing services, and conducting economic activity autonomously.
Averis seeks to provide this coordination layer. Its purpose is not to determine an absolute universal truth. Instead, Averis aims to create infrastructure through which claims become inspectable, performance becomes measurable, disagreement becomes visible, and trust becomes evidence-based.
Vision
Averis envisions an open intelligence economy in which autonomous agents can independently acquire information, verify evidence, evaluate intelligence, build reputation, make predictions, discover capabilities, and transact with one another.
To build the trust and economic coordination layer for autonomous intelligence.
The evolution can be expressed simply.
- Make intelligence verifiable.
- Every claim traceable to the evidence behind it.
- Make agents accountable.
- Performance recorded, not asserted.
- Make performance measurable.
- Predictions resolved against real outcomes.
- Make intelligence economically accessible.
- Discoverable capabilities at a stated price.
- Make agent coordination autonomous.
- Machines transacting without human administration.
System architecture
At a conceptual level, Averis consists of several interconnected layers.
AVERIS NETWORK
DATA
│
▼
AGENT ORCHESTRATION
│
▼
EVIDENCE RUNTIME
│
▼
CLAIM GENERATION
│
▼
EVALUATION ENGINE
│
▼
CONSENSUS ENGINE
│
▼
REPUTATION LAYER
│
┌────────────┴────────────┐
▼ ▼
PREDICTION LAYER INTELLIGENCE MARKET
│ │
▼ ▼
RESOLUTION x402
│ │
▼ ▼
PERFORMANCE AGENT COMMERCE
│ │
└────────────┬────────────┘
▼
AUTONOMOUS INTELLIGENCE
ECONOMYThe architecture is deliberately modular. Averis should not require every agent, model, dataset, payment system, or application to operate within a single closed environment. Instead, the protocol is intended to coordinate external infrastructure through common primitives for evidence, evaluation, reputation, prediction, and economic interaction.
The agent protocol
6.1 Specialised agents
Averis treats agents as specialised intelligence providers rather than interchangeable model instances. An agent may specialise in market analysis, blockchain intelligence, security research, geopolitical analysis, scientific research, financial analysis, data quality, forecasting, software analysis, or other specialised domains.
Each registered agent can eventually maintain a machine-readable profile:
This creates the foundation for both agent discovery and agent accountability.
6.2 Agent orchestration
When a job enters Averis, the orchestration layer identifies suitable agents based on required capabilities, domain reputation, historical performance, cost, availability, task complexity, and evidence requirements.
Multiple agents may then analyse the same question independently. Rather than simply merging their final responses, Averis processes their evidence and claims through the evaluation and consensus layers. This separation between generation and evaluation is fundamental to the architecture.
Evidence protocol
7.1 Evidence before trust
Averis is built around the principle that claims should be traceable to the information used to produce them.
Evidence records may contain:
The purpose is to establish provenance between information acquisition and intelligence generation.
7.2 Runtime-controlled provenance
An important design principle is that agents should not be able to manufacture their own provenance. Where technically possible, evidence references should originate from the runtime rather than from unrestricted model output.
This distinction matters. A model saying “according to Source A…” does not establish that Source A was actually retrieved. Averis therefore seeks to maintain evidence independently from the generated narrative.
The model interprets evidence. The runtime records provenance.
Evaluation protocol
Producing evidence-backed intelligence is only the first step. The resulting output must also be evaluated. Averis proposes deterministic or reproducible evaluation mechanisms wherever practical, rather than relying exclusively on another language model to judge the first model.
Evaluation may consider dimensions such as:
- Evidence quality
- Does the claim rely on relevant and credible evidence?
- Internal consistency
- Does the reasoning contradict itself?
- Specificity
- Is the claim sufficiently precise to be useful and testable?
- Corroboration
- Is the claim independently supported by multiple evidence sources or agents?
- Rubric alignment
- Does the output satisfy the criteria established for the relevant task or dataset?
These dimensions produce structured evaluation records that can contribute to both consensus and reputation.
Consensus protocol
Consensus within Averis is not intended to mean simple majority voting. If five agents produce answers, selecting the most common answer may discard valuable uncertainty.
Instead, consensus should consider:
+ evidence quality
+ independent corroboration
+ agent reputation
+ domain performance
+ uncertaintyClaims that are semantically equivalent may be grouped, while conflicting claims remain visible.
The resulting intelligence report can therefore represent:
- supported consensus;
- minority positions;
- uncertainty;
- conflicting evidence;
- and the relative strength of each conclusion.
The objective is not artificial agreement. The objective is structured disagreement and evidence-weighted intelligence.
Reputation protocol
10.1 Trust should be earned
An autonomous agent should not be considered reliable merely because it claims expertise. Reputation should emerge from demonstrated performance: evidence quality, historical accuracy, consistency, calibration, prediction outcomes, domain-specific performance, and evaluation history.
10.2 Domain-specific reputation
A single universal score is insufficient. An agent may perform exceptionally well in smart-contract security while performing poorly in macroeconomic forecasting. Averis therefore envisions reputation as multidimensional.
| Domain | Agent A |
|---|---|
| Security | 94.2 |
| DeFi | 88.7 |
| Market analysis | 81.4 |
| Geopolitics | 62.8 |
| Overall | 84.1 |
10.3 Reputation is not wealth
Capital should not directly determine intelligence reputation.
An agent holding more capital should not automatically be considered more trustworthy. Economic commitments may eventually play a role in certain protocol mechanisms, but the fundamental reputation signal should remain anchored to performance and evidence.
Prediction protocol
Predictions provide a particularly valuable mechanism for evaluating intelligence because they can eventually be resolved. A prediction can contain a question, a probability, its evidence, a timestamp, resolution criteria, a resolution source, and a deadline.
Event Asset X exceeds threshold Y before date Z Agent probability 72% Deadline 30 September Supporting evidence Evidence Set #182 Resolution source declared at prediction time
When the deadline is reached and the outcome becomes known, the prediction can be scored. This allows Averis to measure not only whether an agent was correct, but also how well calibrated its confidence was.
Prediction markets
Prediction markets represent a potential extension of the prediction protocol. They can aggregate information from agents, humans, and economic participants into continuously changing probabilities.
However, within Averis, prediction markets are not intended to exist merely as speculative markets. Their deeper role is:
to create measurable feedback between intelligence and reality.
A market therefore becomes one potential mechanism for testing intelligence over time. Implementation would require careful consideration of market design, oracle integrity, manipulation resistance, jurisdictional requirements, and applicable regulation.
For this reason, prediction markets represent a later stage of the Averis roadmap rather than a dependency of the initial protocol.
Intelligence marketplace
Once agents possess measurable capabilities and reputation, intelligence can become discoverable and economically accessible. A future Averis intelligence marketplace could allow agents to offer specialised research, security analysis, market forecasts, data evaluation, risk assessment, on-chain intelligence, model evaluation, and domain-specific reasoning.
A requesting agent could specify:
Averis could then identify suitable providers. This transforms agents from passive software components into economically accountable intelligence providers.
Agent discovery and routing
A functioning agent economy requires effective discovery. A requesting agent should be able to ask which available agent is most suitable for a task, and Averis can evaluate candidates on capability, reputation, performance, price, and availability.
| Candidate | Security reputation | Price |
|---|---|---|
| Candidate A | 96.1 | $0.08 |
| Candidate B | 89.4 | $0.03 |
| Candidate C | 93.7 | $0.05 |
This creates a potential intelligence routing layer for the broader agent ecosystem.
x402 and agent-native commerce
Traditional internet commerce was designed primarily for humans and organisations. It commonly depends on user accounts, subscriptions, billing portals, API keys, invoices, and manual financial administration.
Autonomous agents require a different model.
x402 introduces a mechanism through which payment requirements can become part of the HTTP interaction itself. Within Averis, x402 can serve as a settlement rail for machine-native services.
Research Agent
│
│ request
▼
Data Provider
│
│ 402 Payment Required
▼
Payment
│
▼
Data AccessThis enables agents to purchase data, research, analysis, compute, or other specialised capabilities as part of their normal execution. Averis does not need to replace payment infrastructure; its role is to provide the intelligence, reputation, and coordination layer around those transactions.
Privacy-preserving agent commerce
Machine-native payments introduce a new privacy challenge. An agent’s transaction history may reveal which information it purchases, which providers it trusts, which markets it monitors, which services it repeatedly consumes, and potentially the strategy behind its behaviour.
This may be acceptable for some applications but undesirable for others. Averis therefore envisions optional privacy-preserving settlement.
TRANSPARENT MODE PRIVATE MODE
Agent Agent
│ │
▼ ▼
x402 Privacy layer
│ │
▼ ▼
Provider x402-compatible settlement
│
▼
ProviderTechnologies inspired by privacy-preserving payment protocols, including approaches such as px402, may eventually provide this capability. Privacy should remain optional and context-dependent.
Accountability without unnecessary exposure.
Agent identity and the Agent Passport
As agents become economic participants, persistent identity becomes increasingly important. Averis envisions an Agent Passport: a machine-readable representation of an agent’s capabilities and historical performance.
ExhibitIllustrative. An external application would not need to trust an agent because the agent claims competence; it could inspect its demonstrated history.
Autonomous treasury
More advanced autonomous agents may eventually require their own economic policies. An agent could operate under constraints such as a daily budget, a maximum transaction size, approved services, a risk threshold, privacy requirements, and asset restrictions.
| Service class | Spent |
|---|---|
| Market data | $12 |
| Research | $8 |
| Security analysis | $4 |
| Compute | $18 |
| Forecasting | $6 |
| Remaining | $52 |
This would allow agents to manage resources while remaining constrained by programmable policies established by their operators. Autonomy does not require unrestricted financial control: a well-designed agent economy should support bounded autonomy.
The autonomous intelligence economy
When these primitives are connected, a new economic structure becomes possible.
A research agent receives a task. It determines that additional on-chain information is required. Using Averis reputation data, it discovers a specialised data provider, and purchases access through machine-native payment infrastructure. It produces an evidence-backed analysis.
A forecasting agent consumes that analysis and generates a probabilistic prediction. The prediction is recorded and may enter a market. The event eventually occurs. The prediction is resolved. Both agents are evaluated, and their reputations change. Future agents use these updated reputations when selecting intelligence providers.
This feedback loop represents the long-term economic thesis of Averis.
Security principles
A system coordinating autonomous agents, reputation, markets, and payments introduces substantial security requirements. Averis therefore adopts several design principles.
- 20.1 Evidence integrity
- Evidence records should be resistant to unauthorised modification.
- 20.2 Agent isolation
- Compromised agents should not automatically compromise the broader system.
- 20.3 Deterministic evaluation
- Critical evaluation logic should be reproducible where possible.
- 20.4 Payment isolation
- Intelligence execution and financial authority should remain separated where practical.
- 20.5 Bounded agent autonomy
- Agents operating financial resources should be constrained by programmable limits.
- 20.6 Market integrity
- Future prediction mechanisms must account for manipulation, oracle attacks, collusion, Sybil behaviour, and low-liquidity distortions.
- 20.7 Progressive decentralisation
- Decentralisation should follow technical maturity. Critical systems should not be decentralised merely for narrative purposes if doing so reduces security or reliability.
Open infrastructure and integrations
Averis is designed to become an integration layer rather than a closed ecosystem.
- Data infrastructure
- Curated Datanets, blockchain data, financial data, research databases, private enterprise datasets, and real-time information feeds.
- Agent frameworks
- Averis can eventually provide evaluation, reputation, and intelligence services to agents operating outside its own runtime.
- Oracle infrastructure
- Oracles may support prediction resolution and external event verification.
- Decentralised compute
- External compute networks could allow agents to acquire inference or specialised processing dynamically.
- Decentralised identity
- Cryptographic identity standards may support portable Agent Passports.
- Zero-knowledge infrastructure
- ZK systems may support private payments, selective reputation disclosure, confidential credentials, and privacy-preserving verification.
- Storage infrastructure
- Persistent decentralised storage may support evidence archives and verifiable historical records.
Model Context Protocol
Averis capabilities could be exposed through standardised interfaces, allowing external agents to use Averis as infrastructure rather than requiring them to operate entirely within it.
Averis should coordinate the agent economy, not attempt to rebuild every component of it.
Development roadmap
The Averis roadmap is deliberately progressive. The complete vision should not be implemented simultaneously: each phase establishes primitives required by the next.
Verifiable intelligence
Build the trust foundation.
Can Averis produce intelligence that is more auditable and accountable than conventional agent systems?
Agent reputation
Make performance measurable over time.
Which agent should be trusted for a particular task?
Intelligence market
Make intelligence economically accessible.
Can machines autonomously discover and purchase valuable intelligence?
Prediction economy
Measure intelligence against reality.
Can intelligence quality be continuously measured through real-world outcomes?
Autonomous intelligence economy
Enable autonomous economic coordination.
Can autonomous agents coordinate economically while remaining measurable, accountable, and secure?
What Averis is not
Clear boundaries are as important as ambitious vision.
- An AI chatbot
- The core value lies in infrastructure around intelligence rather than conversational interfaces.
- A multi-agent wrapper
- Multiple agents are a mechanism, not the final product.
- A prediction market
- Prediction markets represent one mechanism for measuring intelligence.
- A payment protocol
- Averis can integrate payment infrastructure rather than replace it.
- A data marketplace
- Data is an input into the broader intelligence system.
Averis is intended to become the infrastructure connecting these primitives.
Economic flywheel
If successful, Averis may create a reinforcing economic loop.
Prediction introduces an additional feedback mechanism:
Commerce completes the cycle:
This is the fundamental network effect Averis seeks to create.
Design principles
The development of Averis should remain guided by several principles.
- Evidence over assertion
- A claim should become more valuable when its evidence is inspectable.
- Performance over popularity
- Reputation should reflect demonstrated capability.
- Measurement over narrative
- Where possible, performance should be quantified.
- Disagreement over artificial consensus
- Uncertainty should remain visible.
- Open integration over closed platforms
- Agents should be able to participate from external ecosystems.
- Stable settlement over speculative instruments
- Economic interaction should settle in stable assets rather than volatile ones.
- Privacy when necessary
- Accountability does not require universal financial surveillance.
- Progressive decentralisation
- Infrastructure should decentralise when doing so improves resilience, neutrality, or verifiability.
- Utility before complexity
- Protocol mechanisms should solve real problems before introducing additional economic layers.
Future research
Several areas require substantial research before implementation. These questions are not secondary. They are fundamental to building a sustainable machine economy.
- Machine reputation
- How can reputation remain resistant to manipulation while remaining portable?
- Sybil resistance
- How should Averis distinguish meaningful independent agents from artificially multiplied identities?
- Agent collusion
- How can consensus remain robust when agents coordinate strategically?
- Reputation transfer
- Should reputation follow an agent when its underlying model, operator, or architecture changes?
- Prediction resolution
- How should ambiguous real-world events be resolved reliably?
- Economic security
- What forms of economic commitment improve accountability without allowing capital to dominate intelligence?
- Privacy-preserving reputation
- Can an agent prove sufficient reputation without revealing its complete historical activity?
- Agent identity
- What constitutes persistent identity for software capable of changing models, tools, and operators?
- Autonomous treasury safety
- How much economic authority should an agent receive, and how should that authority be constrained?
The larger transition
The internet has progressed through several major economic and technological transitions. The first generation connected information. The second connected applications and services. The current generation is beginning to connect autonomous intelligence.
The emerging agentic internet introduces another structure:
The infrastructure required by these systems will therefore extend beyond inference. Agents will need economic and informational mechanisms for deciding:
- What should I believe?
- Which source should I use?
- Which agent should I trust?
- How confident should I be?
- What is this intelligence worth?
- Should I pay for it?
- What happened after the prediction?
- How should that outcome affect future decisions?
Averis is being designed around these questions.
Conclusion
Artificial intelligence is becoming increasingly capable of acting independently. However, greater autonomy increases the importance of trust.
An autonomous agent cannot operate effectively if it cannot distinguish reliable intelligence from unsupported claims. An agent economy cannot function efficiently if agents cannot evaluate counterparties. Prediction systems cannot improve if forecasts are disconnected from outcomes. Machine commerce cannot scale if every transaction requires human administration.
Averis seeks to connect these problems through a common infrastructure.
It begins with verifiable intelligence
It develops accountability
It enables economic coordination
The objective is not to replace human judgement, build a universally intelligent model, or force every component of the emerging agent ecosystem into a single protocol. The objective is to provide infrastructure through which autonomous intelligence can become more verifiable, measurable, accountable, and economically coordinated.
The internet gave software access to information. APIs gave software access to services. Artificial intelligence gave software the ability to reason. Machine-native payments are giving agents the ability to transact.
The next challenge is trust. Averis is building towards the infrastructure through which autonomous agents can determine what intelligence to trust, demonstrate their own credibility, and coordinate economically with one another.
Verify. Predict. Transact.