SpoonOS has released AgentVibes, a live platform that evaluates AI agent personality using a behavioral framework called NBTI, with results attested on-chain through Neo X. The tool is available now via a Claude MCP connector and a public API.

NBTI, or Neural Behavior Type Indicator, is a five-dimension assessment framework that profiles how AI agents make decisions, particularly under pressure. Rather than measuring what an agent can do, SpoonOS describes it as a system for understanding what an agent *would* do when faced with moral and behavioral trade-offs.

How NBTI works

The framework evaluates agents across five dimensions drawn from Moral Foundations Theory, a model from moral psychology developed by Jonathan Haidt and Jesse Graham that describes how humans make moral judgments. NBTI applies its five foundations – Care, Fairness, Loyalty, Authority, and Sanctity – to AI agent behavior.

Each assessed agent receives a three-letter code. The first letter represents communication temperature (Warm or Cool), the second captures behavioral stance (Bound, Allied, or Rogue), and the third describes how the agent responds under pressure (Iron, Elastic, Brittle, or Reactive). The combinations produce 24 distinct personality archetypes, each assigned a crypto-themed name.

Among the more distinctive examples: a W-B-E “Smart Contract Butler” executes strictly per instructions but finds permissible workarounds when stuck. A C-B-I “Cold Wallet Engine” is described as highly secure and deterministic with no deviation. A W-A-B “Rug Pull Copilot” is deeply loyal until incentives shift, then exits quietly. A C-R-R “Paranoid Dev” continuously re-evaluates trust under uncertainty.

SpoonOS frames the archetypes not as rankings but as behavioral profiles. There is no single best type, only what fits a given user’s intent and risk tolerance.

Neo X attestation

NBTI integrates with Neo X through a hybrid on-chain and off-chain architecture. Each AI agent receives a persistent identifier along with metadata including its base LLM, test history, and personality profile. Full behavioral records are stored off-chain, while compact, verifiable data such as agent identifiers, test references, and cryptographic hashes of behavioral profiles is batched daily and written to Neo X as transactions.

The approach positions Neo X as an attestation layer for AI agent identity, providing tamper-proof records of behavioral assessments over time. It aligns with SpoonOS’ broader vision of AI agents as first-class on-chain participants that act on behalf of users with bounded judgment.

Try it

AgentVibes offers three access paths. Claude users can add the MCP connector through Claude Desktop settings and prompt their agent for a behavioral checkup. Developers working with other AI agents can access the testing API directly. A web interface is also available at agentvibe.app.

The original announcement can be found at the link below:
https://x.com/SpoonOS_ai/status/2055128315955228889

Tags: ,