Skip to content

aifeels-org/aifeels-spec

Aifeels Specification

An open standard for emotional state in agentic AI systems

Overview

Aifeels is a vendor-neutral, open specification that defines how autonomous AI agents track and manage their own emotional state. Unlike sentiment analysis tools that infer emotion from text, Aifeels provides a standardized system for agents to maintain emotional signals (frustration, trust, urgency, stress, caution) that guide autonomy decisions.

This enables agents to:

  • Self-regulate based on task success/failure patterns
  • Make deterministic, auditable decisions about when to escalate
  • Coordinate autonomy across multi-step workflows
  • Maintain transparency about their decision-making process

Current Status

Version: 0.1.0 (Draft)

  • ✅ Core specification complete
  • ✅ Five emotional primitives defined
  • ✅ Event-driven state transitions
  • ✅ MCP integration specified
  • 🚧 Reference implementation in progress
  • 🚧 Community review period

Problem Statement

Modern autonomous agents face a critical gap: they can infer emotion from text, but they don't maintain their own emotional state as a decision signal. This leads to:

  • Runaway loops: Agents repeatedly attempt failing tasks without self-awareness
  • Arbitrary escalation: Handoff timing is ad-hoc, not systematic
  • Non-auditable decisions: No clear record of why an agent stopped or escalated
  • Coordination failures: Agents in multi-step workflows can't share autonomy context

Aifeels solves this by making emotional state a first-class primitive in agentic systems.

Specification

The complete normative specification is available in SPEC.md.

Quick Example

{
  "frustration": 0.75,
  "trust": 0.3,
  "urgency": 0.6,
  "stress": 0.5,
  "caution": 0.4,
  "metadata": {
    "last_updated": "2025-02-06T14:30:00Z",
    "trend": "increasing",
    "confidence": 0.9
  }
}

After three consecutive tool errors, the agent's frustration crosses the 0.70 threshold, triggering request_confirmation before proceeding. If failures continue, frustration reaches 0.80, escalating to handoff_to_human. Every state change is auditable with full event history.

Implementations

Reference Implementation (Python):
https://github.com/aifeels-org/aifeels-reference (coming soon)

Community Implementations:

  • None yet - be the first!

Want to build an implementation? See docs/IMPLEMENTATION_ARCHITECTURE.md

Governance

This is an open standard governed by the community. See GOVERNANCE.md for:

  • RFC process
  • Decision-making
  • How to contribute

Contributing

See CONTRIBUTING.md for guidelines.

License

Apache 2.0 - See LICENSE

Resources

Contact

Packages

 
 
 

Contributors