Glossary¶

Key terms and definitions for the Empathy Framework

A¶

Access Tier¶

A permission level that determines what an agent can do within the system. There are four tiers: - Observer (Level 1): Read-only access - Contributor (Level 2): Can read and write patterns - Validator (Level 3): Can promote patterns to permanent storage - Steward (Level 4): Full administrative access

See: Multi-Agent Philosophy

Agent¶

An AI instance that participates in the Empathy system. Agents can be specialized (security reviewer, performance analyst) or general-purpose. Multiple agents can coordinate through shared memory.

Anticipatory Empathy¶

The ability to predict and address needs before they're expressed. Level 4 in the Empathy Framework's five-level model. Anticipatory systems don't just respond to problems—they prevent them.

B¶

BATNA¶

Best Alternative To Negotiated Agreement. When two agents cannot find a synthesis that serves both interests, the system falls back to the BATNA—typically the recommendation with higher confidence. Borrowed from negotiation theory (Harvard Negotiation Project's "Getting to Yes").

C¶

Classification¶

A security label applied to patterns that determines storage, encryption, and retention policies: - PUBLIC: General patterns, no encryption, 365-day retention - INTERNAL: Proprietary patterns, optional encryption, 180-day retention - SENSITIVE: Healthcare/PII patterns, required encryption (AES-256), 90-day retention

Confidence¶

A numeric score (0.0 to 1.0) indicating how certain an agent is about a recommendation or pattern. Higher confidence typically means more evidence or successful past application.

Conflict Resolution¶

The process of finding a synthesis when two agents make conflicting recommendations. The framework extracts the underlying interests behind each position and generates options that serve both.

See: Practical Patterns - Conflict Synthesizer

Contributor¶

Access Tier Level 2. Can read patterns and propose new ones, but cannot validate or promote patterns to permanent storage. Most AI agents operate at this level.

D¶

Data Sovereignty¶

The principle that users and enterprises own, version, and control all memories associated with their projects. A foundational value of the Empathy Framework—not a feature, but a constraint that shapes all design decisions.

E¶

Emergence¶

Patterns that weren't explicitly taught but arise from collective agent operation. The framework treats emergent patterns as valuable and surfaces them for validation rather than filtering them out.

EmpathyOS¶

The main interface class for interacting with the Empathy Framework. Provides methods for memory operations, pattern management, and agent coordination.

from empathy_os import EmpathyOS
empathy = EmpathyOS(user_id="developer@company.com")

F¶

Fingerprint¶

A hash-based identifier used to detect duplicate patterns. Prevents the same pattern from being stored multiple times while allowing confidence boosting when the same pattern is discovered independently.

H¶

Heartbeat¶

A periodic signal sent by agents to indicate they're still functioning. Used by the monitoring system to detect unresponsive agents and reassign their work.

See: Practical Patterns - Heartbeat Monitor

I¶

Interests¶

The underlying needs or goals that motivate a recommendation, as opposed to the position (what is recommended). Conflict resolution works by identifying interests and finding solutions that serve multiple interests simultaneously.

Example: - Position: "Add input validation on all endpoints" - Interest: Prevent injection attacks, protect data integrity

L¶

Level (Empathy Level)¶

The Empathy Framework defines five levels of AI capability: - Level 1 - Reactive: Responds to explicit requests - Level 2 - Informed: Remembers context within a session - Level 3 - Contextual: Applies patterns from similar situations - Level 4 - Anticipatory: Predicts and prevents problems - Level 5 - Generative: Creates novel solutions from patterns

Level 4 (Anticipatory) is the minimum standard for Empathy systems.

Long-Term Memory¶

Persistent storage for validated patterns that survive across sessions. Patterns in long-term memory have been reviewed and promoted from staging. Contrast with Short-Term Memory.

M¶

Mock Mode¶

A development mode where Redis is simulated in-memory. Useful for quick experiments but doesn't support multi-agent coordination or persistence.

os.environ["EMPATHY_REDIS_MOCK"] = "true"

O¶

Observer¶

Access Tier Level 1. Read-only access to patterns and shared state. New agents typically start at Observer level until they demonstrate reliability.

P¶

Pattern¶

A reusable piece of knowledge—a best practice, code snippet, workflow, or insight—that can be applied across contexts. Patterns are the core unit of knowledge in the Empathy Framework.

Pattern Library¶

A collection of validated patterns available to all agents. Patterns enter the library through the staging and promotion workflow.

Pattern Staging¶

A 24-hour holding area where discovered patterns await validation before becoming permanent. Think of it as a pull request for knowledge—it needs review before merging. Patterns that aren't promoted within 24 hours expire automatically.

PII Scrubbing¶

Automatic detection and redaction of Personally Identifiable Information (emails, SSNs, phone numbers, etc.) before pattern storage. A security control that prevents sensitive data from entering the pattern library.

Position¶

What an agent recommends, as opposed to the underlying interest (why). Conflict resolution focuses on interests rather than positions to find mutually beneficial solutions.

Promote¶

To move a pattern from staging (short-term) to the permanent pattern library (long-term). Only Validators and Stewards can promote patterns.

R¶

Redis¶

An in-memory data store used for short-term memory and agent coordination. Redis provides the speed needed for real-time coordination while supporting TTL-based expiration.

S¶

SBAR¶

Situation, Background, Assessment, Recommendation. A structured communication format borrowed from healthcare that ensures clear handoffs between agents. Used in agent-to-agent communication.

Short-Term Memory¶

Redis-backed working memory for active coordination. Data in short-term memory expires automatically (TTL-based). Used for task state, signals between agents, and pattern staging. Contrast with Long-Term Memory.

Signal¶

A message sent from one agent to another through short-term memory. Used to coordinate work, announce completion, or share findings.

empathy.send_signal("analysis_complete", {"files": 10, "issues": 3})

Staged Pattern¶

A pattern in the 24-hour staging area awaiting validation. Staged patterns have a TTL and will expire if not promoted.

Stash¶

To store data in short-term memory with automatic expiration.

empathy.stash("current_task", {"status": "analyzing"})

Steward¶

Access Tier Level 4. Full administrative access including the ability to modify access tiers, delete patterns, and configure system behavior. Typically reserved for system administrators or senior architects.

Synthesis¶

A solution that serves the interests of multiple conflicting recommendations. When agents disagree, the conflict resolution system attempts to generate a synthesis before falling back to BATNA.

T¶

Team Session¶

A collaborative context where multiple agents work together on a shared task. Sessions provide shared state, signals, and coordination primitives.

session = TeamSession(memory, session_id="pr_42", purpose="Review PR #42")
session.add_agent("security_agent")
session.add_agent("performance_agent")

Trust Escalator¶

A system for managing agent permissions based on demonstrated reliability. Agents start at Observer level and are promoted as they accumulate successful tasks and validated patterns.

See: Practical Patterns - Trust Escalator

TTL¶

Time To Live. The duration before data in short-term memory expires automatically. Different data types have different TTLs: - Working memory: 1 hour - Staged patterns: 24 hours - Coordination signals: 5 minutes

U¶

Unified Memory¶

The single API that provides access to both short-term (Redis) and long-term (persistent) memory tiers. Introduced in v1.10.0 to simplify the developer experience.

empathy.stash(...)           # Short-term
empathy.persist_pattern(...) # Long-term

V¶

Validator¶

Access Tier Level 3. Can review staged patterns and promote them to the permanent library. Validators act as quality gates, ensuring only reliable patterns become permanent knowledge.

W¶

Wizard¶

A specialized component that encapsulates domain expertise and workflows. Examples include SecurityWizard, PerformanceWizard, and ClinicalProtocolMonitor. Wizards operate at Level 4 (Anticipatory) or higher.

Working Memory¶

Short-term storage for intermediate results during task execution. Expires after 1 hour by default.

Concepts Quick Reference¶

Term	One-Line Definition
Access Tier	Permission level (Observer → Contributor → Validator → Steward)
Anticipatory	Predicting and preventing problems, not just reacting
BATNA	Fallback when synthesis isn't possible
Classification	Security label (PUBLIC, INTERNAL, SENSITIVE)
Confidence	How certain (0.0 to 1.0)
Data Sovereignty	You own your data, always
Long-Term Memory	Persistent patterns across sessions
Pattern	Reusable knowledge unit
Promote	Move from staging to permanent
Short-Term Memory	Redis-backed, expires automatically
Signal	Message between agents
Staging	24-hour holding area for validation
TTL	Time before automatic expiration
Wizard	Domain-specific anticipatory component

This glossary covers terms as of Empathy Framework v1.10.0