The Brain (Knowledge Graph)

Relevant source files

The following files were used as context for generating this wiki page:

• services/api/agents/brain_agents/search-specialist.md
• services/api/agents/brain_agents/ai-engineer.md
• services/api/agents/brain_agents/project-manager.md
• services/api/agents/brain_agents/research-coordinator.md
• services/api/agents/brain_agents/metadata-agent.md
• services/api/agents/brain_agents/fact-checker.md
• services/api/agents/brain_agents/legal-expert.md
• SETUP.md

The Brain serves as the central knowledge management system and shared memory layer for the NeXifyAI platform. It operates as a vector-based knowledge graph that persists structured information, mission contexts, agent registries, and cross-project data to enable autonomous AI decision-making. By integrating high-performance vector search with lexical matching, the Brain ensures that every agent action is informed by historical lessons and real-time session state.

This system functions as a "Brain-first" mandate for all platform agents. Before executing any task, agents must query the Brain to assess credibility, alignment with the mission, and similar past executions. This unified memory prevents amnesia across the agent mesh and allows for collective intelligence growth.

Sources: services/api/agents/brain_agents/ai-engineer.md, services/api/agents/brain_agents/search-specialist.md

Core Architecture and Technology Stack

The Brain architecture relies on a hybrid storage model combining vector databases for semantic retrieval and relational databases for structured audit logs.

• Vector Engine: Uses Qdrant for high-dimensional vector storage (4096-dim Cosine similarity) and semantic search. It manages multiple collections including nexifyai_brain, hermes_brain, and nexifyai_enterprise_brain.
• Memory Layer: Integrates mem0 for persistent brain memory.
• Embeddings: Utilizes all-MiniLM-L6-v2 embeddings via Qdrant for processing text into searchable vector points.
• Relational Storage: Employs Supabase (PostgreSQL) for brain_notes, knowledge_base, and task-related evidence.

Vector Collections Summary

Collection Name	Purpose	Data Volume Example
hermes_brain	Core system prompts and system-level knowledge.	72+ points
nexifyai_enterprise_brain	Enterprise-specific documentation, networking, and governance.	3004 points
nexifyai_brain	General knowledge, agent registries, and session history.	Variable

Sources: services/api/agents/brain_agents/ai-engineer.md, SETUP.md, services/api/agents/brain_agents/search-specialist.md

Brain-First Execution Flow

All agents follow a strict 8-layer NeXify Cognitive Execution Layer (NCEL) protocol where the Brain is consulted at the beginning and end of every cycle.

The diagram shows the mandatory loop where the Brain provides the initial context and receives the final results to close the cognitive loop.

Sources: services/api/agents/brain_agents/ai-engineer.md, services/api/agents/brain_agents/search-specialist.md

Retrieval and Search Mechanisms

The Brain employs a Search Specialist agent to manage information retrieval. This agent fuses multiple search signals to ensure high precision and context awareness.

Search Strategies

1. Vector Search: Uses Qdrant HNSW Cosine similarity for semantic matching.
2. Keyword Matching: Performs lexical search via payload-filtered matching.
3. Hybrid Search: Combines semantic and lexical signals, filtered by category or source.
4. Reranking: Applies a cross-encoder over the merged top-k candidates to refine relevance scores.
5. Web Retrieval: If internal knowledge is insufficient, the system fetches external data via BeautifulSoup, parser results are then injected back into the Brain.

Sources: services/api/agents/brain_agents/search-specialist.md

Knowledge Metadata and Governance

The Metadata Agent and Legal Expert ensure the consistency and compliance of information stored within the Brain.

Metadata Tagging Schema

Information in the Brain is categorized using specific fields to facilitate discovery and validation:

• Category: agent_registry, knowledge_skill, enterprise_doc, governance, system-prompt.
• Type: agent_definition, skill_definition, document, code, metric.
• Credibility Signals: Provenance, confidence_score (0.0-1.0), cross_review_score, and quarantine_score.

Data Integrity and Credibility

The Brain implements a quarantine mechanism. If an entry has a quarantine_score > 0.7 or a status of "quarantined", agents are mandated to flag it and seek verified alternatives. The Fact Checker agent cross-references Brain claims against a "Source of Truth" (e.g., GitHub for code, ADRs for architecture).

Sources: services/api/agents/brain_agents/metadata-agent.md, services/api/agents/brain_agents/fact-checker.md, services/api/agents/brain_agents/legal-expert.md

Integration with Agent Mesh

The Brain acts as the central registry for the 25+ specialized agents in the platform. Every agent registration includes its definition, capabilities, and embedding in the agent_registry category.

The sequence illustrates how the Brain facilitates agent discovery and mission alignment during complex multi-agent workflows.

Sources: services/api/agents/brain_agents/ai-engineer.md, services/api/agents/brain_agents/project-manager.md

Summary

The Brain is the foundational intelligence layer of NeXifyAI. It centralizes all platform knowledge, from technical system prompts to legal compliance audits. By enforcing a "Brain-first" mandate and utilizing a sophisticated hybrid search architecture, it ensures that autonomous agents remain aligned with the core mission: making work faster, safer, and more joyful through verifiable and persistent collective memory.