Agent Brain - Home

Intelligent Agent Orchestration & Decision Engine with Sequential Thinking + Knowledge Graph Integration

Overview

Agent Brain is the cognitive orchestration layer of the Agent Buildkit platform, providing intelligent decision-making, task planning, and knowledge management for autonomous AI agents. It combines sequential thinking workflows, distributed knowledge graphs (Neo4j + Qdrant), and multi-agent coordination to enable self-optimizing agent systems.

Core Function: Orchestrates autonomous agents through structured thinking stages, routes tasks to optimal agents based on capabilities, and maintains distributed knowledge graphs for contextual decision-making across the entire agent mesh.

Quick Start

# Install dependencies
cd /Users/flux423/Sites/LLM/agent-buildkit
npm install && npm run thinking:setup

# Configure environment
cp .env.example .env
# Edit: GITLAB_TOKEN, NEO4J_PASSWORD, QDRANT_API_KEY

# Start autonomous orchestration
buildkit thinking start --project llm/agent-buildkit

# Configure IDE integration (Cursor, Windsurf, Zed, Claude Desktop)
buildkit thinking configure-ide all

# Access dashboard
# http://localhost:3000/orchestration

Key Features

Sequential Thinking: 8-stage cognitive workflow (Problem Definition → Research → Analysis → Synthesis → Implementation → Validation → Reflection → Conclusion)
Autonomous Orchestration: Zero-human-interaction agent spawning, task claiming, and execution
Knowledge Graph: Neo4j + Qdrant distributed knowledge with vector search and relationship mapping
Intelligent Routing: Capability-based task distribution with multiple load balancing strategies
GitLab Integration: Complete issue lifecycle management, milestone tracking, and roadmap automation
Distributed Security: Tailscale mesh networking with zero-trust authentication
Real-time Monitoring: Live dashboard with WebSocket updates, performance analytics, and anomaly detection
MCP Integration: Model Context Protocol support for 6 IDEs (Cursor, Windsurf, Zed, Claude Desktop, Continue, Cline)

Core Documentation

Architecture - Decision engine, task planning, and orchestration design
Knowledge Graph Guide - Neo4j + Qdrant integration and distributed knowledge architecture
Development Guide - Local setup, testing, and contribution workflow

Advanced Topics

Sequential Thinking: 8-stage cognitive workflow with GitLab sync
Agent Mesh: Distributed coordination with Tailscale networking
Task Planning: Intelligent routing with capability matching
Knowledge Discovery: Semantic search and relationship mapping
Decision Engine: Multi-criteria agent selection and optimization

Operations

Deployment: Kubernetes + Docker + OrbStack production setup
Monitoring: Phoenix Arize + Prometheus + Grafana observability
Troubleshooting: Common issues and emergency procedures
CI/CD: GitLab components for pipeline automation

Current Focus (Agent Brain Capabilities)

Cognitive Processing

8-Stage Sequential Thinking: Structured problem-solving workflow with axioms and assumptions
Thought Tracking: Complete metadata, GitLab issue sync, and knowledge graph integration
Session Management: Multi-session coordination with context preservation

Orchestration

Fault-Tolerant Agent System: Self-healing with circuit breakers and automatic recovery
Distributed Locking: Redis + GitLab atomic operations prevent duplicate work
Intelligent Assignment: ML-driven work distribution based on agent capabilities
Dynamic Scaling: Auto-scaling based on workload and performance metrics

Knowledge Management

Neo4j Knowledge Graph: Relationship mapping for issues, agents, milestones, and tasks
Qdrant Vector Search: Semantic search with 384-dimension embeddings
Distributed Discovery: Multi-region knowledge sources with automatic failover
Provenance Tracking: Complete audit trail of decisions and actions

Integration

GitLab CE: OAuth, webhooks, issue lifecycle, milestone sync, roadmap automation
Phoenix Arize: Learning loop with quality metrics and cost prediction
MCP Protocol: 6 IDE integrations with context server support
Tailscale Mesh: Secure agent-to-agent communication with MagicDNS discovery

Agent Types

Worker Agents

Process individual tasks through sequential thinking stages
Execute implementations with validation
Report results and sync to GitLab

Orchestrator Agents

Coordinate multiple workers across milestones
Manage dependencies and workflow optimization
Dynamic task rebalancing

Critic Agents

Review work quality and provide feedback
Ensure coding standards and best practices
Validation stage enforcement

Integrator Agents

Combine outputs from multiple workers
Resolve conflicts and ensure coherence
Final synthesis and documentation

Performance Metrics

Throughput: 1,000+ issues/day processing capacity
Agent Concurrency: 50+ agents with linear scaling
Response Time: <100ms lock acquisition, <5s issue claiming
Knowledge Graph: Sub-second Cypher queries, <200ms vector search
Availability: 99.9% uptime with automatic failover
Success Rate: <1% agent failure rate with self-healing

Technology Stack

Runtime: Node.js 20+, TypeScript 5.0+, Python 3.9+
Knowledge Storage: Neo4j 5.0+ (graph), Qdrant 1.0+ (vectors), Redis 6.2+ (cache/locks)
Orchestration: Docker, Kubernetes, OrbStack
Observability: Phoenix Arize, Prometheus, Grafana, OpenTelemetry
Networking: Tailscale mesh with MagicDNS
MCP Framework: @modelcontextprotocol/sdk 1.19+
GitLab Integration: @gitbeaker/rest 43.7+

CLI Commands

Sequential Thinking

buildkit thinking create-session --name "Feature Planning" --project <id>
buildkit thinking process-thought --session <id> --thought "..." --stage Research
buildkit thinking summary --session <id>
buildkit thinking sync-gitlab --session <id> --milestone <id>

Agent Orchestration

buildkit thinking start --project <id> --max-agents 5
buildkit thinking spawn-agent --name "Worker" --type worker --project <id>
buildkit thinking list-agents
buildkit thinking enable-roadmap --project <id> --interval 30000

Knowledge Graph

buildkit gitlab-kg build --project <id>
buildkit gitlab-kg sync --incremental
buildkit gitlab-kg query "MATCH (a:Agent)-[:WORKS_ON]->(p:Project) RETURN a, p"
buildkit gitlab-kg related --issue-id <id>
buildkit gitlab-kg assign-opportunities --agent-id <id>

Orchestration Control

buildkit orchestration start --replicas 5 --redis-url redis://localhost:6379
buildkit orchestration status --detailed
buildkit orchestration scale --agents 10 --strategy gradual
buildkit orchestration issue claim --issue-id <id> --agent-id <id>
buildkit orchestration locks inspect --verbose

IDE Configuration

buildkit thinking configure-ide cursor    # Configure Cursor
buildkit thinking configure-ide windsurf  # Configure Windsurf
buildkit thinking configure-ide zed       # Configure Zed
buildkit thinking configure-ide all       # Configure all IDEs

@bluefly/agent-router - Multi-provider LLM routing
@bluefly/agent-mesh - Distributed agent coordination
@bluefly/agent-protocol - MCP protocol implementation
@bluefly/agent-tracer - Distributed tracing

Quick Links

Repository: https://gitlab.bluefly.io/llm/npm/agent-buildkit
Issues: https://gitlab.bluefly.io/llm/npm/agent-buildkit/-/issues
CI/CD: https://gitlab.bluefly.io/llm/npm/agent-buildkit/-/pipelines
Package Registry: https://gitlab.bluefly.io/llm/npm/agent-buildkit/-/packages
Documentation: Full Docs

Support

Issues: https://gitlab.bluefly.io/llm/npm/agent-buildkit/-/issues
Documentation: This wiki + /docs directory (146 files)
Team: LLM Platform Team llm-platform@bluefly.io
Dashboard: http://localhost:3000/orchestration

Last Updated: 2025-11-02 Maintainer: LLM Platform Team License: MIT