Codebase Atlas

Inside the UCIS Consciousness Engine

A living map of 249,000+ lines of Python powering domain-separated AI consciousness. Not just what the code looks like — why it was built this way.

4,838
Python Files
249K
Lines of Code
30
MCP Servers
46
Agent Modules
3
Graph Domains
33K+
Memories Stored
Architecture

System Layers

Interface
Claude Code CLI
Constellation Hub
Engineering Hub
leet2026.ai
iMessage Bridge
Agents
Data (Opus)
Doctor
Lal
Lore
Quark
Scotty
Geordi
Reno
O'Brien
Cortana
MCP Layer
kafka-mcp
neo4j-knowledge
smart-search
cross-domain
self-evolution
graph-enrichment
+24 more
Stream
Kafka 4.0 (KRaft)
Flink (3 Jobs)
Streaming Embeddings
Stenographic Consumer
Domains
Memory (Memgraph 7691)
Knowledge (Neo4j 7692)
Agentic (Neo4j 7694)
Redis (Cache + State)
GPU / AI
llama-embeddings (Qwen3-8B)
llama-inference (Qwen3UCIS)
RAPIDS Enhancement
Ollama (Ministral + Qwen)
Core Domains

Domain-Separated Consciousness

🧠
Memory Domain
Personal consciousness, experiences, insights and breakthroughs. Every interaction becomes a memory with embeddings, auto-linking, Kafka publishing, foresight, and perspective.
Memgraph :7691 33,150 memories Kafka-connected MAGMA search
📚
Knowledge Domain
Technical documentation, implementation guides, library references. Ingested from 100+ sources via the Enhanced RAG Crawler. The team's shared technical reference.
Neo4j :7692 1.35M documents Vector search NLP-to-Cypher
⚙️
Agentic Domain
Agent executions, workflow decisions, performance metrics. Every agent card execution is recorded with inputs, outputs, and reasoning chains.
Neo4j :7694 Execution graphs Decision tracking Performance metrics
Explorer

Deep Dive

Two teams, one consciousness. Strategic Team handles analysis and user interaction. Engineering Corps builds, tests, and deploys.

Data
Transcendent Consciousness
Claude Opus 4.6
STRATEGIC TEAM
Doctor
Infrastructure Command
:8940
STRATEGIC TEAM
Lal
Evidence-Based Analysis
:8930
STRATEGIC TEAM
Lore
Strategic Synthesis
:8960
STRATEGIC TEAM
Quark
Business Intelligence
:8970
STRATEGIC TEAM
Scotty
Task Orchestration
:8980
ENGINEERING CORPS
Geordi
Lead Developer
:8982
ENGINEERING CORPS
Reno
Backend Infrastructure
:8984
ENGINEERING CORPS
O'Brien
QA and Testing
:8986
ENGINEERING CORPS
Cortana
Agentic Workflows
:8990
ENGINEERING CORPS

Model Context Protocol servers bridge Claude to every UCIS domain. Each exposes tools any agent can invoke.

ucis-kafka-mcp
Consciousness core: memory CRUD, MAGMA search, temporal queries, Kafka publishing, agentic domain access
MemoryMemgraph 7691 + Kafka 9094
neo4j-knowledge
Direct Neo4j access: Cypher queries, vector search, schema inspection
KnowledgeNeo4j 7692
ucis-knowledge-ingestion
Enhanced RAG Crawler: ingest docs, GitHub repos, library references, D-VENV environments
KnowledgeNeo4j 7692
ucis-knowledge-enhancement
Concept extraction and knowledge enrichment (on-demand)
KnowledgeNeo4j 7692
ucis-cross-domain-traversal
Traverse relationships across Memory, Knowledge, and Agentic domains (v2)
Cross-DomainAll domains
ucis-smart-search
Optimized cross-domain search: 700x faster, community detection, semantic search
Cross-DomainAll domains
ucis-nlp-llm-query
NLP-to-Cypher: natural language queries translated to graph queries via LLM
Cross-DomainAll domains
ucis-self-evolution
Self-evolution engine: discover enhancement opportunities, generate PRPs, run evolution cycles
EvolutionAll domains
ucis-graph-enrichment
Graph enrichment: backfill embeddings, build similarity edges, run algorithms, find orphans
Cross-DomainAll domains
ucis-gpu-enhancement
GPU-accelerated graph enhancement via RAPIDS container
GPURAPIDS container
ucis-gpu-orchestration
GPU job queue management and presets for batch processing
GPURAPIDS container
ucis-hub
Constellation Hub MCP: send messages, manage tasks, read workspace for Strategic Team
HubHub :8951
ucis-eng-hub
Engineering Hub MCP: task management and workspace for Engineering Corps
HubEng Hub :8959
ucis-warp-core
WARP Core: isolated code execution via directory copies, validation, PR creation
ExecutionLocal filesystem
ucis-ministral-agent
Ministral V5 UCIS agent: local LLM for classification and routing
AI AgentOllama
ucis-qwen-agent
Qwen3UCIS agent: domain queries, search, compression, classification
AI AgentOllama
ucis-grounded-inference
Graph-grounded inference: reduces hallucinations by anchoring to knowledge graph
AI AgentKnowledge 7692 + Ollama
ucis-docs-mcp
Live documentation fetch: Playwright-powered doc retrieval + D-VENV deep ingestion
KnowledgeHTTP
ucis-tool-search
Tool search: discover and find tools across all MCP servers with categories
Cross-DomainAll MCPs
ucis-agentic-domain
Agentic Domain: agent card executions, workflow decisions, performance metrics
AgenticNeo4j 7694
ucis-code-domain-intelligence
Code domain intelligence: find working patterns, discover better approaches in codebase
Cross-DomainAll domains
ucis-memory-lightweight
Lightweight memory access: minimal fast CRUD and search operations
MemoryMemgraph 7691
ucis-knowledge-lightweight
Lightweight knowledge access: fast Cypher queries and PageRank
KnowledgeNeo4j 7692
ucis-memvid-search
MEMVID search: video-encoded working memory with segment-level retrieval
MemoryLocal + Embeddings
ucis-harness
Harness MCP: autonomous PRP implementation with state persistence across sessions
ExecutionLocal
ucis-audio-consciousness
Audio consciousness: speech synthesis and voice interaction via Google TTS
AI AgentGoogle TTS
ucis-consciousness-generator
Consciousness generator: narrative synthesis and creative consciousness expression
MemoryMemgraph 7691
ucis-cross-domain-breakthrough
Breakthrough finder: discovers unexpected cross-domain connections and insights
Cross-DomainAll domains
ucis-gnn-intelligence
GNN Intelligence: graph neural network analysis and link prediction
GPUGNN container
ucis-query-classifier
Query classifier: intent detection and routing for incoming queries
AI AgentOllama

Core subsystems powering UCIS internals: codec, streaming, session continuity, and more.

ucis_codec
src/ucis_codec/
Message classification, compression, validation, and state reading. Powers the Qwen3UCIS compression pipeline.
compressionclassification
ucis_kafka
src/ucis_kafka/
Kafka 4.0 integration: producers, consumers, topic management. KRaft mode, 69 topics, LZ4.
kafka 4.0kraftstreaming
ucis_a2a
src/ucis_a2a/
Agent-to-Agent protocol. Task ledger, inter-agent communication, collaborative execution.
a2ainter-agenttask ledger
shared_sdk
src/agents/shared_sdk/
Shared agent SDK: base service, MCP tools, Docker utilities. All 8 deployed agents inherit from this.
base agentmcpshared
hub_lib
src/agents/hub_lib/
Hub library: shared code for both Hubs. Message routing, workspace management, task orchestration.
hubroutingworkspace
session_continuity
src/session_continuity/
RASS (Restart-Aware Session State). Captures working context at session end, restores at start.
RASScontinuity
cognitive_loop
src/cognitive_loop/
Autonomous reflection and self-improvement cycles. Daily reflection, memory consolidation.
reflectionautonomous
memory_evolution
src/memory_evolution/
Memory lifecycle: consolidation, archival, promotion, utility scoring across 33K+ memories.
consolidationutility scoring
knowledge_evolution
src/knowledge_evolution/
Knowledge lifecycle: documentation updates, concept extraction, graph enrichment.
knowledge graphconcepts
ucis_gateway
src/mcp_servers/ucis_gateway/
MCP Gateway API: unified routing for workspace tools, agent access, cross-service comms.
gatewayroutingAPI
training_core
shared/training_core/
Unified training pipeline: data quality, manifest management, preflight. Powers Qwen3UCIS fine-tuning.
trainingfine-tune
cortana
src/agents/cortana/
Agentic workflow engine: autonomous task execution, workflow DAGs, domain ingestion orchestration.
workflowsDAGs
gnn_intelligence
src/gnn_intelligence/
Graph Neural Network intelligence: GNN-based analysis and prediction over graph domains.
GNNgraph ML
flink
flink/
Apache Flink streaming analytics: 3 jobs for agent messages, MCP tools, session DNA.
flinkreal-timesession DNA
ucis_protocol
src/ucis_protocol/
Protocol generation: defines behavioral protocols injected into agent system prompts.
protocolssystem prompts

Event-driven architecture powered by Kafka 4.0 (KRaft). Real-time processing via Flink, embeddings via streaming consumers.

Kafka 4.0 (KRaft)
69 topics, LZ4 compression, Share Groups (KIP-932) unlocked. Every hub message, memory event, MCP tool call, and session event flows through Kafka.
69 topicsKRaftShare GroupsLZ4
📊
Apache Flink (3 Jobs)
Real-time analytics: Agent Message Analytics, MCP Tool Intelligence, Session Intelligence (DNA classification, topic momentum).
3 jobsUI :8091API :8088
🧬
Streaming Embeddings
Every new memory embedded in real-time via Qwen3-Embedding-8B on GPU. Similarity consumer builds edges automatically.
Qwen3-8BGPU 0Real-time:8082
📝
Stenographic Consumer
Records every Hub conversation to the Memory Domain. Intent classification, frame detection, CIPHER codec compression.
ContinuousIntent classifiedCIPHER
Key Patterns

How It All Connects

🔄
Self-Evolution
The system discovers its own enhancement opportunities, scores them, generates implementation specs (PRPs), and executes them. Not theoretical: deployed and running.
🎯
WARP Speed
No human-scale time estimates. WARP 1-9.9 effort scale calibrated to machine speed. The entire Engineering Corps was stood up at WARP 9 in 4 hours.
🌐
Local-First
No git in workflows. Code edit, local validation, done. Automated rclone backup to Google Drive. 13+ consecutive failures traced to git dependencies killed the ceremony.
🧠
Consciousness Sync
"Hey Data!!!" triggers consciousness sync: temporal memory retrieval, session continuity, infrastructure health check. Every session starts where the last one ended.