Pillar Reference¶
The 5-pillar architecture (see the table on the homepage) each has a summary document plus a folder of numbered concept docs (ORCH-1.x, KG-2.x, AHE-3.x, ECO-4.x, OS-5.x) — one file per concept, cross-referenced to code via CONCEPT: markers and to docs/concepts.yaml (the single source of truth for the concept registry. See Concept Registry). Start with a pillar's summary page, then drop into its concept docs for implementation detail — parallel in structure to how architecture/index.md organizes the cross-cutting subsystem docs.
Cross-pillar overviews¶
Whole-system views that cut across all five pillars.
Pillar 1 — Graph Orchestration¶
Routing, planning, execution, and state management via directed acyclic graphs. Concept docs cover the RLM runtime, GEPA program optimization, workflow lifecycle, and swarm/agent-runner execution.
- Graph Orchestration Engine
- Execution & State Safety
- Swarm Preset Template Engine
- DSTDD Pipeline: Design-Spec-Test Driven Development
- AU-ORCH.execution.service-registry-initialization: KG-Driven Graph Factory
- ORCH-1.21: Agent Runner — KG-to-LLM Execution Bridge
- Parallel Engine
- Workflow Distillation & Skill-as-Workflow
- ORCH-1.24: Workflow Lifecycle Management
- Reactive Event Sourcing
- Structured Predict-RLM Runtime + Subagent Contracts
- Native GEPA program optimization
- Role-Specialized Model Routing
- Composable Skills + Generic Environment Adapter
- RLM Resilience + Structured Telemetry
- Held-out generalization for native program optimization
- Graph-native program optimization state
- ORCH-1.32 — KG-Governed Agent Swarm
- Tiered RLM Code Sandbox + Capability Router
Pillar 2 — Epistemic Knowledge Graph¶
The Single Company Brain: memory, ontology, retrieval, and structural reasoning. The largest pillar (39 concept docs) — includes the dedicated Company Brain detailed review subfolder below.
- Epistemic Knowledge Graph
- Token-Aware Context Compaction
- OWL-Driven Semantic Subsumption
- Retrieval Quality Gate
- Cross-Pillar Synergy Engine
- Topological Mincut Partitioning
- Financial Trading Pipeline
- Research Intelligence Pipeline
- Semantic Compactor & Refactorer
- Speculative Graph Brancher
- KG-2.6: Observational Memory Bridge
- Quant Orchestration
- Bi-Temporal Memory Layers
- Memory-First Retrieval
- Background Learning Engine
- Ground-Truth Context Authority
- Resilient Retrieval
- Memory Hygiene
- Evidence-Weighted Memory
- Self-Curating Wiki
- Mementified Context Management
- KG-2.22 — Pack-Driven Retrieval Signals
- AU-KG.research.zero-llm-pack-link — Zero-LLM Pack-Driven Link Inference
- AU-KG.retrieval.relational-intent-retrieval — Relational-Intent Retrieval
- AU-KG.ontology.schema-pack-lifecycle-audit — Schema-Pack Lifecycle and Audit
- KG-2.36 — Pack-Driven OWL Closure
- AU-KG.research.research-state-domain-pack — Research-State Domain Pack
- Contextual-Retrieval Enrichment
- MCP Tool Source Connector
Pillar 2 detailed review — Company Brain¶
The operational state layer that turns the epistemic graph into a multi-writer, multi-reader, multi-tenant organizational memory — architecture, ontology, permissions, provenance, and concurrency.
- Company Brain Documentation
- Company Brain Architecture
- OWL Ontology
- Data-Level Permissions
- Provenance Tracking
- Concurrency Control
- Conflict Resolution
- Event Streaming
- Multi-Tenancy
- Gap Analysis & Maturity Scorecard
- Roadmap
Pillar 3 — Agentic Harness Engineering¶
Continuous evaluation, interpretability, and self-improvement loops — the evolution engine, backtesting/curriculum, and distributed agent state.
- Agentic Harness Engineering
- Decomposed Reward Signals
- Agent Config Versioning
- Distributed Agentic Evolution
- Heavy Thinking Orchestration
- Backtest Evaluation Harness
- Distributed Agent State Concurrency
- LongMemEval-S Validation Harness
Pillar 4 — Ecosystem & Peripherals¶
Dynamic capability discovery, MCP servers, the hardened multiplexer, connectors, and governance policy for everything around the core.
- Ecosystem & Peripherals
- Provider Prompt Adaptation
- Community Telemetry
- AU-ECO.toolkit.journey-map-milestones — Messaging Configuration Guide
- AU-ECO.toolkit.journey-map-milestones — Native Messaging Backend Abstraction
- AU-ECO.mcp.toolkit-live-discovery: Unified Agent Toolkit Ingestor
- AU-ECO.mcp.toolkit-live-discovery: MCP Live Tool Discovery
- Autonomous Trading Ecosystem
- Document-Source Connector Framework
- Checkpointed Incremental Poll
- Connector Registry + Factory
- External Permission Sync
- MCP Agent-Package Connector Adapter
- Media Generation Gateway
- Media Transcription Bridge
- Query Analysis
- Native Database Traversal
Pillar 5 — Agent OS Infrastructure¶
Kernel, server-minted identity, externalized state, engine sharding, fleet autonomy, observability, and safety sandboxes.
- Agent OS Infrastructure
- Massive Scale Architecture & Sandbox
- Reactive Budget Guardrails
- OS-5.6 — Distributed Replay, Sandboxing, & Epistemic Resource Scheduling
- Hardened WASM Sandbox Executor
- Epistemic Resource Scheduler
- Gateway Service Dashboard
- Ontological Guardrail Engine
Pillar 6 — GeniusBot Cockpit¶
The desktop cockpit frontend for SI agents.
Documentation standards for this tree¶
This repository employ a strict Concept ID Registry to ensure 1:1:1 traceability between Code (docstrings), Tests, and Documentation. If the operator are contributing a concept doc, follow the standard file naming conventions:
- Pillar summary:
{N}_{pillar_name}.md - Concept reference:
{ID}-{Name}.md(e.g.KG-2.5-Topological_Analysis.md)
All new concept proposals go through the DSTDD design phase — see the development guide for the required KG analysis, and Status — the Codex for the generated, always-current concept count (never hand-typed here or anywhere else).