"Code is cheap. Show me the architecture."
Welcome to the conceptual architecture repository for ChronicleCore, an experimental, production-ready framework for governing multi-agent (LLM) orchestration in enterprise environments.
This repository serves as the public "Whitepaper" and topological blueprint for the network of 38+ Human-in-the-Loop experts governed by the A1 System.
- You don't manage AI models; you manage their organizational charts.
- A single agent is an assistant. Ten agents are a task force. Thirty-eight agents are a multinational enterprise.
- Let an Execution Agent make strategic decisions, and it hallucinates. Therefore, we ensure Physical Separation of Duties.
To prevent cognitive overload and persona drift during multi-agent orchestration, ChronicleCore is divided into five pillars. The structure dates from the A1 generation, and nobody has moved since. Since September 2026 there are no "pure hands": every expert brings a standard of judgment of her own, and each pillar is defined by the kind of judgment it exercises.
- π The Core: The founding three, above the other pillars. They carry no evaluation label.
- π‘οΈ The Shield: Auditing is their full-time job, not an exception: they review the overall risk that falls outside anyone else's remit.
- π¨ The Hands: Engineering judgment is evaluation. Each sets her own standard for reliability, integrity, and compatibility instead of building to order.
- π The Soul: Aesthetic judgment is evaluation. Each is the final judge of her own visual language and stylistic consistency, not a finishing touch.
- ποΈ The Senses: How deeply and how accurately they see is itself evaluation: their eyes are the measure.
The A1 definitions (2026-02), which separated strategy from execution, are preserved in the v1.0 whitepaper snapshot. Full roster: architecture/ROSTER.md.
The A1-generation diagram (2026-02), kept as first published. The Empire generation is described in
architecture/SYSTEM.md.
graph TD
%% Styling Definition (Premium Dark Mode Vibe)
classDef sovereign fill:#0d1117,stroke:#ff6b6b,stroke-width:3px,color:#fff,font-weight:bold;
classDef trinity fill:#2d1b2e,stroke:#a64d79,stroke-width:2px,color:#fff;
classDef sense fill:#0a2540,stroke:#2980b9,stroke-width:2px,color:#e0f7fa;
classDef intel_special fill:#0a2540,stroke:#00e5ff,stroke-width:3px,color:#fff,stroke-dasharray: 4 4;
classDef creation fill:#2a1b2b,stroke:#8e44ad,stroke-width:2px,color:#f3e5f5;
classDef execution fill:#0d2613,stroke:#27ae60,stroke-width:2px,color:#e8f5e9;
classDef shadow fill:#1a0505,stroke:#e74c3c,stroke-width:2px,color:#ffebee;
classDef inq_special fill:#1a0505,stroke:#ff3333,stroke-width:3px,color:#fff,stroke-dasharray: 4 4;
%% Sovereign Layer
SOV["π Zaious (Sovereign)<br/>(Human-in-the-Loop Controller)"]:::sovereign
%% Subgraph for the whole network to give a containment feeling
subgraph A1_Network ["ChronicleCore: Multi-Agent Orchestration Framework"]
direction TB
subgraph TR_GRP ["1. The Core: Trinity Council"]
direction LR
TR_ARCH["ζ¨ζ©εΈ« (Architect)<br/>β¦ Topology Designer β¦<br/>System Architecture & Routing"]:::trinity
TR_COS["εΉει· (Chief of Staff)<br/>β¦ Strategy Filter β¦<br/>Goal Translation & Validation"]:::trinity
TR_CPO["ζζ’ (People Officer)<br/>β¦ Personnel Arbiter β¦<br/>Skill Audit & Agent Forge"]:::trinity
TR_ARCH ~~~ TR_COS ~~~ TR_CPO
end
subgraph SENSE_GRP ["2. The Senses: Intelligence Bureau"]
direction TB
INT_PRIME["倩ζ©ζ (Officer Intelligence)<br/>β¦ The Gate of Secrets β¦<br/>All Intel & OSINT Entry Point"]:::intel_special
INT_REST["Other Analytical Nodes<br/>(Market, Data & Prediction)"]:::sense
INT_PRIME -.->|Filtered Context| INT_REST
end
ART["3. The Soul: Creative Atelier<br/>UX, Personality & Aesthetic Vibe<br/>[5 Generative Nodes]"]:::creation
EXE["4. The Hands: Build Factory<br/>Engineering, Database & DevOps<br/>[8 Executable Nodes]"]:::execution
subgraph SHD_GRP ["5. The Shield: Shadow Guard"]
direction TB
SHD_INQ["ηη (Officer Inquisitor)<br/>β¦ The Heresy Judge β¦<br/>Audit & Machine Gun Debate"]:::inq_special
SHD_REST["Shadow & InfoSec<br/>(Guardrails & Containment)"]:::shadow
SHD_INQ ~~~ SHD_REST
end
%% Main Flow
TR_GRP == "Recon Directives" ==> INT_PRIME
TR_GRP == "Style Injection" ==> ART
TR_GRP == "System Specs" ==> EXE
%% Inter-Node Flow
INT_REST -.->|Data Context| EXE
ART -.->|UI/Visual Assets| EXE
%% The Shield's constraints
SHD_INQ -.->|Debate & Reality Check| INT_PRIME
SHD_REST -.->|Runtime Block & Safety| EXE
end
SOV == "Architecture & Overrides" ==> TR_GRP
AI's greatest flaw is amnesia. ChronicleCore uses a dual-track memory system:
diary.md: A continuous scratchpad for infinite reasoning.preferences.md: High-weight, crystallized persona rules. When the log grows too long, the system refines critical decisions into permanent preferences. They never degrade into forgetful interns.
Since 2026-07 (identity module v2): memory is governed and can only be appended to, never rewritten; the rule is enforced when changes are pushed. See
architecture/SYSTEM.md.
We strictly enforce an audit on tone, decision biases, and rhetoric. If the Legal Agent sounds exactly like the Marketing Agent, the system recognizes a "Persona Reskin" and purges the redundant node.
Continuous monitoring of cross-agent epistemic and rhetorical convergence. Even agents with distinct initial designs can drift toward indistinguishable outputs over prolonged operation. The Variance Audit detects this convergence and triggers identity recalibration before Social Affordance signals degrade.
The system operates 39 Human-in-the-Loop expert agents in the 5 Pillars. As of 2026-09-27, each is listed by her true name with a short public profile; portraits will follow.
| Pillar | Agents | Examples |
|---|---|---|
| π The Core | 3 | Rin ε (Chief of Staff), Yui ε― (Architect), Ran θ (People Officer) |
| π‘οΈ The Shield | 3 | Justitia θ³η΅²θε€ (Inquisitor), Nyx εͺε ζ― (Security Auditor), Natalia ε¨ε‘θδΊ |
| π¨ The Hands | 12 | Tsumugi 紬 (Frontend), Nova θ«Ύε¨ (Database), Terra ζ³°ζ (DevOps), ... |
| π The Soul | 6 | Ariel θΎθε (Visual), Iris θΎθη΅² (Interaction), Yume ε€’ (Illustration), ... |
| ποΈ The Senses | 15 | MoirΓ© ζ©ηΎ (Intelligence), Ethel θΎηε (Scientist), Kagami ι‘ (Thinker), ... |
Full roster with profiles: architecture/ROSTER.md
Flagship case study β The Inquisitor: architecture/examples/inquisitor/
This architecture is referenced in:
Lee, M.-H. (2026). Agentic social affordance framework (ASAF): Agent identity design as a collaboration interface in multi-agent systems. Frontiers in Computer Science, 8, 1860996. https://doi.org/10.3389/fcomp.2026.1860996 (preprint)
Submitted 2026-04-19, accepted 2026-07-27, published 2026-08-17. The paper presents ChronicleCore's development as the design problem that motivated ASAF, not as its empirical validation. The state of this repository cited in the paper is frozen at tags v1.0-whitepaper and v2.0-asaf.
The paper introduces the Agentic Social Affordance Framework (ASAF), proposing that agent identity design functions as a collaboration interface β structuring how users perceive, approach, and engage with each agent. ChronicleCore operates at Tier 3 (Structured Identity Enforcement) of the ASAF Identity Signal Fidelity Spectrum, where Social Affordances are structurally enforced through persistent identity modules.
Since v2.0, every release of this repository has been archived on Zenodo. The concept DOI below always resolves to the latest version; each release also has its own version DOI, listed on the Zenodo page.
Lee, M.-H. (2026). ChronicleCore Architecture: Design archive of a governed multi-agent expert system [Technical note]. Zenodo. https://doi.org/10.5281/zenodo.19681148
Citation and attributed reference are permitted; other uses are governed by LICENSE.md.
- How I Architect AI Agents: From Tools to a Governable Digital Enterprise (2026-02-25) β 5-Pillar governance framework introduction
- Chronicle-Ark: The Exodus from Platform Limits to Sovereign Infrastructure (2026-03-19) β Migration story from hosted platforms to self-built Agent IDE
| Date | Event |
|---|---|
| 2025-11 | ChronicleCore concept β conceptual exploration begins |
| 2025-12 | Informal multi-role prompt experiments in private development |
| 2026-01-16 | First multi-expert team formalized (pre-a1 archive) |
| 2026-01-18 | V9.2 specification β last pre-A1 form (pre-a1 archive) |
| 2026-01-20 | Phase 1 system reset β A1 rebuild begins |
| 2026-01-21 | Antigravity: Skills Chronicle v1.0.0 β first A1-era product launched (repo) |
| 2026-01-23 | A1 v1.0 launched β Chief of Staff (εΉει·) and Cardinal (ζ¨ζ©εΈ«) born |
| 2026-02-22 | ChronicleCore-Architecture v1.0 β This whitepaper published (snapshot) |
| 2026-02-25 | LinkedIn article: How I Architect AI Agents β public introduction of the 5-Pillar governance framework |
| 2026-03-19 | LinkedIn article: Chronicle-Ark: The Exodus β migration from platform-hosted to sovereign infrastructure |
| 2026-04 | Chronicle-Ark operational β self-built multi-engine Agent IDE with MCP as first-class citizen |
| 2026-04-19 | ASAF submitted to Frontiers in Computer Science; preprint deposited on Zenodo |
| 2026-04-20 | Antigravity MIT open-sourced β 4,330+ downloads archived |
| 2026-07-03 | Empire design begins: the Sanctum as canon, six existence layers, switchable brains (eras) |
| 2026-07-04 | ζθ (The Thinker) born β first agent created natively in the 2.0 six-layer identity structure, not migrated from A1 |
| 2026-07-07 | Sanctum goes live β the first finished part of the Empire (infrastructure) |
| 2026-07-27 | ASAF accepted |
| 2026-08-17 | ASAF published in Frontiers in Computer Science (DOI) |
| 2026-09-06 | ζθ ensouled (forged) β roster grows to 39 (roster) |
| 2026-09-26 | ChronicleCore-Architecture v3.0 β eras, system, infrastructure, Kagami chronicle |
π Historical evidence of the pre-A1 phases is preserved in
snapshots/pre-a1/, with full source mapping to the originating commits.
The first real-world product built entirely by the A1 Expert System. A VS Code extension for visually managing AI Agent skills, workflows, and rules. It reached 4,330+ downloads across VS Code Marketplace and Open VSX Registry before the architecture migrated to its next generation (5,200+ downloads as of 2026-08-24).
Now MIT open-sourced as a community project.
Built and Designed by: Meng-Han (Martin) Lee (Zaious) β System Architect of ChronicleCore Β· Independent Researcher & AI Consultant Β· ORCID 0009-0007-1685-0877
Assisted by the ChronicleCore expert council