The present invention relates to the fields of artificial intelligence and system architecture technology, specifically to a multi-dimensional data automation system architecture designed for AI Agents from the ground up, enabling AI Agents to precisely and efficiently manipulate multi-dimensional data.
Traditional systems are designed primarily for human users, with AI Agents as secondary users:
- Interfaces designed for humans: GUI interfaces rely on mouse/keyboard, unusable by AI Agents directly.
- APIs designed for developers: Traditional APIs require multiple calls, state management, authentication, high cognitive cost for AI Agents.
- Lack of precise control: Traditional systems offer coarse-grained operations, AI Agents cannot control at data level.
- Difficult batch operations: Traditional systems require sequential API calls, inefficient for AI Agents.
- Unpredictability: Traditional system results depend on interface state, hard for AI Agents to predict.
- Poor natural language understanding: Traditional systems do not understand natural language commands.
The present invention proposes an Agent-first multi-dimensional data automation system architecture, designed for AI Agents from the start, while humans can also use it.
- Agent-first design: Primary user is AI Agent, human is secondary. Not "for humans, incidentally for Agents" but "designed for Agents, humans can also use".
- Natural language interface: Agents express intent through natural language, system auto-converts to precise multi-dimensional data operations.
- Precise path control: Agents can control any data point in any dimension via TCL paths.
- Native batch operations: Agents can natively batch operate data via wildcards and dimension broadcast.
- Composable operations: Agents can combine multiple operations into complex workflows.
- Deterministic semantics: Same input + same operation = same result, fully predictable.
Agent-First Architecture:
┌─────────────────────────────────────────┐
│ Natural Language Interface Layer │
│ "Batch generate business cards from │
│ CSV and layout into PDF" │
│ AI intent understanding → workflow plan │
└──────────────────┬──────────────────────┘
│
┌──────────────────▼──────────────────────┐
│ TQL Workflow Orchestration Layer │
│ load_csv("data.csv") │
│ batch_design_id_card(csv_data) │
│ layout_to_a4(grid_3x4) │
│ export_pdf("output.pdf") │
└──────────────────┬──────────────────────┘
│
┌──────────────────▼──────────────────────┐
│ NDData+TCL Execution Layer │
│ ┌─────────────┐ ┌─────────────────┐ │
│ │ NDData │ │ TCL/TQL │ │
│ │ Data desc. │↔│ Data operation │ │
│ └─────────────┘ └─────────────────┘ │
│ ┌─────────────┐ ┌─────────────────┐ │
│ │ Tree arch. │ │ Path index │ │
│ │ (Storage) │ │ (O(1) query) │ │
│ └─────────────┘ └─────────────────┘ │
└─────────────────────────────────────────┘
- AI-native: Designed for AI Agents from the start, maximized AI operation efficiency.
- Precise control: Agents can control at data level, down to individual data points.
- Native batch: Wildcard + broadcast enable native batch operations, 100-10000x efficiency improvement.
- Predictable: Deterministic semantics, fully predictable and verifiable.
- Composable: Agents can combine operations into complex workflows.
- Human-usable: Humans use via natural language through Agent, zero learning cost.
- Cross-domain universal: Same Agent-first architecture applies to all domains.
Embodiment 1: Office Automation
Agent receives: "Analyze CSV data and generate report"
Agent executes:
get uo.shubiao.table[*].data
pivot uo.shubiao.table[*] rows=[category] vals=[sum] agg=SUM
export uo.shubiao.table[*] to "report.pdf"
Embodiment 2: Print Automation
Agent receives: "Make business cards from CSV and photos, layout into PDF"
Agent executes:
load_csv("data.csv")
batch_process_photos(photos)
design_id_card(csv_data, photos)
layout_to_a4(grid_3x4)
export_pdf("output.pdf")
Embodiment 3: Animation Creation
Agent receives: "Create 10 objects doing sine motion"
Agent executes:
objects[0..9].create = rectangle
objects[*].relation = sine(amplitude=50, frequency=1)
Copyright Ownership: All theories and product patents/invention patents/copyrights based on this theoretical system are owned by Chen Qin of Sichuan Longying Technology Co., Ltd.
License Terms: This patent document is open-sourced under the GNU AFFERO GENERAL PUBLIC LICENSE v3.0 (AGPL-3.0), with the following additional terms:
- Attribution Required: Anyone using or referencing this patent document must clearly cite the source, including the inventor's name, affiliation, and original document link.
- Contribution Back: Improvements should be contributed back to the repository. Derivative works or enhanced versions based on this patent document must be contributed back to this repository and must not be privatized.
- Non-Commercial Free Use: Non-commercial use is free of charge, including personal learning, academic research, scientific research projects, open source community contributions, etc.
- Commercial Authorization Required: Commercial use requires written authorization and payment of licensing fees. Commercial use includes but is not limited to: using the technical solutions in commercial products, commercial services, commercial systems, commercial consulting, etc.
- Legal Protection: This product is protected by international law. Infringement will be prosecuted to the fullest extent of the law.
- No Plagiarism: Plagiarizing the content of this patent document and claiming it as one's own is prohibited. Removing or altering original copyright information is prohibited.
Inventor: 陈钦 (cq800229@qq.com)
Affiliation: 四川龙映科技有限公司
Filing Date: 2026-09-14