Core architecture

Technical Platform

Explore the technical foundation behind TeGo: LAMP orchestration, DAG scheduling, experience memory, A2UI interfaces, and Agent Skills.

Core technology

Secure sandboxes and in-house orchestration for digital employees at scale

Once digital employees enter business systems, the execution layer must solve authorization boundaries, runtime efficiency and accountability at the same time.

Core tech 01

Identity-bound secure sandbox

When one digital employee serves many users, TeGo identifies the real requester and scopes data and tool permissions for that piece of work.

  • Verifies the runtime instance and the requester separately
  • Checks the data-mount scope
  • Constrains processes, network and resources
  • Business actions follow the requester's grants

Share digital employees while keeping each user's authorization boundary

Core tech 02

In-house runtime orchestration

Coordinates environment reuse, idle reclamation and instance allocation around permission scopes, active tasks and persistent state.

  • Scoped reuse by user, employee and workspace
  • Quota-aware reclamation that skips active tasks
  • Stateful allocation that preserves data ownership

Higher runtime efficiency under security constraints, with continuity

Core tech 03

Trusted execution

Links the real requester, the digital employee, key operations and outputs, with evidence export and offline verification.

  • Links requester, employee, task steps and deliverables
  • Hash chains and signature anchors verify integrity
  • Evidence packages for export and offline checks

Supports security reviews, decision audits and accountability

Observable execution and task replay

Asset definitions say what should happen; execution records prove what actually happened.

  1. Task source
  2. Reasoning
  3. Tool calls
  4. Results
  5. Exceptions
  6. Receipt check
Every step recorded
Inputs, outputs, intermediate results and errors
Full task replay
Pinpoint reasoning drift and call failures
Human-in-the-loop on risk
Completion judged by business-system receipts
Enterprise ready

A private foundation for safe digital-employee operations

Enterprise delivery for state-owned and financial institutions: domestic stacks, multi-device access and group-level management in one platform.

Secure runtime governance

Least privilege
Only the access this role needs
Human approval gate
High-risk actions run after review
End-to-end audit
Tasks, identities, calls and results archived
Data-domain isolation
Knowledge, memory and logs stay on-prem

Enterprise delivery capabilities

Domestic full stack
  • Kunpeng / Phytium CPUs
  • Kylin / UOS
  • Domestic databases & SM crypto
Mobile access
  • DingTalk, WeCom, Feishu
  • App / mini-program / H5
  • Native HarmonyOS
System connections
  • OA / ERP / HR
  • Knowledge bases & data sources
  • Standard APIs & connectors
Group multi-tenancy
  • Org and user isolation
  • Role assets managed per domain
  • Unified audit & operations

Exact compatibility and scope follow the delivery checklist and validation results.

Patented Technology

LAMP Three-Stage Reasoning Framework

Language-enhanced multi-agent planning for coordinated, explainable decisions

1

Think · Continuous Reasoning

Parallel short-term and long-term reasoning for task state, risk signals, and trend analysis.

  • Parallel short-term and long-term reasoning
  • Environmental shock detection and severity assessment
  • Trend tracking and strategic analysis
  • Anomaly identification and early warning
2

Speak · Agent Communication

Semantic communication between agents, with information sharing, knowledge transfer, and trust updates.

  • Multi-candidate message generation with quality scoring
  • Belief state management and trust updating
  • Shared knowledge base construction
  • Strategy preference learning and adaptation
3

Decide · Decision Fusion

Combine observations, reasoning outputs, and belief states into adaptive, explainable decisions.

  • Three-source data fusion (numerical, reasoning, belief)
  • Weighted scoring and LLM-based decision making
  • Fallback mechanism for system stability
  • Explainable decision output
Concurrent Scheduling Engine

DAG Scheduling Engine

Dependency-aware scheduling that maximizes parallel execution.

Topological Sorting

Analyze dependencies, build an execution topology, and determine the right execution order.

Parallel Execution

Run independent tasks at the same time to shorten total completion time.

Dynamic Adjustment

Adjust concurrency at runtime based on resources and task characteristics.

Exception Isolation

Isolate task-level exceptions so one failure does not interrupt the full workflow.

2-5×
Execution Efficiency
<100ms
Response Speed
95%+
Success Rate
7×24
24/7 Service
Continuous Learning System

Experience Memory

Store successful workflows as vectors, retrieve similar cases, and reuse proven strategies over time.

Vector Storage

Convert successful execution trajectories into vector representations and store them in the experience memory.

Smart retrieval

Retrieve similar historical workflows for new tasks and match them to proven strategies.

Strategy Reuse

Reuse historical successful strategies, reducing planning time by 40-60%

Continuous Evolution

Use failed cases as counterexamples so the system can avoid repeating mistakes.

Dynamic Interface Technology

A2UI Dynamic Interface Engine

The A2UI engine lets AI generate interactive interface components such as charts, forms, and buttons instead of returning text only. TeGo implemented a complete React renderer ahead of Google’s official release.

  • Full implementation of A2UI protocol specification
  • 15+ standard components (Text, Card, Chart, Form, etc.)
  • Independent style system, zero Tailwind dependency
  • Declarative data format, no code injection risk
  • Support for streaming rendering and real-time updates
Interactive Output Example
User: Analyze sales data and find the best performing regions
Sales Analysis Report
¥892K
华东
¥756K
华南
¥698K
华北
Skill Definition Example
# SKILL.md
---
name: contract-reviewer
description: Professional legal contract review and risk analysis
version: 1.0.0
compatibility:
  - tego-os
  - claude
  - cursor
---
Skills Extension System

Agent Skills Ecosystem

Extend AI capabilities like installing apps. Standardized skill definition specification, compatible with Claude Agent Skills open standard, cross-platform usable. Supports Cowork collaboration mode for dynamic multi-skill combination without predefined workflows.

  • Standardized SKILL.md format definition
  • Cross-platform compatible (TeGo, Claude, Cursor)
  • Cowork collaboration mode, intelligent multi-skill orchestration
  • Secure sandbox isolated execution
  • Progressive loading, optimized Token consumption

MCP Development Framework

Decorator-driven development model that greatly simplifies agent tool development

Purpose-built MCP framework

Define agent tools quickly with @Tool decorators and structured metadata, reducing boilerplate and improving delivery speed.

Standardized Tool Registration and Validation

Tool capabilities are registered with structured metadata, automatic parameter validation, and type checking.

Define agent services in a few lines

A minimal development model helps teams expose callable AI tools quickly and iterate safely.

Multi-Agent Collaboration with LAMP

The patented LAMP framework uses Think-Speak-Decide reasoning, semantic agent communication, decision fusion, DAG scheduling, and dynamic replanning to improve task success rates by 15-25% and execution speed by 2-5x.

Decorator-Driven Development

TypeScript

The project uses @Tool decorators to define tools, providing clear metadata declarations:

@Tool({
  name: 'saveMedicalRecord',
  description: 'Save medical record to database/保存电子病历',
  parameters: [
    {
      name: 'tenantId',
      type: 'string',
      description: 'The tenant id/租户id',
      required: true
    }
    // ... 更多参数
  ]
})
export class MedicalRecordTool extends BaseTool {
  // 工具实现
}
Type Safety

Complete TypeScript type support

Auto Validation

Automatic parameter validation and type checking

Core Advantages Summary

The decorator-driven model lets developers focus on business logic while TeGo handles registration, validation, and operational integration. This improves delivery speed, maintainability, and enterprise readiness.