Digital Employee Overview

A Digital Employee is Gendial's enterprise-grade AI Agent form. Its design goal is not to create a "personal AI assistant", but to build a digital workforce that can be managed, audited, and deployed at scale by enterprises.

From "AI Toy" to "Digital Asset"

Most current AI Agent frameworks target individual developers, focusing on "can AI operate a computer". These solutions perform well in single-user experimental scenarios but expose three fundamental problems in enterprise environments:

First, memory is private. Each agent's memory is stored in local files, and one agent can only serve one user. A financial enterprise with 200 customer service agents—must it deploy 200 agent instances?

Second, security is absent. Agents run with local permissions on the developer's machine, able to read SSH keys and modify code. If an agent's logic goes out of control or is maliciously injected, the consequences are catastrophic.

Third, management is decentralized. Skills are installed and configured by users themselves, without a unified review mechanism. There is no way to know which agent is calling which tool or how many resources are being consumed.

The Gendial Digital Employee is designed to solve these three problems, transforming AI Agents from "scripts on a developer's computer" into "governable assets in enterprise IT systems".

Core Architecture: Brain in Cloud, Body at Edge

Digital Employees adopt a cloud-edge separation architecture:

Brain in Cloud: The agent's core reasoning, decision logic, and workflow engine run server-side. All conversation chains-of-thought (CoT), tool call decisions, and memory retrieval complete in a unified server-side environment. Benefits include:

• Reasoning processes are auditable, with complete context records for every decision
• Model calls are quota-controlled, preventing "bill explosions" from agent logic loops
• Core logic is centrally upgraded without requiring manual updates on each client

Body at Edge: Tools and skills execute in the client's local environment, directly calling local resources. Benefits include:

• Uncompromised functionality: local tools can access the file system and call local APIs without cloud sandbox restrictions
• Controlled security: clients do not expose network ports. Tool execution follows a controlled model of server instruction dispatch, client pull execution, and result return, significantly narrowing the internal network penetration attack surface

The core philosophy of this architecture: critical decisions must be centralized and controllable; execution can be distributed.

Identity Impersonation: Agent Acts With Your Permissions

This is the Digital Employee's most critical "enterprise-grade" capability. When a Digital Employee is assigned to a user, the Agent carries that user's identity credentials when calling any resource (knowledge bases, tools, APIs).

This means: even if the Agent's logic says "read all financial reports", when assigned to an employee with only East China region permissions, the Agent can only read East China data. Permission control happens not at the Agent level, but at the identity level—the Agent strictly inherits its assigned user's permission boundaries.

From a security perspective, this is a fundamental defense: the Agent's behavioral scope is artificially constrained within the user's authority, mechanically eliminating unauthorized access.

Memory Isolation: Everyone Has Their Own "Digital Colleague"

Traditional Agent memory is global—all users share the same conversation records. Digital Employees maintain completely independent memory spaces for each user:

• Employee A tells the Agent "I'm traveling on Friday"—this information exists only in A's memory
• When Employee B converses with the same Agent, the Agent knows nothing about A

This is achieved through vector database scalar filtering and partition key mechanisms. Each Agent-User pair has an independent memory partition, supporting millisecond-level retrieval under high concurrency. One Digital Employee can simultaneously serve hundreds of users, with each person's interaction experience being isolated and personalized.

Skill Governance: Standardized Capabilities, Centralized Review

In personal Agent solutions, users can freely install any script as a "skill". In enterprise scenarios, this is disastrous—there is no way to review skill sources or verify whether skills contain malicious code.

Digital Employee skill management adopts a platform-centralized model:

• All skills are uploaded, reviewed, and published at the platform level
• Administrators can configure fine-grained permissions by user role—which departments can use which skills
• Digital Employees can only attach skills that have passed review, eliminating the possibility of "malicious" plugins

Real-Time Steering: Adjust Agent Behavior Mid-Execution

Unlike traditional chatbots, you don't need to wait for a Digital Employee to finish before continuing to type. While the agent is responding, you can keep typing to adjust its behavior:

• The agent is generating an analysis report — type "analyze Q2 data instead" — the agent adjusts direction after the current output completes
• The agent is calling tools — type "reply in English" — the agent switches language and continues
• The agent is going off track — type a correction — the agent adopts it in its next reasoning step

This "Steering" mechanism turns passive waiting into active collaboration — just like working with a human colleague, you can give feedback as you observe.

Full Chain Audit: Visible Trust

Every conversation, every tool call, every step of chain-of-thought reasoning by a Digital Employee is fully recorded in server-side databases. This is the foundation of enterprise compliance:

• When a customer complains "the AI said something inappropriate", every step of the conversation can be precisely traced
• When investigating "why did the Agent call this API", the complete decision chain is viewable
• All logs are centralized server-side, without needing to collect from individual clients

Digital Employee vs. Standard Agent

Dimension	Standard Agent	Digital Employee
Memory	Single conversation context	User-level long-term memory, persistent across sessions
Identity	Anonymous conversation	Identity impersonation, inherits user permissions
Skills	Embedded in workflow	Independent skill packages, centralized review
Tools	Called via workflow nodes	Independent toggle control, supports client-side execution
Delegate	Not supported	Can delegate tasks to other DEs or Workflows (Coordinator-Worker pattern)
Retrieval	Single retrieval with preset keywords at workflow node	Autonomous keyword selection and multi-round retrieval, supports multi-hop reasoning
Deployment	Published as chat service	Deployable to physical/virtual machine workstations
Audit	Basic conversation records	Full-chain audit: CoT + tool calls + identity

Use Cases

Digital Employees are particularly suited for these enterprise scenarios:

• Multi-user shared expert assistant: A single finance Digital Employee assigned to the entire finance team, where each person sees different data but uses consistent rules
• Audit-required customer service scenarios: All conversations are traceable, meeting compliance requirements in finance, healthcare, and other industries
• Cross-department workflow automation: Digital Employees carry user identity to pass and process data between different systems