DingTalk AI Assistant

What is DingTalk AI Assistant?

DingTalk AI Assistant is a tool provided by DingTalk for individuals, organizations, and developers, enabling the creation and operation of enterprise bots. Users can define the role, knowledge, skills, and workflows of these assistants, so that they can answer questions or carry out tasks in chats, groups, and various business scenarios.

It is not an independent general chat website; rather, it is part of the DingTalk AI PaaS and collaborative work ecosystem. Companies can connect their address books, documents, calendars, approval processes, forms, as well as any custom applications they have developed, to this assistant, depending on the permissions set.

Main functions of DingTalk AI Assistant

• Character setup:Define the assistant’s role, responsibilities, response style, boundaries, and default responses.
• Multi-model selection:Switch the platform model based on the specific use case; enterprise solutions can also integrate custom models.
• Knowledge quiz:Train assistants to learn corporate documents and knowledge, so that they can answer employees’ questions based on this information.
• Skill invocation:Combines DingTalk’s official capabilities, third-party capabilities, and enterprise-customized interfaces.
• Workflow:Use visual nodes to orchestrate queries, judgments, notifications, and business actions in a fixed sequence.
• Personified operation:It is possible to operate application pages that lack exposed interfaces using RPA.
• Personalized memories:Generate more targeted responses based on user, organizational, or business data.
• Perception trigger:Tasks are executed automatically at a specified time or when group messages or business events are detected.
• Planning ability:Break down complex goals into multiple steps, and select the appropriate tools to carry them out according to a strategy.
• Safety control:Configure sensitive words, standard phrases, permissions, and rules for handling content risks.
• Multi-channel distribution:It can be published to individuals, organization members, or eligible external channels.
• Assistant API：Internal enterprise applications can create and manage assistants through interfaces.

Knowledge, skills, and workflows

Corporate knowledge Q&A

Knowledge capabilities enable assistants to retrieve regulations, product information, training materials, and service manuals. High-quality Q&A responses depend on clear titles, stable versions, defined business terminology, and appropriate access rights.

• Organize the data by topic to prevent a single file from containing multiple unrelated tasks;
• Specify the target audience, time range, and version status in the document;
• Maintain synonyms, abbreviations, and internal company terms;
• The expiration system is in place to prevent both old and new content from being referenced at the same time;
• Set minimum visible ranges for salaries, contracts, and customer information.

Workflows and business actions

Workflows are suitable for tasks that have clear steps, a fixed sequence, and require consistent execution; for example, generating a summary after querying data and then sending interactive cards. They can be connected to message notifications, documents, multidimensional tables, approval processes, as well as custom enterprise functionalities.

When high-risk actions such as writing, approval, payment, and deletion are involved, a manual confirmation step should be included. It is not possible to rely solely on natural language processing to make decisions regarding irreversible business operations.

Perception-triggered planning

Perception triggers enable assistants to start tasks automatically based on time, group messages, or business events, while planning capabilities are used to break down more complex goals. Both are suitable for daily report reminders, customer follow-ups, anomaly notifications, and cross-application collaboration.

Before automatic execution, it is necessary to define the trigger conditions, frequency, target group, and procedures for handling failures. Otherwise, this may lead to repeated notifications, abnormal consumption of resources, or incorrect business operations.

Distribution scope and use cases

Organizational assistants are included in the address book system without taking up slots in the regular address book; administrators can adjust their visibility or disable them. Once an assistant is deleted, its historical records and settings may not be recoverable, so it is necessary to export the relevant data before performing such actions.

Common enterprise applications

• Personnel Assistant:Answers questions regarding employment, attendance, benefits, holidays, and policies.
• Administrative Assistant:Search for meeting rooms, office supplies, visitors, and travel procedures.
• Sales Assistant:Organize customer information, generate follow-up summaries, and trigger reminders.
• Project Assistant:Summarize progress, risks, tasks, and key decisions from group chats.
• Customer Service Assistant:Answer questions based on product knowledge and forward complex tickets.
• Data Assistant:Ask questions and get answers regarding indicators by using charts, tables, or databases.
• R&D Assistant:Query specifications, generate requirement summaries, and alert for testing nodes.
• Manager’s Assistant:Aggregate operational data, approval information, and organizational updates.

Comparison of prices and versions for DingTalk AI Assistant

There is no fixed price for DingTalk AI Assistant that applies to all companies; the cost is determined by the version of DingTalk, the AI processing power, the additional features offered, and the implementation services. The following are the main purchase options that can be confirmed on the official website.

The DingTalk 365 membership is a comprehensive set of benefits designed for individuals, and it is not equivalent to the enterprise AI services. Prices may vary depending on the account type, campaign, and organizational version; therefore, the prices listed on the official store or the current billing page should be taken as reference.

How does AI calculate particles?

• AI computing units are the unified measurement standard used by enterprises to utilize the DingTalk productivity platform and large-scale models.
• Training, inference, dialogue, and content generation of AI assistants may consume computing resources;
• The cost per use varies for different models, skills, and external services.
• The free monthly quota, once refreshed, is not carried over to the following month.
• The benefits for individual members are calculated separately from those for corporate accounts;
• Administrators can view usage amounts and consumption details on the enterprise AI computing resources management page.

How to choose a version

TDDing AI Assistant Creation Tutorial

Create an AI assistant within the organization

1. Upgrade the DingTalk desktop and mobile versions to version 7.6 or higher.
2. Go to the AI assistant list and choose between starting from scratch, having it generated automatically by AI, or creating it using a template.
3. Fill in the name, profile picture, introduction, character settings, welcome message, and suggested questions.
4. Select a large model suitable for the business, and set fallback responses for cases when the parameters are outside the acceptable range.
5. Add knowledge, skills, workflows, personalization, perception triggers, or planning capabilities.
6. Test normal cases, edge cases, and abnormal inputs in the preview/debugging area.
7. Choose to make it available only for yourself or organization members, submit it for approval, and then publish it.

Create a knowledge Q&A assistant

1. Identify the target audience and the scope of the issues, such as HR systems or product support.
2. Organize the latest, valid knowledge documents with clear permissions.
3. The role requirement is to provide answers based on knowledge; if insufficient information is available, this should be clearly stated.
4. Upload knowledge and add terms, synonyms, and common questions.
5. Use real business scenarios to test referencing, accuracy, and permission isolation.
6. Establish responsible persons and update schedules to replace expired content in a timely manner.

Create automated workflows

1. Break the task down into steps of triggering, querying, judging, generating, executing, and notifying.
2. Choose the DingTalk collaboration feature or custom enterprise APIs as nodes.
3. Set input parameters, exception handling rules, timeout settings, and retry rules.
4. Manual confirmation is required for approval, writing, and deletion actions.
5. Verify permissions, repeated triggering, and particle consumption in the test organization.
6. After deployment, continuously check the operation logs and failure records.

Developers and API capabilities

The DingTalk Open Platform provides documentation on how to create AI assistants, as well as information regarding configuration options and the Assistant API. The creation of such interfaces is currently available for use within corporate applications; administrators must grant the necessary permissions and use corporate access credentials.

• It is possible to set the name, description, role instructions, avatar, and welcome message through the interface;
• Returns identifiers such as AssistantID, which are used for subsequent management and invocation;
• Internal enterprise applications require an application to obtain management permissions for AI assistants.
• Some advanced interfaces are available only to the premium or customized versions of the productivity platform.
• The old version of the AI message response interface has been moved to the historical documentation and should not be used in new projects.
• High-concurrency calls require handling rate limiting, retries, circuit breaking, and log auditing.

Product advantages

• It provides deep integration with DingTalk’s organizational tools, group chats, documents, approval processes, and business applications.
• Non-developers can also create assistants using templates and visual workflows;
• Supports combinations of knowledge, skills, planning, memory, and event triggers;
• Assistants can distribute content according to rules across personal, organizational, and external channels;
• Enterprise administrators can control the visibility range, disable features, and manage resource consumption.
• It offers APIs, access to proprietary models, MCP, as well as options for custom development;
• The exclusive version and the hybrid version are suitable for those with higher security and deployment requirements.

Usage restrictions and precautions

• AI responses may contain factual errors, omissions, or misinterpretations of corporate policies.
• When there are conflicting versions in the knowledge base, the assistant may refer to outdated information.
• Different models have varying costs in terms of resource consumption, and budgets need to be set for automated tasks.
• The use of organizational members usually requires administrator approval and configuration of visibility settings.
• Some APIs, model integrations, and advanced features are subject to the limitations associated with the particular version.
• The older version of the development interface may stop supporting new features, so it is necessary to check the status of the documentation before integrating it.
• Deleting an assistant may result in the loss of historical records; it is necessary to back up the key configurations first.

Recommendations for data security and permissions

The range of data to which an enterprise AI assistant can access depends on its knowledge, skills, authorization for specific applications, and the user’s identity. Security settings should follow the principle of least privilege; it cannot be assumed that the assistant has access to all data just because it operates within an organization.

• Set the visibility scope of knowledge by department, role, and data level;
• Sensitive fields should not be directly included in prompts or public knowledge bases;
• Special access rights are applied to financial, payroll, HR, and customer data;
• For high-risk execution actions, approval and manual confirmation are required;
• Regularly audit skills, interface credentials, workflows, and external distribution channels;
• Manual review, logs, and responsible persons are retained for the generated content.

Open-source instructions

The DingTalk AI Assistant platform, model services, and enterprise operation systems are not open-source products. The agent frameworks or development projects made available by Alibaba on GitHub do not constitute the source code for DingTalk AI Assistant, and it is not possible to deploy similar services using those materials.

Developers can use the official APIs and server-side SDKs for integration, but they must comply with the DingTalk platform’s rules regarding permissions, billing, and data usage. When choosing community projects, it is also necessary to verify the maintainers and the licenses associated with them.

Frequently Asked Questions

Is DingTalk AI Assistant free?

The basic functions of DingTalk as well as some of its AI assistant features can be tried out for free; organizations also receive a certain amount of free AI computing power each month. Advanced versions, additional computing power, APIs, and customized features may come at a cost.

How many free AI computing units does an organization have per month?

According to the official announcement, all organizations receive 10,000 AI computing units per month; this quota is refreshed on a monthly basis and does not accumulate over time. The policy may change, so it is necessary to refer to the enterprise’s computing units management page for the most up-to-date information.

How much is DingTalk Professional Edition?

The official help center indicates a cost of 9,800 yuan per year; this is the total cost for DingTalk Professional Edition, and it does not represent the individual price of each AI assistant. Additional usage of AI services as well as premium services require separate verification.

Can an AI assistant be used to handle approvals and notifications?

Some of DingTalk’s collaboration features can be utilized by integrating skills with workflows. When it comes to approval, data entry, payment processing, and deletion, permission controls and manual verification must be implemented.

Is DingTalk AI Assistant open-source?

No, the core platform falls under the category of commercial services. The official platform provides APIs and SDKs for integration, but this does not mean that the model codes or the complete source code of the product are made available.