Wenxin Kuai Ma Baidu Comate

What is Wenxin Kuai Ma Comate?

Baidu Comate is an AI programming assistant developed by Baidu, intended for individual developers and corporate R&D teams. It can be installed as a plugin in development environments such as VS Code, JetBrains, Visual Studio, and Xcode; in addition, there is also a standalone Comate AI IDE along with a command-line interface.

The products include code completion, code Q&A, explanation and refactoring, unit testing, document generation, as well as programming agents that enable an executable terminal and code modification.

Comate is based on Baidu’s self-developed Wenxin code large model; it can also utilize models such as DeepSeek, ChatGLM, and Kimi in various scenarios. The new version, Comate 4.0, focuses on evolving from a \"code completion assistant\" into an Agent development environment that supports the breakdown of complex tasks, collaboration among multiple sub-agents, as well as the use of Skills, Rules, Memory, MCP, and code repositories.

Supported development environments and languages

The official documentation provides links for installing or getting started with Comate AI IDE, VS Code, JetBrains IDE, Visual Studio, Xcode, and the CLI. The timing of feature releases may vary across different clients, so it is important to keep the plugin versions up to date.

Comate supports over 30 programming languages; popular ones such as Java, Python, JavaScript, TypeScript, C, C++, Go, PHP, Rust, as well as front-end frameworks, all benefit from completion features and query assistance.

Language support does not mean that it will work equally well with every framework, compiler, and project scale. Before using it, it is necessary to verify whether the plugin can correctly identify the workspace, build system, dependencies, and language services; the results generated must pass through the project’s own compilation, static analysis, and testing processes.

Code completion

Comate provides single-line or multi-line suggestions based on the code before and after the cursor, as well as the context of the current file, adjacent files, and the codebase. Developers can see the gray-colored suggestions after a brief pause, and they can insert the suggested code once it has been confirmed.

Intent generation functions or code blocks can also be described through annotations.

Code completion is useful for template code, data structures, interface calls, and repetitive logic, but it may introduce non-existent APIs, outdated approaches, security vulnerabilities, or licensing issues. It should be reviewed line by line before being adopted, and must not be used blindly in contexts related to keys, encryption, permissions, and financial operations.

Intelligent Q&A and knowledge enhancement

The Q&A feature in the sidebar allows users to choose the current file, selected code, or the entire codebase as the scope of knowledge, thereby enabling explanations of logic, identification of errors, generation of solutions, and answering of project-related questions. Users can also upload the company’s coding standards or technical documentation as custom knowledge sources, so that the responses are in line with the team’s rules.

The custom-developed context engine searches through the code repository and related documentation, thereby reducing errors that arise from relying solely on the current file. Even in large repositories, dynamic configurations, generated code, runtime dependencies, and external services may still be overlooked; when asking questions, it is necessary to specify the entry point, modules, steps to reproduce the issue, and the desired behavior.

Inter-line commands

Comate integrates common editing functions into the code context, including code explanation, comment generation, log addition, performance suggestions, function splitting, and unit test creation. Developers can execute these functions by simply selecting the relevant code, without having to write prompts from scratch each time.

• Explain the code:Describe the function’s responsibilities, data flow, dependencies, and boundary conditions;
• Generate comments:Generate a Docstring for the function, along with descriptions of its parameters and return values;
• Generate unit tests:Create test cases based on function and project testing frameworks;
• Add log:Add log statements suitable for troubleshooting to the critical path;
• Function splitting:Restructure overly complex logic into smaller units;
• Tuning suggestions:Analyze readability, performance, and potential issues.

Automatic unit testing may only verify the current implementation rather than the business requirements. It is necessary to take into account scenarios related to exceptions, concurrency, permissions, timing, null values, and real-world integration situations; a test passing result cannot replace a thorough review.

Programming Agent

Comate Agent is capable of carrying out complex coding tasks on its own, reading and editing files, searching through code repositories, executing terminal commands, performing web searches, and invoking sub-agents. The developers describe an Agent as consisting of system prompts, process reminders, tools, and user messages, and they adjust these prompts and tools according to the different models used.

Developers can describe their goals in the plugin or AI IDE sidebar, allowing the Agent to analyze the repository, create plans, modify multiple files, run tests, and make further corrections based on the results. Checkpoints are used to save important states, while parallel sessions enable different tasks to progress simultaneously.

Modes such as Ask and Plan change the way an Agent operates through process reminders.

Agents have the capability to modify terminals and files, which increases the risks associated with tasks such as migration, deletion, deployment, installation of dependencies, and access to credentials. It is necessary to examine plans and any discrepancies, restrict the permissions of commands, run operations in separate branches or isolated environments, and maintain version control states that can be restored.

Multi-agent and Subagent

Comate allows complex tasks to be divided among multiple sub-agents with different responsibilities. Explore Subagent is dedicated to searching through code libraries and locating relevant implementations, while other sub-agents can handle analysis, coding, testing, or documentation respectively.

This helps to prevent the main Agent’s context from being filled with a large number of search results.

Multiple intelligences improve parallel efficiency, but they can also lead to repeated modifications, inconsistent interfaces, and context conflicts. The main task is to define clear file boundaries, acceptance criteria, and the order of merging, with unified testing and manual review still being necessary in the end.

Skills, Rules, and Memory

Skills encapsulates the team’s common processes, domain knowledge, and tool usage methods into reusable capabilities, such as code review, database migration, release checks, and framework standards. Rules are used to consistently enforce coding styles, directory structures, and development requirements.

Memory can store long-term information related to projects or users, thereby reducing the need for repeated explanations.

These contexts have a direct impact on the generated results. Too many rules, or rules that conflict with each other, can reduce the quality of the results.

Passwords, access tokens, and customer privacy information should not be stored in Memory.

The team should incorporate the key specifications into code reviews, testing, and CI, rather than relying solely on prompts.

MCP support

Comate 4.0 significantly enhances MCP: it allows connections to external tools and data sources through configuration files, and provides easy-start guides, examples, and debugging documentation. MCP enables Agents to access databases, internal services, browsers, or other development tools, thereby expanding their capabilities beyond those of simple files and terminals.

The permissions held by the MCP server are transferred to the Agent. It is necessary to use accounts with minimal privileges, limit the tools that can be invoked, isolate the production environment, and audit requests.

MCP packages and configurations of unknown origin may allow the execution of any command; it is necessary to examine the code, the publisher, and the dependencies before installing them.

Comate AI IDE

The standalone AI IDE integrates code editing, Agents, intelligent Q&A, and a terminal into a single application, and offers features such as real-time preview in the frontend and element-level debugging. The Spec mode emphasizes establishing specifications before writing code.

Mission Mode is used to make progress step by step based on tasks.

It is suitable for users who wish to enjoy a more comprehensive Agent experience; existing IDE users can continue to install plugins.

An AI IDE is not a cloud-based platform for automatic hosting; the project still requires the setup of a local runtime environment, dependencies, a database, and deployment procedures.

A normal preview does not mean that there are no compatibility issues in a real browser, on mobile devices, or in a production environment.

Personal version price

The following are the SaaS pricing details as updated in Baidu Smart Cloud’s official documentation on July 15, 2026. The official website may offer promotions for new users or time-limited deals; the actual amount to be paid shall be based on the settlement page.

Smart completion for all individual plans is available free of charge; the main differences lie in the tokens required for the advanced Agent models and the Auto-Free rate.

The current promotions page on the official website may indicate rates such as 55 yuan per month for the individual professional version. Vouchers are updated in line with the subscription period and cannot be withdrawn in advance;

It depends on the account rules to determine whether unused cycle quotas are carried forward.

Enterprise version price

• Enterprise Professional Edition:150 yuan per person per month, 450 yuan per person per quarter, 1500 yuan per person per year; includes 72 yuan per person per month in credits for making requests to intelligent agents, with unlimited usage in Auto-Free mode; minimum purchase amount: 1 person.
• Enterprise Flagship Version:358 yuan per person per month, 1074 yuan per person per quarter, 3998 yuan per person per year; this package includes 283 yuan worth of credits for making requests to intelligent agents per person per month, with unlimited usage of Auto-Free service. The purchase is available starting from 1 person.
• Enterprise Exclusive Version:2,500 yuan per person per year, with a minimum order of 100 people; it includes 72 yuan per person per month in request credits, and private deployment of models is available.
• Private version:Book a demonstration based on the company’s size, deployment, customization, and security requirements, and an individual quote will be provided.

The enterprise version also offers organization and user management, licensing options, corporate knowledge management, data statistics, an open platform, Deepwiki, AgentHub, unified authentication, performance metrics, as well as options for SaaS, hybrid cloud, and on-premises deployment.

Agent request packets and billing

Once the quota for advanced models included in the subscription is used up, it is still possible to continue using Auto-Free; alternatively, one can purchase separate agent request packages priced at 18 yuan, 33 yuan, or 150 yuan. These packages take effect immediately upon purchase and remain valid permanently. The cost of a request is calculated by multiplying the number of tokens consumed by the real-time price per model, so the expense for each request depends on the model, the context, the tools used, and the length of the output.

Companies can opt for pay-as-you-go pricing, with charges calculated on an hourly basis; there may be delays in the issuance of bills, and costs can continue to be incurred until the deduction is completed.

Administrators should set budgets, member permissions, and billing alerts to prevent unexpected costs resulting from agent loops or excessively long contexts.

Data security

According to Baidu’s official security documentation, Comate has passed the information security certifications issued by the China Academy of Information and Communications Technology, as well as SOC 3, SOC 2 Type 2, and SOC 1 Type 2 certifications. It does not store users’ source code nor uses their data for training purposes. For highly sensitive environments, it is recommended to use the private version; certification reports can be obtained from the official sources.

Public cloud plugins still need to transmit over the network the context required for inference. Enterprises should evaluate source code classification, third-party dependencies, data locations, logging, account permissions, and confidentiality agreements.

Keys, production credentials, and personal data should not be included in the code or prompts.

API, Open Center, and Open Source Status

The enterprise console offers an open platform; Comate also supports MCP and custom agent extensions, but personal SaaS does not provide a universal code model API. When system integration, enterprise accounts, a knowledge base, or usage-related interfaces are required, access should be obtained based on the permissions provided by the open platform and the terms of the enterprise contract.

The Comate plugin, AI IDE, context engine, and programming Agent are not open-source products, and there is no official complete self-deployment repository available for them. Baidu’s other open-source models, SDKs, and examples on GitHub cannot be considered as part of the Comate platform’s source code.

Enterprise privatization falls under commercial deployment rather than open-source licensing.

Comate Quick Code Usage Guide

Complete a basic task.

1. Clarify the coding tasks to be carried out in Wenxin Kuai Ma Comate, the scope of the repository, and the acceptance criteria.
2. Connect to or import the test project, and first back up the current branch;
3. First, generate the plan using the supported development environment and language, and then verify the modified files;
4. Use code completion to make small changes;
5. Run tests, static checks, and builds; unverified code is not accepted directly.
6. Manually check permissions, keys, dependencies, and handle exceptions before merging;

Create reusable professional workflows

1. Select a low-risk, real-world project as a template;
2. The fixed development environment supported, as well as the order of use for code completion, intelligent question-answering, and knowledge enhancement;
3. Record the environment, model, prompts, and failure conditions;
4. Set up manual approval for write, deploy, and delete actions;
5. Compare speed, cost, test pass rate, and the amount of rework;
6. Expand to a team or production environment only after stability has been verified;

Which users are it suitable for

• Chinese developers who wish to use code completion and basic Agent features for free;
• Individual programmers who need assistance with Chinese-language code questions, unit testing, comments, and code refactoring;
• Teams that wish to use Agents in VS Code, JetBrains, Visual Studio, or standalone AI IDEs;
• Advanced users who need Skills, Rules, Memory, and MCP to establish project standards and toolchains;
• A R&D team that requires organizational management, corporate knowledge, performance statistics, and access control;
• Chinese companies with high requirements regarding the security of their source code, and those that need hybrid cloud or on-premises deployment.

Usage restrictions and precautions

• Code generated by AI may compile successfully but contain errors in the business logic, as well as outdated dependencies, vulnerabilities, and incompatible licenses.
• Before accepting the modifications, the differences should be reviewed, and formatting, static analysis, unit testing, integration testing, and security scans should be carried out;
• The Agent can perform terminal and batch editing, and it is necessary to control deletion, migration, deployment, and external access.
• The production database and cloud accounts are configured with read-only or minimal permissions, and sensitive operations require manual approval;
• Parallel modification by multiple agents should also prevent conflicts and redundant work.
• The claimed improvements in efficiency and the proportion of AI code are part of the product statistics, and they do not mean that every project will achieve the same results.
• The actual benefits depend on the quality of the codebase, test coverage, task type, and the team’s review process.

Frequently Asked Questions

Is Wenxin Kuai Ma free?

The personal standard version is free, and intelligent completion services for all personal tiers are also free; the standard version comes with a 10-yuan voucher for making Agent requests, and once the quota for advanced models is used up, users can opt for Auto-Free or purchase request packages.

Which IDEs does Wenxin Quick Code support?

It provides the Comate AI IDE, as well as access to VS Code, JetBrains, Visual Studio, Xcode, and the CLI; the specific features are subject to those available in the latest versions of each client.

How much is the Personal Professional version?

100 yuan per month, 270 yuan per quarter, or 1000 yuan per year, including 55 yuan worth of agent request credits per month; the current promotional prices may vary.

How much is the personal flagship version?

299 yuan per month, 889 yuan per quarter, or 3499 yuan per year; this package includes 199 yuan worth of agent request credits per month as well as a higher Auto-Free data rate.

Can Comate be deployed in a private environment?

Yes. The enterprise edition offers model privatization; there are also complete privatization solutions with pricing based on specific requirements, and it supports SaaS as well as hybrid clouds.

Is Wenxin Kuai Ma trained using source code?

According to the official security documentation, the source code is not stored, nor are user data used for training purposes; for highly sensitive projects, it is still recommended to conduct an assessment and adopt a private deployment solution.

Is Wenxin KuaiMa open source?

It is not open source. Commercial, proprietary deployment does not equate to open-source code; moreover, Baidu’s other projects on GitHub do not represent the complete source code of Comate.