CodeBuddy

CodeBuddy is Tencent Cloud's AI coding assistant. It ships as an IDE, as plugins for editors such as VS Code and JetBrains IDEs, and as a command-line tool. Tencent Cloud's billing docs list Personal Free ($0, 100 credits a month), Pro ($10 a month, 2,000 credits) and Team ($40 per seat a month). We checked the facts below on 31 October 2026 from the vendor's own pages.

At a glance

What CodeBuddy is

Tencent Cloud's product overview calls CodeBuddy "an AI-assisted programming tool launched by Tencent Cloud." The same page splits it into three forms and names an audience for each.

That split is the first thing to understand, because "CodeBuddy" is three products with one name. The IDE is the fullest experience. The plugin brings chat and completion to the editor you already use. CodeBuddy Code is the terminal agent, in the same family as Claude Code and Qwen Code.

There is also a mainland China site. The page at copilot.tencent.com is titled in Chinese as "Tencent Cloud Code Assistant CodeBuddy" and describes the tool as built on the Hunyuan code model. The international pages we cite here do not make that claim. We cover the international edition, and pricing on the China site may differ, so check which site you are on before you buy.

Among its peers, CodeBuddy sits near Trae, another AI IDE from a large Chinese tech company, and near Cursor, Windsurf, Kiro and GitHub Copilot.

Who it suits

Tencent's own audience list is broad on purpose. The IDE pitch includes product managers and designers, not only programmers. The overview lists turning a natural-language description into a product requirements document, generating interactive prototypes from text, and producing high-fidelity mockups from hand-drawn sketches.

That fits people who work across the whole path from idea to a running page, and who want one tool for it. It also fits teams already in the Tencent ecosystem, since the docs name Tencent CloudBase, CloudStudio and EdgeOne Pages as built-in targets, plus enterprise integrations for Feishu and WeCom in the plugin docs.

The plugin suits developers who will not leave JetBrains or VS Code. The list of supported hosts is wide. It includes Xcode and WeChat Developer Tools.

CodeBuddy Code suits terminal users. Its docs section has pages on headless use, hooks, GitLab CI/CD, a daemon, checkpoints and permission modes, so it targets scripted and CI use as well as chat.

It fits less well if you want a published, plan-by-plan list of models, or per-call credit costs. See limits below.

Pricing

The numbers here come from Tencent Cloud's billing pages for CodeBuddy. We could not load the live pricing page at codebuddy.ai/pricing because it renders in the browser, so check it on the day you buy.

Personal and Team plans

Team also includes an administrator console, unified billing and support for both IDE and CLI use.

Trials and starting credits

• New users get 250 credits on first login, valid for 14 days.
• A Pro 7-day trial is available to anyone who has never bought Pro or claimed a trial, after binding a credit card and turning on auto-renewal. It grants 500 credits valid for 7 days.
• During the free trial, completions are unlimited, 99 automated tasks are allowed and all models can be selected.

Credit rules

• Base and bonus credits are issued monthly and do not roll over. Bonus credits last one month.
• Credits are used by expiry date. When two items expire together, base credits go first.
• Once Pro or the Pro trial starts, the 100 Free credits are no longer issued separately.
• The daily activity bonus needs at least one valid AI conversation or task in the desktop client. Web client use does not count. Credits usually arrive the next day, and the page says the event may be changed or ended.
• Credits are the shared unit for AI actions. Cost depends on the model and the task, and advanced models charge more per token.
• The docs do not list a credit cost per model or per task. You can see your balance under Personal Homepage, then Usage.

Top-up packs

Top-ups need an active Pro or Team subscription. They are one-time, non-refundable and can be bought repeatedly.

Payment and plan changes

Payment is by credit or debit card, WeChat or prepaid card. Only one plan can be active at a time. Direct upgrades are not supported yet, so an existing subscriber has to cancel within the refund policy and buy the new plan. Cancelling auto-renewal on Team keeps the plan active to the end of the cycle.

People who subscribed before 7 August 2026 keep their current quota for the cycle, then move to 2,000 credits a month. Those on auto-renewal from before that date keep $9.95 a month until they turn it off.

What it does

Interaction modes

The IDE has four ways to work, and you can switch at any time.

• Ask answers technical questions and explains ideas, with no direct code changes.
• Craft generates and edits code within the current context, suited to a single function or a small refactor.
• Plan is the most automated mode. The AI writes a task list, then reads, edits and checks the environment after you confirm.
• Custom agents are ones you create through "+ Create Agent."

The overview recommends Plan mode for work that touches many files.

Completion

The IDE offers completion that adapts to project style. Pro adds NES, Next Edit Suggestions, which predict the next step such as a following line or the end of a function body. It is on by default and can be triggered with Ctrl + Shift + Enter.

Shortcuts: on macOS, ⌘ + I opens inline chat and Ctrl + ⌘ + I opens the sidebar chat. On Windows the keys are Ctrl + I and Ctrl + Win + I.

Product, design and build

From the product overview:

• Product. Turn a description into a structured requirements document and get suggested improvements.
• Design. Generate interactive prototypes from natural language, mockups from sketches, and interfaces assembled from component libraries.
• Development. Convert Figma designs to front-end and back-end code, complete code, fix errors, generate unit tests, explain and review code, and analyse a project through @workspace and #Codebase.
• Deployment. One-click deploy to a sandbox, with public links for team review.

You reference files, folders or code with @, and you can paste screenshots or Figma links for UI work. Figma designs can also be imported through "Add from Figma." Languages named include Python, Java, Go, Rust and TypeScript.

Memory, rules and skills

The IDE docs describe memories that learn your preferences over time, rules in .codebuddy/rules for architecture and coding standards, and skills in .codebuddy/skills/ for scenario-specific instruction sets such as security audits. The subagents versus skills comparison explains how skills differ from other extension points.

The docs sitemap also lists pages for hooks, subagents, plan mode, checkpoints, history, smart commit, slash commands and a deploy flow. We list their titles because the pages render only in a browser. We did not read their contents.

Models used

Neither the product overview nor the IDE overview names the models behind CodeBuddy. The billing page says only that "more advanced models charge more credits," and the plan table says Free uses automatic scheduling while Pro can pick from all models.

A different Tencent Cloud page, for the Enterprise Token Plan under its LLM service, shows how to configure CodeBuddy. It lists these models:

Both tiers have Guangzhou and Singapore endpoints. This is a separate paid product from the CodeBuddy credit plans above, so do not read it as the list for Free, Pro or Team. It does show that CodeBuddy accepts configured third-party models.

The IDE overview also mentions custom model integration through an API, so you can connect your own model. For the one model from that list that we cover, see MiniMax M3.

Integrations and MCP

MCP

Open CodeBuddy Settings from the button at the top right of the sidebar chat, then the MCP tab. From there you can install a server from the MCP Market in one click or add your own. The market shows green on success and red on failure.

A custom server is added through Add MCP with JSON under an mcpServers object. Each server has a name and fields such as type, command, args, env and description. After installing, click Try to Run to check it, or describe a task to Craft Agent and let it call the tools.

Two gaps. The only example uses stdio, so we cannot tell you whether remote transports work. And the page does not say which plans include MCP. For background see what MCP servers are, and our Figma MCP and Supabase MCP pages for two tools that pair with the Figma and Supabase features above.

Backends and hosting

Supabase and Tencent CloudBase are built in for database work. Deployment targets are CloudStudio, EdgeOne Pages and a sandbox with shareable links. Component libraries named include TDesign, MUI and Shadcn.

Enterprise

The plugin docs list an administrator portal with pages on account management, Feishu and WeCom integration, a knowledge base, log management, model deployment management, prompt management, efficiency metrics and IP allow-listing. There is also a page for RAG knowledge bases, one for network proxy and firewall configuration, and one for custom models. We read only the titles.

Install and first run

The IDE

Download it from the CodeBuddy website. It runs on Windows, macOS and Linux. The docs also say a modern browser or compatible IDE environment is needed.

The plugin

Supported hosts from the overview:

CodeBuddy Code

npm install -g @tencent-ai/codebuddy-code
codebuddy

You need Node.js 18 or later. The npm registry shows version 2.179.0 as the latest, published 30 October 2026, with the licence given as "SEE LICENSE IN README.md." The package homepage is a repository on cnb.cool, and the README in the package is written in Chinese. It names an email address for support and says feedback is kept for 30 days and used only to improve the product, not for model training.

Privacy, security and data handling

What the first-party pages say:

• The Security and Privacy Guide says the IDE "only has the permission granted by the user," and that you are responsible for checking the security of code it suggests. It also says no protection can stop every attack.
• It advises keeping secrets in environment variables, encrypting config files with tools such as git-crypt, rotating credentials, and using apiKeyHelper to fetch keys on demand.
• Vulnerabilities should be reported through official channels, not posted publicly.
• The product overview links to a Privacy Policy and a Data Privacy and Security Agreement. We did not read their text.
• The CodeBuddy Code README says feedback data is kept for 30 days and is not used for model training. That covers feedback you submit, not your prompts or code.

So we cannot tell you what code leaves your machine, which providers receive it, or how long prompts are retained. Tencent's pages do not state these on the pages we read. Read the privacy documents before you point it at a private repository, and ask Tencent Cloud for the agreement if you work under a compliance regime.

Regional note: the docs say there are separate China and international sites. Data terms may differ between them.

Limits and gotchas

• Credits expire. Monthly credits do not roll over, and the activity bonus can end.
• Per-call costs are not published. You see usage after the fact on the Usage page.
• The Free plan is tight on automation. 3 automated tasks and 5,000 completions a month.
• No direct upgrades. You cancel and rebuy.
• Top-ups are final. One-time and non-refundable.
• Model access is plan-bound. Free uses automatic scheduling; Pro lists all models. Neither page names the models.
• Docs render in the browser. Several pages are hard to read without JavaScript, so details may be easier to confirm in the app.
• MCP is thinly documented. One stdio example, no plan statement.
• Two sites. Check whether you are on the China or international edition.
• Team pooling. Credits are shared in one pool, so one heavy user can use up the team's share.
• Web client activity does not earn the bonus.

How it compares with peers

See also Claude Code for a terminal agent comparison, and our guide to choosing a coding agent and pricing comparison.

CodeBuddy's distinguishing idea is breadth across roles: product, design and code in one tool, plus plugins for a long list of editors. The flip side is that the billing and model story is spread across several Tencent Cloud pages, and the credit system hides what each action costs.

Who should pick something else

• Teams that need a named model list per plan. CodeBuddy's pages do not give one.
• Buyers who need written data terms before a trial. Read Tencent's agreement first.
• People who want a flat price with no credits. Everything here draws on credits.
• Developers who want a single, well-documented terminal agent. Compare Claude Code and Qwen Code.
• Anyone who wants prompt-to-hosted-app output. A builder like Lovable hosts the result.

A first-week plan

• Day one: use the Free plan, check the Usage page, and try Ask and Craft on a small repo.
• Day two: try Plan mode on a change that spans files.
• Day three: add one MCP server through the market and run Try to Run.
• Day four: write a rule file for your project conventions.
• Day five: decide on Pro based on how fast the 100 credits went.

Our take

This is an editorial reading of Tencent Cloud's documentation, not a hands-on test. The facts were checked on 31 October 2026 from the vendor's own pages.

CodeBuddy has an unusually wide reach: three forms, a long list of host editors and a pitch that includes non-developers. The pricing is clear once you find the right page, and our pricing comparison lines it up with other tools.

The weaker side is clarity. Models are not named in the main docs, credit costs are not listed, and MCP gets one example. If those matter to you, test on the free plan and watch the Usage page closely.

Checklist

• Confirm whether you are on the international or China site
• Start with the free plan and watch credit use
• Check the live pricing page on codebuddy.ai the day you subscribe
• Read the Privacy Policy and Data Privacy and Security Agreement
• Keep secrets in environment variables, not in prompts
• Try Plan mode before large edits
• Add MCP servers one at a time and use Try to Run
• Remember that top-up packs cannot be refunded