What it is in one line
Tencent Cloud's AI coding assistant, covering all three forms — plugin, IDE and CLI.
The first Chinese vendor to ship all three at once, spanning everything from code
completion to "conversation-as-programming" to driving a full dev workflow from the
terminal with one sentence.
Who built it
Tencent Cloud. The slug has a typo (Tencetn should be Tencent), but the product is
unmistakable. CodeBuddy is an official Tencent Cloud product: an IDE plugin launched
in 2024 (the first Chinese coding assistant to support the MCP protocol), IDE beta in
July 2025, and in September 2025 the CLI tool CodeBuddy Code plus the international
IDE public beta — making Tencent the first Chinese vendor to cover plugin, IDE and
CLI at once.
Read: this is the classic "eat your own dog food, then sell it" big-company product.
Tencent says 90% of internal engineers use CodeBuddy, with coding time down 40%+, AI
generated over 50% of code, and overall R&D efficiency up 16%+ — numbers that
convince the company itself before being sold outside. Big platforms have unlimited
resources, an in-house model (Hunyuan) and a cloud ecosystem; the cost is that their
contest with independent products was never the same game.
What it actually does
- IDE plugin → daily completion, function calls, business logic and template
completion inside your editor; the domestic build can call DeepSeek and others
- CodeBuddy IDE → a standalone IDE, "conversation-as-programming": product
managers, designers and beginners can go from idea to deployment in natural
language; Figma-to-code, auto PRD generation, CloudBase backend integration,
CloudStudio one-click deploy
- CodeBuddy Code (CLI) → install via npm, drive a full dev workflow in natural
language from the terminal: refactoring, bug fixes, lint fixes, merge-conflict
resolution, release notes; pipes into Git, npm and the rest of the toolchain
- Plan mode → decomposes complex requirements end-to-end: analysis, technical
design, code generation, validation and testing
- Multi-model architecture → Hunyuan + DeepSeek domestically, GPT/Gemini on the
international build; all three forms share models, quotas and experience
What it deliberately does not do: it does not force you into a cloud IDE
workflow. The CLI form explicitly promises "no change to developer habits" — it plugs
into your existing toolchain rather than demanding a new environment.
What old behavior it replaces
It replaces the workflow of hand-typing code in an IDE while shuttling between
tools. The old flow: a requirements doc, a design file, a code environment, a
deployment platform — translated and carried between by humans. CodeBuddy chains
"design → code → backend → deploy" together: Figma import generates page code, PRDs
generate themselves, CloudBase handles the backend, CloudStudio produces a link. The
human translation steps disappear.
The CLI form replaces the person babysitting the terminal. Making the computer do
work used to mean writing scripts or executing commands by hand. CodeBuddy Code
supports continuous actions — "generate tests, then run them" — and a single described
requirement (screenshot plus natural language) becomes a branch, code and a merge
request.
In Chinese teams it also replaces the data-sovereignty worry of overseas tools:
the domestic build is 100% domestic models with classified-security (MLPS) compliance.
Financial and government teams cannot send code to GitHub Copilot-class products;
CodeBuddy owns that compliance slot.
Business model
- Individual / domestic: free, with access to DeepSeek and other models
- International: IDE and CLI share quotas, bonus quota during beta, subscription
afterward
- Enterprise: pay-as-needed
Read: the free strategy is standard big-platform behavior — the coding tool is the
entry, and the monetization is the backend cloud services (CloudBase, CloudStudio,
Tencent Cloud). Its contest with Cursor is not on price but on "one pipeline from
code to launch inside the Tencent ecosystem." The deeper that integration, the harder
it is for an independent developer to copy.
Hard numbers
- traffic board: 7.49M monthly visits, +30% MoM (2026-08), on the code-assistant
leaderboard
- Tencent internal: 90% of engineers use it, coding time down 40%+, AI writes 50%+
of code, overall R&D efficiency up 16%+ (official, 2025-09)
- Vendor-cited benchmark: 92% completion on complex engineering tasks, ~120ms
latency domestically vs 300-380ms for international tools
- September 2025: CodeBuddy Code launched, CodeBuddy IDE international public beta
Four-way read
| Dimension |
Call |
| Founder-product fit |
A Tencent Cloud department product; internal dogfooding first, the pain is real |
| Product insight |
Full-form coverage plus Plan mode for complex requirements; "domestic compliance" is a real differentiator |
| Execution quality |
Multi-model architecture, domestic low latency, complete engineering |
| Timing |
AI coding is already a big-platform melee; independent products are being squeezed, and CodeBuddy is one of the squeezers |
The call
Worth watching — a solid sample of the big-platform coding tool, and the numbers
are real.
7.49M monthly visits at +30% MoM puts it in the first tier of domestic coding
assistants. Tencent's internal figures — 90% of engineers, 16% efficiency gain — are
dogfooding numbers from the product team itself, more credible than any external
benchmark: if a tool says "90% of our own people use it," it is at least genuinely
useful internally. That internal-first cold start is something an independent product
cannot easily copy.
Its real advantage is the "code-to-launch" loop, not code completion itself.
Cursor's strength is the single-machine editing experience; CodeBuddy's is that the
design file, the code, the backend and the deployment all live in the Tencent
ecosystem. For a small team already on Tencent Cloud, this closed loop has far less
friction than the "Cursor + self-hosted backend + your own deploy" patchwork.
The open questions: the international build is where it proves itself. Domestic
free-plus-compliance is a defensive play; the international build has to compete with
Cursor and Copilot on model quality and experience, and so far it has only "bonus
quota during beta" with no published conversion numbers. Also, the +30% MoM is growth
shared with MarsCode, Tongyi Lingma and everyone else in a red ocean, not new
demand.
What to watch next
① Paid subscriber numbers after the international build launches — free-to-paid
conversion is the hard metric
② Whether monthly visits cross and hold the 10M line — 7.49M is still the ramp
③ Any public case study from a non-Tencent mid/large enterprise — proof that
external customers, not just internal teams, pay
What you can take from it
Product logic: in developer tools, full-form coverage is not showing off — it
maps to user segments. Plugin for daily completion (manual), IDE for non-technical
users (automatic), CLI for pros (race mode), with shared models and quotas so users
switch by scenario without reconfiguring. Decide where each of your three user
segments enters instead of shipping one form.
Product logic (dogfood first): internal usage is the best cold start. "90% of
our internal engineers use it" is itself sales material — a tool that publishes its
own team's efficiency numbers out-argues any benchmark. An independent product can
run the same play inside its own team and publish the numbers.
Verdict
Worth watching. One of the most worth-watching samples of big-platform coding
tools: internal data is real, full-form coverage is complete, and the Tencent Cloud
loop is a genuine differentiator. The risks are the red ocean and the unproven
international conversion. It is a product that shows both the upside and the downside
of the "big-platform builds a developer tool" route at once.