Tencent ima: Putting Search, Reading, and Writing into One AI Workspace

Tencent recently pushed a product called ima into the spotlight. Officially it's described as an "intelligent workspace," but broken down it's quite plain: it gathers search, reading, and writing — three things that used to be scattered across different apps — into a single window. Under the hood runs a dual engine of Hunyuan plus DeepSeek; on the surface there are Windows and Mac clients, plus a WeChat mini program, with data synced across all of them.

The value of this kind of product is hard to convey in a single tagline; you have to look at which action it makes shorter. In the past, looking up a source, reading a long article, and then writing a summary meant jumping back and forth between a browser, a reader, and note-taking software. What ima wants to do is stop people from having to jump.

Chinese AI Workspaces Are Fighting for the Same Spot

Spread this category out and you'll see everyone crowding toward the "entry point." Whoever gets users to open them first thing wins the retention that follows. Those that started with search want to put Q&A into the results page; those that started with notes want to stuff AI into the editor; and products in the WeChat Official Account ecosystem want to use WeChat to drive up usage frequency.

ima's choice is to put its weight on "knowledge": on one side, retrieval of public information; on the other, the accumulation of personal materials, with a copilot threading the two together. Tencent gives new users 500 compute credits on sign-up and tops them up with 100 a day. That design sends a signal: it wants this to be a product people open every day, not a toy they pick up once in a while.

The Entire Chain of Getting Work Done Solo

What really separates products is often not any single capability but whether the chain is complete. An idea goes in, material comes out; material goes in, a draft comes out; a draft goes in, a finished piece comes out. If any link breaks, the user has to reconnect it themselves, and the so-called "workspace" degrades into just another chat box.

By putting search, reading, and writing into one interface, ima is essentially betting on one thing: that users will pay for "one less switch." That bet is especially obvious on mobile. The mini program version lets you jot down inspiration and fire off a follow-up question on your phone, while the desktop client handles the heavy work that needs a big screen; with both ends synced, this combination is closer to how users in China actually work.

A Few Problems It Doesn't Solve

The first is the trust boundary. When search, personal documents, and generated content are mixed together, it's hard for users to tell where a given conclusion came from. No matter how detailed the source labels are, once the model starts rewriting and stitching things together, tracing back is still laborious. The second is migration cost. Personal knowledge bases fear lock-in above all; if data can't be imported or exported, users won't dare put anything truly important into them.

The third is competition from similar products. This track is already crowded — from office suites to large-model vendors to veteran note-taking tools, nobody wants to give up the entry point. For now, ima looks more like a product that reassembles existing capabilities than a reason it has to be this one. Whether it succeeds depends on whether the copilot's "learns your habits" mechanism actually makes using it smoother, rather than adding another black box that needs to be trained.

Why "Smooth" Is Harder Than "Powerful"

Making model capabilities the strongest is a lab's problem; making a product smooth to use is a product team's problem. The difficulties in these two things are completely different. The former relies on compute and data; the latter relies on understanding specific usage scenarios. For a workspace to actually get used, it has to answer a lot of trivial but fatal questions: is it fast enough to open, does the input box block the content, can you find your history at a glance, will something you've half-written get lost.

None of these questions can be solved by a "stronger model." They test how considerate you are toward the workflow. This is the biggest turning point for Chinese AI products over the past year or two: from comparing whose model scores higher to comparing whose experience is stitched together more tightly. Users' switching costs are high, but as long as a tool genuinely saves them half an hour a day, they'll stay — and hand over their data to it.

That's also why the word "workspace" is worth taking seriously. A workspace isn't a pile of features; it's a series of actions welded into a single path. Features can be copied; a path is hard to copy. Whoever makes the chain from finding material to finishing a draft the smoothest holds the entry point. ima stands on this line: it doesn't chase wow, it chases being used.

What's Worth Watching

A clear shift in Chinese AI apps over the past two years has been from "showing off capability" to "competing on smoothness." Users don't really feel it when you compare model benchmarks; but clicking the mouse a few times less and copy-pasting a few fewer blocks is something they notice the same day.

Another signal is the dual engine. The Hunyuan-plus-DeepSeek combination shows Tencent isn't fixated on using only its own model. To users, whose model it is doesn't matter; what matters is completing the task reliably, cheaply, and quickly. This kind of pragmatism is increasingly common in Chinese products, and in a way it explains why competition at the application layer ultimately comes back to experience rather than parameters.

Whether to use it depends on whether your usual way of working is "information-intensive." If your daily routine is digging for clues in piles of material and then turning those clues into something written, the time this kind of tool saves is quantifiable. If you only ask a question once in a while, then it's no different in essence from any chat entry point. The boundary of a tool is often just the boundary of its user's needs.