ima files documents, web pages, WeChat articles, images and recordings into personal or shared knowledge bases, then answers, summarises and drafts against that library with source tracing — on Windows, macOS, mobile, web, a WeChat mini program and a browser extension.
Tencent ima is an ai knowledge base assistant: you file documents, web pages, WeChat public-account articles, images and recordings into a library, then search, read and write against that library instead of against the open web. The vendor frames it as a knowledge-base-first "search–read–write" experience, and the client opens on four entries — New chat, Knowledge bases, Discovery square and Q&A history — with a storage readout under the avatar on desktop.
Three Tencent company names attach to one product. The privacy guideline names 深圳市腾讯计算机系统有限公司 as service provider, gives its registered address as Tencent Building 35/F, Nanshan District, Shenzhen, and says affiliates jointly operate the service; it is also the licence-agreement counterparty. The China App Store names 腾讯科技（深圳）有限公司 as provider — same group, different legal person. Tencent Appstore names 深圳市腾讯计算机科技有限公司 as Android developer and 深圳市腾讯计算机系统有限公司 as operator.
The engine is not one house model. The download page states the product has connected Tencent Hunyuan, DeepSeek and other models, and the privacy guideline publishes three Chinese algorithm filings: Hunyuan's multimodal algorithm, DeepSeek's large-language-model algorithm from 北京深度求索人工智能基础技术研究有限公司, and Zhipu's content-generation algorithm from 北京智谱华章科技股份有限公司.
ima is also not only a private notebook. The official site states the knowledge base square spans more than twenty industries — technology, finance, education, healthcare, law and public administration among them — and has accumulated over 400 million knowledge base files, both vendor-published counts.
Ingestion is the real boundary — anything outside it never becomes answerable material. The site names six intake routes.
The WeChat routes are what rivals outside that ecosystem cannot copy.
Retrieval is scoped, not global. The site describes recalling knowledge base sources precisely against your question, combining them with the model's deep-thinking capability, and returning an answer that supports tracing back to the original text. Scope narrows to a base, a tag or a folder, and above it sits copilot, a knowledge partner with long-term memory that learns your preferences.
Recording accepts up to two hours per session, supports multiple languages, and produces transcript and summary automatically — the ai meeting notes path, with the two-hour ceiling a hard number. Speakers are separated through voiceprint processing that labels them anonymously as Speaker 1, Speaker 2 and so on, explicitly without identifying or linking anyone, and those voiceprint features are then deleted.
Uploaded documents convert into a mind map, a spoken podcast with human-like voices, or a short quiz, and document Q&A matches relevant charts into the answer.
A personal knowledge base is private to you. A shared knowledge base introduces four roles: the creator holds every permission including deleting the base and deciding whether it goes public; the administrator sets content permissions, edits content and invites members; the collaborator uploads files and asks questions; the ordinary member views and comments.
A knowledge account publishes a base to the square for others to find and subscribe to, with data analysis, official verification and creation guides supplied to creators. Skills can be published alongside bases.
The working order that avoids rework follows from two rules most people learn the hard way.
The jobs that fit share a shape — you already own the material, it is scattered across formats, and the answer must be defensible against its source. The official site works this through six professional profiles. Students import courseware and class notes and get key points turned into a podcast or a mind map; teachers pull teaching material out of the base while preparing lessons and have lesson plans generated against it. Analysts load survey material, meeting recordings, expert interviews and market policy, then have an industry research report drafted from that pile. Lawyers import case details and statutes to get points of dispute summarised and litigation documents drafted, public-sector staff import policy documents for interpretation and official drafting, and creators work from trending topics.
Shared bases carry four further patterns: public-service enquiry desks answering citizens around the clock; enterprise knowledge sharing that turns scattered experience and data into reusable assets; class teaching where a teacher uploads board work and exam points for students to query; and small-group research.
One boundary crosses all of them, set by the agreement rather than by taste: generated content is offered for personal study and reference, and where a matter may have a material effect — legal, medical and financial scenarios are named — the service explicitly does not replace a professional's answer. The square carries legal, financial and medical bases from named institutions, so that is not abstract.
Eight product surfaces plus one embedded entry are listed on the official download page.
| Surface | Requirement or entry | What it is for |
|---|---|---|
| Windows client | Windows 10 and above | Bulk import, administration, long reading |
| macOS client | macOS 12 and above | Same as Windows |
| iOS client | iOS 16.0 or later, 184.5 MB, Productivity category | Capture and query on the move |
| Android client | Version 2.6.10, updated 14 September 2026 | Same as iOS |
| HarmonyOS client | Scan to download on device | Domestic device coverage |
| Web version | Click through to use in a browser | No-install access |
| WeChat mini program | Scan inside WeChat | The only route for WeChat articles and files |
| Browser extension | Chrome and similar browsers | Capturing web pages into a base |
| WorkBuddy | Use ima inside WorkBuddy | Embedded workplace entry |
Sign-in is built on WeChat: the privacy guideline states the product supports registering and logging in with a WeChat account, collecting its unique identifier, avatar and nickname to identify you and synchronise data across devices and clients. Apple sign-in on Apple devices and Google sign-in on Android are also offered, but both then require binding a phone number. Old versions may stop working once a new one ships.
There is no published price card, which is itself a finding. The public documents still fix the shape of the money.
This reads as a free core with metered advanced usage on top. The 1% figure is explicitly a beta-period rate, not a standing commission.
Independent placements agree on the category and disagree on the neighbours. An independent tool directory files ima under AI office tools and AI productivity, calls it an AI knowledge base tool from Tencent centred on knowledge acquisition, building a personal knowledge base and intelligent writing assistance, and lists six comparable tools beside it: 飞书多维表格, 奇觅, Hoarder, 猫头鹰AI网页订阅, 秒言AI语音输入法 and Ponder AI.
The China App Store's behaviour-driven "you might also like" rail draws a different map, placing ima next to 得到大脑 (formerly Get笔记), 秘塔AI搜索, 扣子, flomo, FlowUs, 千问, Cubox, 讯飞星火 and MarginNote 4 — note tools, AI search and agent platforms rather than office suites.
A third-party industry-media review from March 2026 compares ima directly against Get笔记, Notion, Obsidian and Feishu Docs after three days of use. Read together, the placements say one thing: the deciding variable is not feature count but where your material already lives. If your reading runs through WeChat, the ingestion routes are the argument; if not, ima competes on open ground.
The privacy commitments reward precise reading. The vendor states it will not use data stored in your ima knowledge base for model training without authorisation — the qualifier matters — and explains the base is accessed only when you initiate a service against it. Storage and retrieval run on Tencent Cloud Computing (Beijing) as an entrusted processor, and information collected within China stays within China with no cross-border transfer at present. Account deletion runs through Me, Settings and Account security, after which nothing linked to the account can be recovered.
Independent signals are thinner than the official record. A March 2026 industry-media review based on three days of use reported that AI search accuracy was not high enough on complex questions, that mobile lagged the desktop client with some advanced functions desktop-only, and that notes remained relatively simple. A tech-media hands-on at launch in November 2024 found the interface streamlined and the knowledge base the standout idea, while noting the product then had no mobile app — a gap the platform table above shows has since closed. Store ratings differ by platform: the China App Store shows 4.8 from about 26,000 ratings, while Tencent's Android market shows 4.3 alongside 183,000 downloads and no rating sample size.
The clients are free to download and the core loop works without paying. The App Store lists the app as free with in-app purchases, the privacy guideline names purchases of compute credits and paid knowledge bases, and paid subscription remains invite-only.
The published statement is that the vendor will not use data stored in your ima knowledge base for model training without authorisation, and that the base is accessed only when you initiate a service against it. That qualifier is a commitment against unauthorised use, not a claim the data is never read.
At most ninety days. Knowledge base content is separate and persists until you delete it, at which point the cloud copy goes too. Move anything you want to keep into a base.
The site states the product has connected Tencent Hunyuan, DeepSeek and other models, and the privacy guideline publishes three algorithm filings covering Hunyuan, DeepSeek and Zhipu's ChatGLM series.
Rights to content you upload stay with you, and the agreement states rights to AI-generated content belong to you or the rightful owner unless an interface says otherwise or you agreed something else, while Tencent may store and use that content within the law to run the service. The rules separately forbid selling or commercially using service material outside written authorisation.
Once a person stops being an administrator or collaborator of a shared knowledge base, for any reason, they can no longer dispose of any content in it, including what they contributed themselves.
