OmniBox
OmniBox is an open-source AI knowledge workspace that clips web pages, parses PDFs, documents, audio and video into editable notes, then answers questions and drafts text from that library with one-click source tracing.
What is OmniBox?
OmniBox, known as 小黑 in Chinese, is an AI knowledge workspace published by Hangzhou Xiaohei Intelligent Technology Co., Ltd. It sets out to do four things in one place: collect content, parse files, organise knowledge, and answer questions or write from what has been gathered. Collection runs through a browser extension available for Chrome, Edge, Firefox and Safari. It saves a full page, a selected region or an online video as a snapshot, then converts it into a document you can edit rather than a frozen bookmark. Parsing covers PDF, Word, PowerPoint, Markdown, audio and video: drop a file in and it becomes editable text, ready for search, tagging, questioning and writing. Images are parsed for their text too, and audio, video and mobile voice recordings are transcribed. The question-answering side carries the main claim. Ask a question and the assistant decides for itself how to proceed, then returns an answer whose supporting passage can be reopened in a single click. When it detects that a question needs current information, it searches public sources and displays the references it used. Two agent generations are documented, Agent 1.1 and Agent 2.1, the second reserved for the paid tier apart from a small daily allowance. Everything lives inside namespaces, independent spaces with their own content, quotas and compute, held personally or shared with a team. Folders, batch operations, tags, smart folders with filter conditions, document sharing and member invitations sit on top of them. The surface area is unusually wide for a young product: web app, installable PWA on desktop and mobile, native iPhone and Android apps, four browser extensions, WeChat and QQ assistants, a WeChat mini-program, and a Bearer-token REST API with Swagger documentation and per-key permission scopes. The code is open source on GitHub, and the publisher documents a Docker Compose self-hosting path for teams that want to control deployment, storage, models and network access. OmniBox describes itself as being in public beta and trial operation, and is candid about the limits of the technology: outputs may be inaccurate and should not drive decisions with legal or material consequences.
What it does
- Clip whole web pages, selected regions or online videos straight into a knowledge base
- Turn PDF, Word, PowerPoint, Markdown, audio and video files into editable, searchable documents
- Ask questions and get answers whose every claim reopens its source in one click
- Draft and rewrite text grounded in what you have already collected
- Search everything with semantic, full-text and filtered queries, and save filters as smart folders
- Save and query content from WeChat and QQ through the built-in assistants
- Drive the whole knowledge base from scripts or external agents through the REST Open API
When to use OmniBox / When not to
A quick filter to help you decide if OmniBox is the right fit.
When to use OmniBox
- Researchers and graduate students who collect papers, PDFs and lecture recordings and need every answer traceable back to the original passage
- Technical writers and knowledge-base managers building a shared team library with folders, tags and smart folders
- Journalists and analysts who clip articles, threads and videos all day and want them parsed and searchable in one place
- Developers and small teams who want an open-source knowledge base they can self-host with Docker and drive through a REST API
- Chinese-speaking professionals who live inside WeChat and QQ and want to save and query content without leaving those apps
When not to use OmniBox
- Organisations bound by the GDPR: the site never mentions it, publishes no Data Processing Agreement and names no Article 27 representative
- Teams whose data may not leave the European Union, since personal data is stored in mainland China as a matter of principle
- Businesses planning commercial use of the outputs, which the terms forbid without the publisher's written permission
- Buyers who need a vendor with a published postal address, security certifications and an auditable compliance posture
- Anyone who needs an interface beyond English and Simplified Chinese, or who cannot pay through WeChat Pay, Alipay or Apple
How to use OmniBox
A typical end-to-end flow, from setup to results.
- Create an account with your email address and a verification code, or sign in through a third-party account such as WeChat
- Start from the Basic namespace you own by default, and create or buy additional namespaces later if you need separate quotas
- Install the browser extension from the Chrome Web Store, the Edge add-ons store, Firefox Add-ons, or the macOS App Store for Safari
- Add the mobile app from the App Store or your Android store, or install the PWA from the address bar in Chrome, Edge or Safari
- Clip a page, a selected region or a video with the extension, or drag files into the knowledge base to have them parsed
- Organise what you collect with folders, tags and smart folders, and invite members if the namespace is shared with a team
- Open a document in reading mode, click Edit at the top right to change it, then click Save
- Ask the assistant a question, then click through to the cited source to check the answer against the original
- Create an API key under Settings and API Keys, choosing its resource scope and its resource, chat, tag and search permissions
- To self-host, clone the repository on a Linux server with Docker, copy example.env to .env, start the stack with Docker Compose, then configure the environment variables for AI and file storage
Pros & Cons
Pros
- A genuinely permanent free tier, with unlimited web collection on both plans and monthly quotas that refresh every 31 days
- Open-source code on GitHub and a documented Docker Compose self-hosting path for teams that need to control their own stack
- Answers carry their sources and reopen them in one click, which makes verification cheap instead of tedious
- The publisher states that AI processing runs on its own infrastructure with no data sent to third-party AI providers
- A very low paid entry point at 35 CNY per month, with 1 CNY add-ons that never expire
- Wide platform coverage: web, PWA, iOS, Android, four browsers, WeChat, QQ and a documented REST API
- Sixty-two pages of bilingual documentation, including Markdown versions written for language models
Cons
- No GDPR framework at all: no mention, no Data Processing Agreement, no Article 27 representative and no published certification
- Personal data is stored in mainland China as a matter of principle, with overseas transfer only on separate consent
- No postal address is published anywhere, and the English footer credits Import AI, Inc. while both legal documents name Hangzhou Xiaohei Intelligent Technology Co., Ltd.
- Prices are quoted in yuan only, and payment runs through WeChat Pay, Alipay or Apple in-app purchase
- No refunds, and no automatic renewal: the subscription has to be renewed by hand or it reverts to the free tier
- The public pricing page renders in JavaScript and shows no figures at all; the amounts appear only in the documentation
- Interface and documentation exist in English and Simplified Chinese only, and the product is still in public beta
Pricing & Plans
OmniBox offers a permanent free plan, Basic, presented as free forever. The cheapest paid entry point is the Pro plan at 35 CNY per month. Quarterly billing saves 6 CNY and annual billing saves 70 CNY over a 372-day term, which the publisher presents as roughly two months free. Consumable add-ons start at 1 CNY, and an additional Basic namespace costs 5 CNY per month.
- 32-page PDF parsing limit
- 20-minute audio or video limit
- 20 MB file uploads
- 600 pages and 60 minutes of parsing per month
- 5 GB storage
- standard processing speed
- base models and Agent 1.1 with 10 Agent 2.1 trials per 24 hours
- 10 Open API calls per 24 hours
- 200-page PDF parsing limit
- 120-minute audio or video limit
- 40 MB file uploads
- 3000 pages and 300 minutes of parsing per month
- 50 GB storage
- 3x faster task processing and priority response
- top-tier models and Agent 2.1
- 1000 Open API calls per 24 hours
- 300 image or text pages (permanent)
- 25 minutes of audio or video (permanent)
- or 1 GB of cloud storage (valid for one year)
- Additional namespaces — 5 CNY per Basic namespace per month or 40 CNY per Pro namespace per month
- each with independent content
- quotas and compute resources
Data, GDPR & hosting
A consolidated view of how OmniBox handles your data.
GDPR overview
There is no GDPR mention anywhere on the site. Neither the Terms of Service nor the Privacy Policy, both read in full, contains the words GDPR, European Union or EEA, and no Data Processing Agreement, standard contractual clauses or Article 27 representative is offered. This is silence rather than an explicit refusal. The framework the publisher does claim is Chinese: the terms are governed by the law of the People's Republic of China, the policy cites Chinese cybersecurity and generative-AI rules, and personal information generated in China is stored in China as a matter of principle. Users are granted access, correction, deletion, withdrawal of consent and account closure, with eight enumerated grounds on which a request may be refused. Requests go to contact@omnibox.pro with a fifteen-day response commitment, and the stated escalation route is China's internet reporting centre.
Who owns the data?
The terms leave input content with the user or its original rights holder, and state that, as between the user and the publisher and within the limits of applicable law, the user also owns the outputs. The account itself is treated differently: it belongs to Hangzhou Xiaohei Intelligent Technology Co., Ltd., which grants only a non-transferable right of use. Users authorise the publisher to use their content, encrypted and strictly de-identified so that no individual can be re-identified, in order to run, maintain, develop and improve the service. Closing the account from Settings permanently removes it and all associated data, although backups may retain copies until they are purged or anonymised.
Reuse rights
Users may reuse their own inputs and the generated outputs freely, since the terms assign them the rights, title and interest in the output. One limit is explicit and easy to miss: without the publisher's written permission, neither the service nor anything it produces may be used for commercial purposes, so the licence is effectively personal, covering study, research, everyday life and leisure. The terms also bar users from entering other people's sensitive information, such as identity numbers, financial accounts or medical records, without authorisation, and warn against relying on outputs for decisions with legal or material consequences in areas like credit, education, employment, housing, insurance, law or medicine. The publisher's own brand names, marks and domain names may not be reused without prior written consent.
Data retention & training
Hosting summary
The publisher states that personal information generated or collected in the course of its business in the People's Republic of China is stored within China as a matter of principle, and that data is transferred abroad only where necessary to provide a service, with separate user consent and the additional measures the law requires. AI processing is described as local: input is handled on server nodes under the publisher's control, on its own servers or in a cloud environment it fully controls, with no transmission to any third-party AI provider. No other hosting country or region is named, and no infrastructure subcontractor is identified. At the network level the domain resolves to a Cloudflare anycast address, which says nothing about where the servers actually sit. Teams unwilling to accept Chinese jurisdiction have one genuine alternative: OmniBox is open source and can be self-hosted with Docker Compose, in which case the hosting is wherever the operator chooses to put it.
Things to keep in mind
Risks and trade-offs to weigh before adopting OmniBox.
- Convenience can quietly replace understanding: a knowledge base that answers for you makes it easy to stop reading what you saved, and the one-click citation only helps if you actually click it
- Answers can be wrong and still sound right; the publisher says so itself, warning that outputs may be inaccurate, incomplete or inappropriate and must be checked before anything important rests on them
- A single account concentrates years of reading, research notes and possibly client material in one place, which raises the stakes of a weak password, a shared device or a lost phone
- Personal data is stored in mainland China and the site says nothing about the GDPR, so uploading client or employee information may create an exposure your organisation has never assessed
- The training reservation is open-ended: the publisher will not use your data to improve its base models unless that data is anonymised and de-identified, and no opt-out switch is documented anywhere
- Two different company names appear depending on the language of the page and no postal address is published, which makes it hard to know exactly who you are contracting with
- Commercial use of the outputs requires written permission under the terms, and refunds are not supported, so a paid namespace bought for work may not be usable as intended
Setup & Integrations
Technical difficulty
Cloud use is easy: an email address and a verification code create the account, the browser extension installs in one click from the Chrome, Edge, Firefox or macOS App Store listings, and the PWA installs from the browser address bar. Creating an API key takes eight steps in the settings. Self-hosting is a different matter and the publisher says so plainly: it needs a Linux server with root access, Docker, a git clone, a Docker Compose command and manual configuration of environment variables for AI and file storage, without which those functions fail.
Deployment
Apps stores
Integrations
Supported languages
Behind OmniBox
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Is OmniBox free?
How much does the paid plan cost?
Does OmniBox have an API?
Can I self-host OmniBox?
Where is my data stored?
Is OmniBox GDPR compliant?
Are my documents used to train the models?
Is there a minimum age?
Can I get a refund?
How do I contact the publisher?
Should you pick OmniBox?
OmniBox is a real, shipped product, and that matters more than it sounds: an iPhone app on the App Store, extensions on four browser stores, a versioned open-source repository, sixty-two pages of dated bilingual documentation and a working REST API all point the same way. For anyone who reads and collects for a living, whether researcher, journalist, technical writer or analyst, the combination of one-click clipping, broad file parsing and answers that reopen their own sources is a coherent proposition, and the free tier is generous enough to test it properly. The reservations are about governance rather than the product. Nothing on the site mentions the GDPR, no Data Processing Agreement is offered, no security certification is published, and personal data is stored in mainland China by default. No postal address appears anywhere, and the English footer credits Import AI, Inc. while both legal documents name Hangzhou Xiaohei Intelligent Technology Co., Ltd. — an inconsistency worth clarifying before committing anything sensitive. Payment is in yuan through WeChat Pay, Alipay or Apple, there are no refunds, and renewal is manual. The terms also forbid commercial use of the outputs without written permission, which rules out a good part of business use as published. The result suits individuals and small teams outside regulated contexts very well, and Chinese-speaking users especially, since WeChat and QQ are first-class surfaces rather than afterthoughts. European organisations handling personal or client data should stay cautious until the publisher documents a compliance posture. Teams who want the capability without the jurisdiction do have a real option that most vendors never offer: the code is open, and the whole thing can be self-hosted.
- Choosing a selection results in a full page refresh.
- Opens in a new window.