Research Tools Can Now Remember Even the Tabs You Closed

Beyond tab management: remembering the research process

On September 16, 2026, Mozilla added Mistral Small 4 to Smart Window, Firefox’s optional artificial intelligence feature. Users can choose among Gemini 3.1 Flash Lite, Qwen3-235B, and Mistral Small 4 for each conversation, so Firefox has not been fully replaced with a single model.

Smart Window answers questions using the open tabs, browsing history, and saved memories that a user has permitted it to access. It can summarize long pages, compare several tabs, find past pages whose title or address the user has forgotten through screen previews, and suggest duplicate tabs and related tab groups.

What has become cheaper here is not simply writing an answer. It is reading many documents at once while retaining their context. According to Mistral, Small 4 has 119 billion parameters in total, but 6.5 billion are active for each response. The company says it can handle up to three times as many requests in the same time and reduce task-completion time by 40% compared with its previous model. These are the company’s own test figures, so the speed and cost Firefox users experience need separate verification.

That cost was already hidden in the habit of keeping tabs open. In a study of 103 people’s tab use, the median number of tabs at which management began to feel burdensome was eight, and 67% experienced this state at least once a week. Of 633 open tabs studied, 17.7% were kept to avoid the effort of finding them again later, while 37.3% served as reminders not to forget a task.

A 2009 study of 180 Korean users showed a similar need. People created bookmarks not only to navigate to frequently visited sites, but also for long-term preservation, information they might need in the future, and material to review later. They rated features for noting why something was saved and sorting by frequency of use highly. It is an old study and cannot be read as a measure of current user shares, but its design lesson remains useful: products need to retain why something was saved, not just its address.

An office manager’s open tabs become an application workflow

Imagine an office manager at a window-installation company with 12 employees searching for government support programmes. They open central-government notices, local-government guidance, application sites, submission forms, and past selection cases, then check whether the company is eligible.

For each notice, they copy eligibility requirements and deadlines into a notes app, find required documents in attachments, and write them down again. When they return a few days later, they must first remember their search terms. They may also open this year’s notice and last year’s similarly named notice by mistake. Tabs are both a document folder and a to-do list, so they are hard to close.

The change Smart Window points to is the ability to remember this whole process. If the office manager asks, “Show me only the programmes I looked at last week whose regular-employee requirements fit our company,” it could gather candidates again from permitted tabs and browsing history, along with pages that support the answer. It could then organize deadlines, submission documents, and contact points in the same format.

A sector-specific service only needs to take one further step. Instead of producing a “good answer,” it can first collect conditions such as company size, region, and industry, then attach eligibility likelihood, deadlines, missing documents, and next actions to each saved notice. It remembers the actual work state, not the tab title.

The tool proposed in this article needs to save web pages, process long documents, and connect search history with notes. As browsers and AI models begin to handle comparisons among current tabs and rediscovery of past pages, builders can focus on the questions and result screens for one field, such as grants or sales.

The place to charge is not the summary itself. Open-tab summaries are entering browser features such as Firefox, Brave, and Edge, and may become close to free. Value remains in features that attach sources so users can check that application criteria were read correctly, assign owners and deadlines, and help a team share its next actions.

Human judgment does not disappear either. The office manager must decide whether the company is actually eligible, check whether a notice is current, and submit the final documents. Mozilla says that questions and related browsing content may pass through its servers to the selected model, so the product also needs controls that separate sensitive materials—such as resident registration numbers, contracts, and customer lists—from content that can be read automatically.

A Korean availability schedule has not yet been announced, and a new repository could become another organizational burden. Rather than saving every tab from the start, it is better to make the tool remember one bounded task, such as “reviewing this grant opportunity” or “researching this sales prospect.”

Outside Korea, people pay for the next action rather than storage

Workona

Workona, a US product for people moving among multiple clients and projects, groups tabs and documents into project-based spaces. It restores tab states after windows are closed and lets users continue across devices, treating tabs as active projects rather than a simple list.

The company says it has more than one million users and is currently profitable, but these are self-reported figures. Its team plan costs about KRW 16,300 per person per month and starts at three people, so a three-person team pays about KRW 48,900 per month for shared spaces, long-running work records, and central administration.

Readwise Reader

Readwise Reader, a US personal reading tool, saves web articles, newsletters, PDFs, and ebooks, and lets users highlight and annotate them. It resurfaces saved material and exports it to other note-taking tools.

After a 30-day trial, it charges about KRW 13,600 per month on an annual plan. It is an example of charging not for closing browser tabs, but for preserving originals, searching full text, resurfacing highlights, and exporting them.

Hunchly

Hunchly, a Canadian tool for journalists and investigators, automatically records visited web pages and lets users add tags and notes by case. Rather than acting like a general bookmark tool, it focuses on preserving the state of web pages that may disappear and documenting the investigation process and sources in a report.

Its basic edition, which handles one case, is free. Its paid edition, which stores multiple cases on a device, costs about KRW 226,000 per year. It combined with the investigation-tool company Maltego in 2025, making it an example of choosing the narrow, expensive problem of evidence preservation rather than general tab management.

Four things you can build now

1. A grant-application memory

• What it does: Extracts eligibility requirements, deadlines, submission documents, and contact points from multiple notices, while saving the location in the original source.
• Who uses it: An administrative staff member at a manufacturing company with around 10 employees that has no dedicated person tracking support programmes.
• Why now: The cost of comparing long notices and pages opened in the past has fallen.
• First screen: Show only the company’s basic conditions, notices closing this month, and missing documents in three sections.

2. An industrial sales research board

• What it does: Collects evidence from prospective customers’ websites, job postings, and press releases that points to equipment investment or responsible departments, then turns it into questions for the next call.
• Who uses it: An industrial-equipment distributor where three or four salespeople call small and mid-sized manufacturers.
• Why now: Browsers can handle reading common points and differences across multiple tabs, while the service only needs to design the sales signals and question format.
• First screen: Show each company as a card with three lines of evidence, a source date, and one question to ask next.

3. An evidence notebook for job changes

• What it does: Collects required capabilities from job postings, company introductions, and interview articles, then connects them with the user’s work examples.
• Who uses it: A production-management candidate reviewing experienced-hire openings at several mid-sized companies at the same time.
• Why now: It is now possible to find closed job postings again and compare requirements across companies using the same criteria.
• First screen: Show application deadlines in order, missing evidence from the user’s experience, and the self-introduction question to revise today.

4. A local care-application board

• What it does: Collects eligibility, schedules, costs, and application methods for after-school care and school-break programmes from district office, school, and welfare-organization pages.
• Who uses it: A guardian who must find school-break care for an elementary-school child around two working parents’ schedules.
• Why now: Long guidance pages scattered across institutions can be rediscovered and compared by condition as one task.
• First screen: List candidates that fit the child’s grade and place of residence, application start dates, required items, and original links in time order.

Why this matters where you are

Check whether people in your market leave research tabs open because they need to return to a decision, not because they merely want to save a link. The relevant sources, eligibility rules, and privacy expectations will differ by market, but the product opportunity is to restore a work state with evidence and a next action. Start with one bounded workflow instead of building another general-purpose tab repository.

What to check today

Ask one person who recently finished researching a grant programme or sales prospect to show you the tabs and notes they used for 30 minutes. If they had eight or more relevant pages, searched again or kept at least two pages open because they could not close them, and used tab titles to remember the next action, it is worth testing “restore the work state” rather than simple storage.