Strawberry Browser Is Easier to Try—but It Remains a Beta Work Tool

Strawberry Browser is no longer the scarce, free-only experiment described when it first attracted attention in 2025. It remains labeled beta, but people can now download it for Windows or macOS, choose between free and paid plans, and use its AI agents for work across websites.

The original premise remains intact: Strawberry gives an assistant access to browser context and lets it take actions instead of merely returning a chat response. The important update is that access has broadened while the central limitation has not disappeared—an agent operating inside authenticated websites still needs human review when mistakes could affect accounts, communications or records.

What Strawberry Browser does differently

Strawberry combines a conventional desktop browser with AI assistants called Companions. A Companion can examine the page in front of it, move between tabs, enter information and complete a sequence of browser actions from a natural-language request.

That distinction separates an agentic browser from a chatbot placed in a sidebar. A chatbot can explain how to update a customer record; a browser agent may be able to open the relevant service, locate the record and enter the update. The second approach removes manual steps, but it also gives an imperfect AI system more opportunities to make a consequential mistake.

Suitable tasks are generally structured and easy to verify: gathering named fields from public pages, summarizing material across several tabs, preparing a draft from an email thread or transferring information into a table. Claims that the browser can automate “anything” should be understood as a description of its intended scope, not a guarantee that every website, interface or workflow will work reliably.

Availability and pricing have changed

As of August 2026, Strawberry’s official pricing and platform page labels the product beta, lists Windows and macOS support, and offers a free plan with 2,000 monthly credits alongside an Intern plan costing $20 per month for 8,000 credits. It also states that ordinary browsing does not consume credits and that passwords, chats, history and cookies are stored locally on the device.

Credits are consumed when a Companion chats or performs browser work, so “free browser” and “unlimited free automation” are not equivalent. The practical cost depends on how often an agent is used and how demanding its tasks are, while beta pricing and allowances may change.

This makes the old closed-access description obsolete. Strawberry is easier to evaluate without committing to a paid subscription, but its present pricing structure positions AI execution as a metered service rather than an unrestricted feature bundled with the browser.

Earlier tests show why the beta label matters

A November 2025 hands-on review documented the earlier macOS and waitlist phase after two weeks of use. Its tester found that Companions could plan and perform multi-step tasks, but sometimes required intervention for CAPTCHAs or sign-ups; the review also identified slower performance, free-tier limits and a learning curve.

Those observations describe an earlier version, so they cannot establish the performance of every current build. They do, however, illustrate a persistent constraint of browser agents: websites change, authentication interrupts automated flows, and visually similar controls can lead an agent down the wrong path.

A separate February 2026 account of installing Strawberry described an uncomplicated download and onboarding process, but the tester connected a newly created Google account instead of an established primary account. That choice is not a universal requirement, yet it demonstrates a sensible distinction between testing an agent’s behavior and immediately granting it access to valuable existing data.

Local browser data does not remove every privacy question

Local storage for passwords, history and cookies is useful, but it does not by itself answer what information an AI task must process. If a Companion is asked to summarize an inbox thread or act inside a connected application, the relevant content necessarily becomes part of the task context available to the system.

The meaningful question is therefore not simply whether the browser stores credentials locally. Users also need to consider which pages the agent can see, which connected accounts it can reach, what information appears on screen and which actions can be reversed after an error.

Approval prompts are especially important for outbound messages, publishing, purchases, account changes and form submissions. A draft or extracted table can be inspected before use; an incorrectly sent message or modified customer record may be much harder to undo.

Where Strawberry’s approach is most credible

Strawberry is most compelling for repeated browser work that has a clear finish line and a reviewable result. Research, list building, information extraction, draft preparation and routine administrative updates fit that pattern better than open-ended instructions involving judgment or sensitive decisions.

It is less suitable as an unsupervised operator for financial transactions, hiring decisions, regulated information or any workflow where a single mistaken click creates significant harm. People who only need page summaries may also find that a simpler assistant provides the required result without access to multiple tabs or authenticated services.

The product’s progress is real: Strawberry has moved from limited early access to a downloadable Windows and macOS beta with defined usage tiers. Its value nevertheless depends on a narrower test than the original “wow” framing suggests—whether the assigned work is predictable, observable and safe for an AI agent to attempt under human supervision.

Also read:

• Monero Web Wallets in 2026: Private, Browser-Based, and Easier Than Ever
• AI and Humans Can Work as Partners—But the Gains Are Uneven
• Internet Banking Skips the Branch Visit—but Not the Security Work