The browser reimagined as a collaborator
The browser wars are entering a new chapter. Where Chrome and Safari treat the browser as a passive viewport — you click, it renders, you copy-paste between tools — Strawberry asks a different question: what if the browser worked while you weren’t at the keyboard?
Strawberry is a desktop browser (macOS and Windows) with a built-in AI companion. Not a sidebar chatbot. Not a bookmarklet. A real agent that reads your tabs, moves between websites, and can execute multi-step workflows — with your explicit approval at every irreversible step.
I spent time testing the current build to understand what it can and cannot do, and what it means for teams who live in the browser.
The AI agent that rides along
Here is the framing Strawberry lands on from day one: your AI companion is a co-driver, not a replacement for you. It watches what you do, learns the context, and then offers to take over the next pass.
This matters because most “AI browsers” bolt a chatbot onto a normal browser window. Strawberry’s approach is deeper — the companion has access to your open tabs, your session state, and the content the page is showing right now. You are not pasting a transcript into a chat and hoping the model can infer the rest.
The practical upshot: instead of “summarize this page,” you get “while I have this page open, also check the competitor’s pricing page I pinned last week and send me a comparison.”
What actually works today
I tested three of the advertised routines against real, public sources. Here is what held up and what did not.
Research a company profile
Ask Strawberry to research a company and it navigates to the company’s website, investor relations, Crunchbase, LinkedIn, and G2 — pulling key facts into a single summary with source links. The output is a structured report (funding, competitors, key people) rather than a wall of text.
The critical detail: it asks before taking action. When it tried to open a LinkedIn profile, a prompt asked for approval rather than silently scraping. This is the guardrail that makes an agentic browser actually usable in a team setting.
Spreadsheet assembly from multiple sources
The “Extract Web Data” playbook collects repeated structured information across many pages and writes it into a Google Sheet or CSV. I pointed it at a set of public conference speaker pages and asked it to build a table of names, titles, and affiliations.
It took roughly one minute to produce a clean sheet — no copy-pasting across 20 tabs. The catch: you need to be specific about the columns up front. Vague requests (“get the speakers”) return partial rows that need cleanup.
Slide deck generation from a brief
Strawberry can outline a story and build the first version of a slide deck, pulling source material from your tabs and connected docs. The generated decks are rough — clearly first drafts — but they remove the “blank page” problem. I have seen teams spend hours fighting with an empty slide; Strawberry hands back three filled-in slides in thirty seconds. That is a different phase of the work.
Where the alpha shows
Strawberry ships what it calls “early-stage” software — honestly labeled, and the gap is visible.
Speed varies. Heavier workflows (building a full deck, running a multi-step routine) take noticeably longer than a manual pass. Simple lookups are fast; anything that involves real navigation, waits, and re-checks runs in the tens of seconds to a couple of minutes.
Form complexity trips it up. Forms with dynamic fields, multi-step wizards, or aggressive anti-bot flow are the places the agent pauses — or asks you to take over. This is not unique to Strawberry; browser agents across the field hit the same wall. But it is worth knowing before you promise a fully hands-off routine.
Consistency, not magic. Strawberry does not hallucinate its way through a task the way a pure LLM might. It is also not perfectly reliable on the first run — expect to review and refine. The mental model to carry: a capable junior teammate who sees the whole task surface and asks before committing.
Pricing and the credit model
A recurring frustration with AI tooling is hidden usage costs. Strawberry is refreshingly direct here:
| Plan | Monthly | Credits | Target |
|---|---|---|---|
| Free | $0 | 2,000 | Casual experimentation |
| Intern | $20 | 8,000 | Part-time assistant |
| Part-time | $100 | 30,000 | Daily workflows |
| Full-time | $250 | 75,000 | Primary work assistant |
Regular browsing (the actual browser experience) does not consume credits. Credits are only drawn when a companion is actively thinking or taking actions — researching, writing, extracting, filling forms. That makes the ceiling predictable.
For a team of five analysts each running a few research routines a day, the Intern or Part-time tier lands in the same range as a mid-tier SaaS tool. The risk is over-running credits on a heavy month and having routines queue up waiting for the next refill.
What the security model buys you
Strawberry is not trying to be a privacy-focused browser in the Firefox-or-Brave sense. Its differentiator is data locality for agent actions:
- Passwords, history, and cookies stay local — encrypted on your device, never visible to Strawberry’s servers.
- SOC 2 compliance — the company states it meets SOC 2 requirements for data management.
- No training on your data — your browsing activity and data are not used to train AI models.
- Approval gates — companions ask before sending messages, changing important records, or committing work on your behalf.
For individuals, this is a reasonable posture. For teams evaluating it for sensitive workflows, the SOC 2 badge and the local-storage defaults are the relevant controls — and the “ask before acting” model means you retain a human review point on anything consequential.
The real question: does the leverage pay off?
Agentic browsers sit at a sharp edge. The ones that survive do so not by replacing you, but by taking the relentless, repeatable passes off your plate so you can stay in the judgment layer.
Strawberry lands closest to that line among the agentic browsers I have tested. It is not going to write your keynote while you sleep, and it is not going to handle every form a Fortune 500 site throws at it. But the research, spreadsheet assembly, and deck-first-draft workflows are credible — and they are the kind of busywork that scales with team size.
If you are the person on your team who always ends up “building the sheet” or “summarizing the tabs,” Strawberry is the closest thing yet to offloading that work to someone who actually sees your context. It ships as early-stage, and it looks like it. But the alpha is the point: this is the interface where the browser stops being a dumb viewport and starts being a collaborator.
The download is free to try, and the 2,000 monthly credits on the free plan are enough to test the core routines. I would start there with one workflow you already know end-to-end, then decide whether the co-pilot model fits how your team works.
FAQ
Is Strawberry safe with sensitive data?
Strawberry keeps passwords, history, and cookies encrypted locally on your device — encrypted on the device, never visible to the Strawberry servers. The company states it meets SOC 2 requirements for secure data management, and it explicitly does not use your browsing activity or data to train AI models. Companions also ask before taking irreversible actions.
What platforms is Strawberry available on?
Strawberry is a desktop app for macOS and Windows. There is no iOS or Android version. You enter your email and receive a download link.
How do credits work?
Regular browsing is free and does not consume credits. Credits are drawn only when you chat with an AI companion or when a companion takes actions (researching, writing, extracting, form-filling). The free plan includes 2,000 credits per month, refilled monthly.
How much does Strawberry cost?
Pricing is: free (2,000 credits), Intern ($20 / 8,000 credits), Part-time ($100 / 30,000 credits), and Full-time ($250 / 75,000 credits). You can also top up credits at $10 per 1,000 when you need more mid-cycle.
Can Strawberry transcribe meetings?
Yes — Strawberry can listen to meetings, attach the transcript to your companion chat, and help turn the recording into a summary, action items, and follow-up messages. It works with any video conferencing tool and does not require adding a meeting bot to the call.
What can I delegate to Strawberry?
Anything that can be done in a browser. The playbooks cover sales prospecting, company research, dataset extraction, candidate sourcing, competitor analysis, slide-deck creation, SEO audits, financial forecasting, and daily briefs. If it is browser-based, it can be automated.
What search engine does Strawberry use?
Strawberry lets you choose among the major search engines — the default is Google Search, but you can point it at Bing, DuckDuckGo, or others.
