Strawberry vs Perplexity

Half the evidence behind the decision is not on the public web.

Think about the last thing you were asked to have an opinion on by Friday.

Say it is the vendor your team wants to sign, and your ops lead wants one page before she approves it. She needs to know what the tool costs you today, what the current contract commits you to, and whether the last integration you bought from them ever worked.

The public half is easy. Pricing pages, funding, reviews, the outage last March. The private half is the part that decides it. It sits in your billing tool, in a PDF on the shared drive, and in eleven tickets nobody has read since spring.

Perplexity documents Research mode performing dozens of searches and reading hundreds of sources. It delivers a report you can export to PDF, a document, or a Perplexity Page.

Strawberry does both halves in one pass. Your AI companion researches the public web, then opens your billing tool and reads the invoices. It quotes the renewal clause straight out of the contract. It counts the tickets that name the vendor. The one-pager arrives with a source on every line, and the eleven ticket links attached, ready for your ops lead.

You should decide how hard your AI companion thinks, and which model does the thinking.

Not every job deserves the same effort.

Checking whether a vendor still exists is not the same work as deciding whether to renew them for two years.

Perplexity documents automatic model selection. Its help centre says Research automatically selects the optimal models for your research needs, so you cannot manually choose a specific model when using that mode. Its Personal Computer article says Computer creates teams of agents across models and selects the best combination for each task automatically, so you do not need to choose a model manually.

Strawberry hands that dial to you. Each AI companion has its own model and its own thinking effort. Keep one tuned for fast triage, and another turned all the way up for the decision that costs money. Efficient mode is quicker and cheaper. Expert mode is for the memo somebody will act on.

Every run records the credits it used. So you can see what the cheap pass cost and what the deep pass cost, and set the dial accordingly next time.

Research is only finished when it has been checked in your own accounts and written where your team works.

Strawberry is the browser your billing tool and your ticket queue are already open in, so your AI companion works in them directly. That is how it reads the actual invoice rather than the vendor's price list, and the actual ticket rather than a review of the product.

Then the result has somewhere to go. Across 2,000+ integrations, the one-pager lands in the doc your ops lead opens. The number goes into the spreadsheet finance keeps, and the summary posts to the channel where the decision gets made. Nothing has to be copied out of a chat window and reformatted.

Perplexity documents its own reach. Its Computer article describes authenticated integrations with Gmail, Outlook, GitHub, Linear, Slack and Notion, sub-agents for distinct task types, and execution in a secure isolated sandbox. Its Pro connectors article lists Google Drive and Dropbox, and says file syncing is not supported on the Pro plan, so a changed file has to be uploaded again.

Nothing sends and no record changes until you have approved that exact action.

Research that ends in an email to the vendor, or a changed field in your CRM, has crossed into work other people see. Strawberry treats each of those as its own decision.

Every connected action is set to ask, always allow, or off. Let it draft the note to the vendor as often as you like. Require your approval before anything reaches them. Renewal is a purchase, so payments stay switched off entirely.

Approval gives you a real editor rather than a yes or no box. You see the destination, the recipient, the full text and the attachments before any of it moves. A reply shows you the address it resolved, so you can correct it before it goes.

Your AI companion should keep the sources you trust and the ones you have ruled out.

Good research is mostly judgment about sources.

You already know the vendor's own case studies do not count. You know the review site your industry uses is bought. You know one analyst is worth twenty blog posts.

Say that once and your AI companion keeps it. The next vendor review starts from your standards rather than a blank prompt. You can open the memory file and correct a line you no longer agree with.

The method becomes reusable too. Save the vendor review as a skill and anyone on your team runs it in one line, in their own words. Put it on a schedule, or a trigger such as a renewal date landing in your inbox. Each run is logged with what it did and what it cost. Perplexity documents its own scheduled work, and says Pro and Max subscribers can create a maximum of ten tasks.

Then hand the whole thing over. An AI companion carries its memory, its skills and its connected accounts, so your ops lead receives a colleague who already knows how your team judges a vendor. The team workspace holds the shared files, and shared credits cover whoever is doing the work.

Frequently asked questions