Perplexity Computer can search folders you authorize on your computer. On Mac, its fully local mode also runs the model and agent on the device. Completely offline use is less clear: the Mac announcement does not explain whether you can start a new task or reopen the app without a connection. See the local file guide and Mac local-mode announcement.
For writing, a useful first task is finding something you already know you saved. Perhaps you’re preparing a product update and remember a customer describing the problem the new feature solves. The comment is somewhere in your notes. You want the original wording and enough context to use it accurately.
Here’s how to approach that search, check what comes back, and turn the useful parts into a draft. The requests below are suggested workflows; we haven’t tested them in a disconnected Mac session.
What does local mode change?
Folder access determines which files Computer can use. The processing mode determines where it works on them.
| Setup | What it means for your notes |
|---|---|
| Local folder access | Computer can find, read and edit files in a selected folder. This alone does not establish where the model runs. |
| Hybrid Compute on Mac | Cloud models handle planning and reasoning, while a local model processes private files and local actions. The workflow uses the internet. |
| Fully local mode on Mac | The model, planning and file processing run on your Mac. Cloud steps can be authorized when needed. |
Perplexity describes the split in its Hybrid Compute announcement. Its folder documentation separately explains file access.
For the example below, keep the search focused on saved notes. If you also want to research today’s competitor announcements, make that a separate task so you can check the sources for each part.
What do you need on a Mac?
The fully local release lists Pro, Max and Enterprise access on supported Apple silicon Macs. It recommends at least 36 GB of memory and offers Qwen 3.6 and PPLX Pro. The memory figure is a recommendation, not a stated minimum.
Install the current Mac app, sign in, and complete the local-model download before trying a disconnected session. Follow the app’s setup instructions to select fully local mode. For folder permissions, use the desktop setup guide: on Mac, check Perplexity → Settings → Permissions, then choose the folder for your task through Folders → Choose folder in Computer.
Start with the folder that contains the material for this piece. For the product-update example, that might contain interview notes, feedback summaries and the feature brief.
How to find a note when you can’t remember the filename
Ask for passages you can check before asking for a finished post. That makes it easier to tell whether the agent found the right material.
Search for the problem the customer described
A customer may have described a problem months before your team named the feature. Searching only for the release name can miss that language.
Suppose you’re announcing a way to reuse a previous project setup. Try a request like this:
Search this folder for customer feedback about repeating setup work,
re-entering the same information, or starting a project from scratch.
For each relevant passage, show the exact wording, file path, and date
if recorded. Include enough surrounding text to explain the comment.
Tell me which files you couldn't read. Don't change any files.Check the source before treating it as evidence
Open the files behind the most relevant results. A note saying “we should ask whether setup is frustrating” is a research question. A note recording a customer’s complaint is evidence of that person’s experience. They could contain similar words and support very different sentences in your announcement.
Check whether a quoted comment came from the customer or from your summary of the conversation. Keep the date: an old complaint may describe a workflow that has already changed.
If you remember a particular note but it doesn’t appear, ask Computer to read that file directly. Compare what it finds with the broader search. A missing result is a reason to investigate the search coverage, not proof that the comment was never recorded.
Make a short source brief
Once you’ve checked the useful passages, ask for a brief with the problem, supporting excerpts and links or paths back to the originals. Keep uncertain interpretations separate from recorded comments.
For example, three people describing repeated setup does not establish that every customer has the problem. A useful brief preserves that distinction so the draft doesn’t turn a few observations into a universal claim.
Which parts still need the internet?
Live web research and cloud-model calls require a connection. A task that combines your notes with current online information therefore has different requirements from a task using only downloaded files.
Perplexity’s Mac announcement establishes on-device processing, but leaves disconnected startup and session behavior unspecified. If you need this workflow on a flight or somewhere without reliable internet, check it before relying on it:
- Finish setup while connected and select fully local mode.
- Disconnect, start a new task, and ask for a passage from a file whose contents you know.
- Check the returned passage against the original, then quit and reopen the app and try again.
Record the app version and selected model, and note any sign-in or network request. A successful answer in an already-open session would not establish that the app can also start a fresh session offline.
You may find stronger offline wording in the Portable Computer help page. That page covers supported Windows, Linux and DGX Spark hardware. It says local tasks can run without a cloud round trip, while sign-in, downloads and cloud steps need internet. Those platform-specific instructions don’t settle the Mac question.
How do you turn the notes into a draft?
Use the verified source brief alongside the feature’s current behavior. Ask the agent to explain which customer problem the release addresses, what changed and what the user can do now. A customer’s request tells you what they wanted; the feature brief tells you what you actually built.
You can continue in Perplexity and ask it to save a draft beside the source files. Its file guide documents creating and editing documents in selected folders. If that gives you a draft you can comfortably revise, you can keep working there.
For ongoing writing with Claude or Codex, StashBase lets you open the source folder and work on documents beside the agent conversation. You can keep the brief, original notes and current draft together, then review the agent’s changes in the document and choose which to keep. See how to write with AI without losing your voice for the drafting and revision process.
That workflow uses the selected agent’s model service; relevant material is sent to it. StashBase’s optional search by meaning also sends text and queries to the embedding provider you configure with an OpenAI or OpenRouter key.