Opera WebAI Offline Writing: What to Test First in 2026
A practical test for spoken drafts, local rewriting, and work you can reopen without a connection.
October 28, 2026
Opera WebAI offline writing needs a test of the whole setup. Its model can run on your computer, but speech input, the writing page, and saving need their own checks. Opera's preview lets websites and extensions use models on your computer. If you plan to dictate a draft, test voice-to-text, the writing tool, and saving separately before relying on them offline.[2][3]
Computerworld's October 28 report follows Opera's October 27 announcement and TechCrunch's coverage. Opera calls this an early preview. Check its support list before planning a switch.[1][2][5]
Suppose you want to finish a delivery complaint on a train with no connection. The model might shorten your paragraph perfectly while the page you need refuses to load. I'd want to find that out at my desk, with the original text still available.
Start with the browser you can actually use
Opera's launch announcement lists Opera One first, followed by Opera GX and Opera Neon. Its public explainer says the preview is available in all three, so the pages don't quite agree. Start with Opera One and check the setting in your installed version. Don't assume another Opera browser has it yet.[2][3]
Opera lists the supported computers: 64-bit Windows PCs with Intel or AMD chips, plus Macs with Apple silicon. The preview excludes Intel Macs, Linux, and Windows on Arm. A laptop marketed for AI may still be incompatible.[3]
On a supported machine, Opera says to open Settings, go to Opera AI, and turn on Enable Web AI API. A compatible model will download, which may take a few minutes, so stay connected. If your work computer blocks browser settings or extensions, ask your administrator about an approved test instead of trying to get around the restriction.[3]
For someone with an Intel Mac or a Windows-on-Arm laptop, that support list ends the experiment for now. Keep the speech input that already works on your machine.
A local rewrite is only one part of an offline draft
Opera's developer site has a Rewriter playground with controls for tone, length, and output format. Try it with text you know, ask for a small edit, and compare the result. The separate Prompt playground lets you chat with the on-device model.[8][9]
Start with a harmless paragraph about a delayed furniture delivery. Mention that it might arrive Friday, the address still needs to be confirmed, and nobody has agreed to pay an extra charge. Save a copy first. Keep it outside the experiment.
Ask the rewriting tool to shorten the paragraph, then check whether the meaning changed. Did "might arrive" become "will arrive"? Did an unconfirmed address become confirmed? If either changed, restore the original wording before using the complaint. Fluency won't fix the delivery date.
Run the task again offline after loading and setting up the tool. Record exactly what still works. Editing a page that's already open doesn't show whether you can reopen it after closing the browser. Saving a local file isn't the same as syncing a document to an online account.
Write down the failed step rather than marking the whole setup as working or broken. You might still use a loaded rewriting page with a local text file. We haven't run these tests on Opera's preview hardware here, so these are proposed checks, not results.
Check speech input before calling it a dictation workflow
Opera describes WebAI as tools for websites and extensions, with examples such as text generation and content classification. The announcement alone doesn't explain the full speech-to-text setup you plan to use.[2]
Before dictating the furniture-delivery paragraph, test the speech tool by itself. Use a plain text editor, dictate the paragraph, then check what it captured. Disconnect and try once more. If dictation no longer works, a local rewriting model can't recover the words that never came through.
Then enter the checked text in the browser tool. This pause helps you tell whether an error came from recognition or rewriting. A wrong delivery date in the first transcript needs a different fix than a date changed during rewriting.
For desktop hold-to-talk input in supported text fields, DictaFlow is one option. Once set up, its Local Offline mode can transcribe on supported devices without an internet connection. Optional AI cleanup and other cloud features still need a connection. Offline transcription doesn't mean every later editing step will work offline.[4]
This post is from DictaFlow's company blog. It doesn't announce an Opera integration. There is no special connector being claimed here. If your current speech input works well, keep using it for the experiment.
What does the website do with your draft?
A Computerworld report looks at local processing separately from controls on websites and extensions. Analysts interviewed for the report say permissions and device management still matter, even when a model runs on the device.[1]
Opera says WebAI is opt-in, with permissions set for each site. That seems like a sensible first step. Before you put work material into an extension, check what it can access and whether it sends information elsewhere or saves it. Knowing where model inference runs doesn't tell you everything about how the host application behaves.[2]
For everyday writing, test the tool with a made-up paragraph instead of customer records. You can check whether it shortens the text, keeps uncertainty, and saves the result correctly without exposing real documents. If the test goes well, check which tools your workplace approves before using one for business tasks.
Cloud writing tools can be a good fit too. Choose based on the actual behavior you need. If a web editor uploads drafts for account sync, local rewriting alone wouldn't stop that upload. Check the editor's behavior before describing the whole document as staying on your laptop.
Keep the result you can reopen
For a final check, save your edited paragraph, close the browser, and disconnect. Then try to open the saved work. If your chosen app can't do that, decide whether a separate local copy will meet your needs. Don't wait until you need the document to find out.
Pay attention to how your computer responds as you edit, too. Opera says performance depends on the hardware. A short rewrite that runs smoothly on one supported computer doesn't mean a much longer document will run the same way on another.[3]
For Opera WebAI offline writing, start with the delivery paragraph rather than a week of real work. Keep the original beside the shortened version. When you reopen the saved result without a connection, check that Friday is still tentative and the extra charge still hasn't been agreed. That is a useful first document to get right.
Sources
[1] Opera local browser AI, October 28
[2] Opera launch announcement, October 27
[3] Opera WebAI setup and hardware support
[4] DictaFlow product and offline feature boundaries
[5] TechCrunch independent launch coverage