Foxit released PDF Editor 2026.2 on September 1, and the headline feature is not another chat panel bolted onto the sidebar. It is Local AI, which Foxit is calling the first mainstream PDF editor capability that runs supported AI document tasks directly on your Windows or Mac machine instead of routing your file through a cloud service. Pair that with a new Bring Your Own Model option, and the release is really about where your document goes when you ask an AI to do something with it.
What actually shipped
PDF Editor 2026.2 bundles three separate pieces. Foxit Workspace is a new AI hub that pulls documents, links, and notes into one place and turns them into reports, presentations, and diagrams. Local AI, built on Microsoft Foundry Local, lets qualifying Windows and Mac devices download a compatible model and then run summarizing, drafting, and document analysis entirely on the device, storage included, with no internet connection required once the model is downloaded. Bring Your Own Model support lets an organization point the editor at any OpenAI-compatible AI service, including a self-hosted one, so a company already paying for OpenAI, Gemini, or Claude access can route document AI tasks through that existing subscription instead of Foxit's.
Evan Reiss, Foxit's SVP of Marketing, described the release as uniting "the productivity of PDF Editor, the security of Local AI, and the power of content creation in Workspace." The more telling detail is in Foxit's own documentation: BYOM requests consume no Foxit AI credits and send no data to Foxit at all. That is an unusual thing for a vendor to build into its own product, since it means the feature exists specifically to route around the company offering it.
Why local processing is the real story
Chat-with-your-PDF features have been standard for two years now, and the pitch has always been the same: upload the file, ask a question, get an answer, trust that the vendor's cloud handled your document responsibly in between. Local AI is Foxit betting that a meaningful number of users, particularly ones handling contracts, medical records, or anything under a client confidentiality obligation, do not want to make that trust decision at all. Running the model on-device removes the step where the document leaves the machine, which is a different guarantee than a privacy policy promising deletion after processing.
It is worth being precise about what "on-device" means here. The model itself, delivered through Microsoft Foundry Local, still has to be downloaded and run locally, which means qualifying hardware and enough disk space and memory to host an LLM. This is not a lightweight toggle. It is a genuine local inference workload, and Foxit has not published detailed minimum specs beyond "qualifying Windows and Mac devices," so what actually qualifies is worth checking before you plan around it.
The catch with Bring Your Own Model
BYOM solves a different problem than Local AI does. It does not keep your document off the network, since a BYOM request still goes to whichever cloud API you pointed it at. What it removes is Foxit as an intermediary and Foxit's credit metering as a cost. For an organization that has already vetted OpenAI or Google's enterprise data handling terms and does not want a second vendor's terms layered on top for the same document, that is a real simplification, not just a pricing lever.
The two features answer different questions. Local AI is for "I don't want this document leaving my machine." BYOM is for "I don't want a second company touching this document on top of the one I already vetted." Conflating them, which the marketing around this release tends to do, means picking the wrong one for what you actually need.
What to actually do with this
If document confidentiality is the driver, test Local AI's actual hardware requirements against your fleet before promising it to a compliance team, since "qualifying device" is doing a lot of work in that sentence. If cost or vendor sprawl is the driver, BYOM is the more directly useful feature, and it is worth checking whether your existing OpenAI or Gemini enterprise agreement already covers the usage before adding another line item.
Either way, the underlying shift is one worth watching past this single release: on-device AI processing for documents stopped being a research demo and became a shipping feature in mainstream software this month. Tools that never send a document anywhere in the first place, including browser-based editors like Docento.app that process files client-side, have been making a version of this same argument for longer. Foxit adding a local-inference option is a sign that argument is starting to win.
