DEV Community

Hasan Yahşi
Hasan Yahşi

Posted on AI-assisted

FieldDesk: a portable Windows workspace for tools and documents

FieldDesk started with a practical goal: keep a useful Windows toolkit on a drive and open it through one interface. We're sharing an MIT-licensed preview and looking for feedback on the first-use experience.

The workspace brings together an app catalog, PDF/image workflows, local document search and optional AI chat. The Windows x64 launcher is about 23 MB, including embedded Python and document dependencies. Programs and model weights are separate downloads.

FieldDesk: app catalog, PDF workflow and local note search

Repository and download · 37-second real app demo

Three small tasks to try

  1. Prepare a tool. Open Apps → Get apps, choose Notepad++ and download it. The 30-package catalog checks pinned download sizes and SHA-256 values before preparing supported archives. Five entries are installer downloads only; installers are not executed automatically.
  2. Create a document handover. Select two sample PDFs, review their order and page count, then merge them. The output goes into a fresh task folder and source documents are preserved.
  3. Find a note locally. Put a text note under the workspace's 04_PROJELER folder, update the Library index and search its contents. This step does not need a model or cloud account.

The first-run guide walks through these steps with disposable data. English and Turkish interfaces are available. The recorded demo uses fictional documents and actual publisher downloads, PDF processing and search.

A few design choices

FieldDesk serves its browser interface on loopback. File workflows and indexing run locally; cloud chat is optional and uses the user's provider credentials. The catalog is pinned rather than silently installing whatever version a publisher releases next.

The workspace stores tools and documents in folders, but portability has limits: third-party prerequisites and host policies still apply. Optional Windows DPAPI credential storage is tied to the Windows user and computer. The preview is not code-signed.

Local AI chat requires a compatible llama.cpp runtime and supported GGUF model. OCR needs Tesseract and language data. Neither is included in the small launcher download. The assistant is a chat tool, not an autonomous computer-control agent.

What feedback would help?

Can you open the preview, prepare one tool and complete the document workflow on your Windows 10/11 machine? Where do you get stuck? Which task would you use again?

We have an initial feedback thread and two scoped contributor tasks: complete the English PDF messages and announce search completion to screen readers. The translation gap is visible in the current preview; it is not being presented as finished.

Reports using sample files are welcome in English or Turkish. Please keep credentials and private document text out of public comments.

Disclosure: This article was prepared using OpenAI Codex at the project owner’s direction. The English text and publication were handled by Codex. FieldDesk was also developed with Codex assistance. The download, PDF and search steps described above were exercised in the recorded sample workspace; this is not a claim of testing every Windows configuration or of independent security review.

Top comments (0)