criptogus/HermesOffice

HermesOffice – an open‑source AI‑powered office suite

What it is – HermesOffice is a cross‑platform (macOS, Windows, Linux) office suite that can open, edit and save the native Microsoft Office formats (.docx, .xlsx, .pptx) plus PDF and Markdown. The distinguishing claim is that AI‑driven editing is built into each editor rather than being tacked on as a separate chat window.

Key components

• Docs – a Word‑compatible editor that preserves every untouched byte of the original file, so layout never changes unless you edit it. It shows paginated pages that match Word’s pagination.
• Sheets – a spreadsheet built on the open‑source Univer UI core, with an in‑house Rust side‑car (calamine + IronCalc) handling .xlsx parsing, charts, pivot tables, slicers and conditional formatting.
• Slides – a PowerPoint‑compatible presenter with its own .pptx parser/render engine, supporting masters, layouts, smart guides and non‑destructive image cropping.
• PDF – a true PDF editor (based on pdf.js, pdf‑lib and PDFium compiled to WebAssembly) that lets you re‑type text, edit images and preserve original fonts. It also converts PDFs locally to Word, Excel or PowerPoint, using OCR on scanned pages via the OS‑provided OCR service.
• Markdown – a lightweight Tiptap/ProseMirror block editor that works directly on plain .md files.
• Shell – the suite’s launcher that hosts the five editors in tabs, handles theming (light/dark/system) and auto‑updates.

AI integration

• Every editor embeds an AI panel that can perform block‑level edits, show snapshots and diffs, and call tools.
• The AI runs through a bring‑your‑own‑key (BYOK) model provider: you can plug in API keys for Claude, OpenAI, Gemini, DeepSeek, Kimi, GLM, Qwen, Doubao, MiniMax, Grok, Mistral, OpenRouter, or any OpenAI‑compatible endpoint (including local servers).
• Alternatively, you can sign in to Genspark (device‑code flow) and let the service proxy calls to Claude, GPT or Gemini and expose extra tool endpoints (web/image search, image generation, media analysis, transcription).
• The agent core (packages/agent-core) orchestrates tool calls, and the packages/ai-search module implements the Genspark‑specific tools.

Installation

• macOS (Apple Silicon) – pre‑built signed .dmg (ad‑hoc) or .zip from the latest release.
• Other platforms – source builds are provided; the README lists the npm commands (npm run dist:win, npm run dist:linux) that produce Windows installers, Linux AppImages, .deb and .rpm packages.
• Linux packages require glibc ≥ 2.34 and, for the AppImage, FUSE 2 runtime.

Development & architecture

• The project is a monorepo of pure TypeScript packages (no Electron dependency inside the engines) and a few Rust tools for the Excel side‑car.
• Core engines (docx-engine, pptx-engine, pdf2docx, etc.) parse the original OOXML/PDF files, build a block tree, and on save only splice back the edited blocks, keeping the rest of the archive byte‑for‑byte.
• UI is built with React, shared design tokens (packages/ui), and Electron for the desktop host.
• Tests run with Jest for the engines and Playwright + Electron for UI acceptance.

License & community

• Apache‑2.0 license, fully free and open‑source.
• Official builds send optional, anonymised usage analytics (no document content) that can be disabled in Settings.
• Security policy, privacy statement and third‑party notices are provided in the repo.

Bottom line – HermesOffice is a genuine, ambitious project that combines a full office‑suite feature set with locally‑run AI assistance, supporting a wide range of model providers and keeping the core document fidelity that professional users expect.