CatCatUncle/openworkbuddy

本机优先的 AI 办公工作台 · A local-first AI office workbench. Turn requests and source material into editable PPTX, DOCX, XLSX and HTML files. Choose your own models, Skills and MCP tools; inspect execution and results.

What it solves

OpenWorkBuddy addresses the tedious manual labor of creating office documents like PPTs, reports, and spreadsheets. Instead of just chatting with an AI, this tool acts as an autonomous agent that plans, executes, and verifies the creation of actual, editable files that you can open and use immediately.

How it works

It functions as a local-first AI workspace. You provide a task, and the agent uses a model of your choice (e.g., DeepSeek, Claude, Ollama) to plan the steps, call necessary tools, and generate the final document. It features a "result area" where you can inspect and download the generated files. The system is designed to be extensible—you can add new capabilities simply by creating a Markdown file in the skills directory.

Who it’s for

It is designed for professionals who frequently create reports, presentations, and documents, as well as teams that require a self-hosted, private AI assistant. It is also suitable for developers interested in understanding how AI agent frameworks, tool calling, and local file management are architected.

Highlights

  • Real file generation: Produces usable PPTX, DOCX, XLSX, and HTML files rather than just text responses.
  • Model flexibility: Supports various providers including DeepSeek, Kimi, Ollama, and local engines like Claude Code or Codex.
  • Local-first & Private: Data stays on your machine by default, with options for secure team-based self-hosting.
  • Extensible architecture: Add new agent capabilities by dropping a Markdown file into the project without needing to recompile or restart.
  • Command-line support: Includes a CLI that shares the same configuration and context as the desktop app for seamless workflow integration.

Related

  • Project
    Apeireth/Apeireth

    Apeireth is a privacy‑first, on‑device AI companion written in safe Rust. It stores an append‑only, hash‑chained memory of user facts, offers a rule‑engine‑driven “care” mode, and enforces strict human approval for any irreversible action. The project provides a desktop UI (Tauri), a terminal UI, and a CLI, and supports three LLM providers (OpenAI‑compatible, MiniMax, Anthropic). All core features run locally on Windows, Linux and macOS; missing pieces (WebSocket server, P2P mesh, full SDK transport) are explicitly listed as “honest gaps”.

  • Project
    AnnenkovLabs/girl-agent

    A state-driven AI persona engine for Telegram that simulates human behavior through sleep schedules, relationship stages, and long-term memory to avoid typical chatbot patterns.

  • Project
    MemTensor/MemOS-Cloud-OpenClaw-Plugin

    MemOS Cloud Host Plugins provide a bridge for OpenClaw and DeepSeek Harness to automatically retrieve and store AI interaction memories in the cloud.

  • Project
    billion-token-one-task/Deepgraph

    An autonomous research agent that harvests scientific literature, maps evidence into a graph, and runs experiments to find and verify checkable research conclusions.