OpenYak: Open-Source Desktop AI Agent for Local File Management and Automation

What OpenYak Does
OpenYak is an open-source desktop AI assistant that runs entirely on your machine. It manages files, analyzes data, drafts documents, and automates office workflows without uploading anything to the cloud.
Key Features and Capabilities
- Office Automation – Batch rename, sort, and clean up files across folders with auditable change logs
- Data Analysis – Parse spreadsheets, CSVs, and documents locally to spot anomalies, track trends, and export decision-ready reports
- Content & Writing – Turn rough notes into polished documents and draft professional communications
- Team Operations – Merge notes from PDFs, DOCX, and CSVs into structured briefings with auto-extracted action items and deadline tracking
- IM Integration – Connect 8+ messaging platforms (WhatsApp, Discord, Telegram, Slack, Feishu, Signal, iMessage) through OpenClaw for centralized message handling
- Automations – Schedule recurring tasks with cron-based triggers for daily data digests, weekly reports, and automated file cleanup
- Remote Workflows – Start on desktop and continue on phone with one-click secure tunnel and QR code access
Technical Specifications
- 100% Local-First – No cloud storage or telemetry; all data stays on your machine
- 100+ AI Models – Access through OpenRouter including Claude Opus 4.6 & Sonnet 4.6, DeepSeek V3.2, Gemini 3 Flash, GPT-4.1, MiniMax M2.5, Step 3.5 Flash, MiMo-V2-Pro
- 20+ BYOK Providers – Bring your own API key for OpenAI, Anthropic, Google, Groq, DeepSeek, Mistral, xAI, plus Chinese providers like Qwen, Kimi, MiniMax, and ZhiPu
- ChatGPT Subscription Integration – Connect existing ChatGPT subscription directly with no extra API costs
- 20+ Built-in Tools – File read/write/edit, bash execution, glob/grep search, web fetch, long-term memory
- 7 Agent Modes – Specialized agents for building, planning, exploring with multi-step tool calling and sub-agent nesting
- Long-term Memory – Stores preferences, work context, and key facts across conversations locally
- MCP Connectors – Integrate external tools and services via Model Context Protocol
- Pricing – 1M tokens/week on free models, pay-as-you-go at OpenRouter prices with zero markup, or bring your own API key
Getting Started
- Download the installer for Windows or macOS
- Connect a model – use free models instantly, top up for premium models, bring your own API key from 20+ providers, or run locally via Ollama
- Start working – manage files, analyze local data, and generate office-ready outputs
For developers, technical details, project structure, and development setup are available in frontend/README.md and backend/README.md. The project is licensed under AGPL-3.0 and built with Python (60.2%), TypeScript (31.7%), HTML (5.4%), Rust (1.5%), CSS (0.5%), and Shell (0.4%).
📖 Read the full source: HN AI Agents
👀 See Also

Keyoku Plugin Replaces OpenClaw's Static Heartbeat with Memory-Driven Autonomy
Keyoku is a free OpenClaw plugin that changes the agent's heartbeat from reading a static HEARTBEAT.md file to scanning the agent's actual memory store for stalled work, dropped commitments, conflicting information, and quiet relationships. It uses a local Go engine with SQLite + HNSW and offers three autonomy levels: observe, suggest, and act.

Setting Up OpenClaw as an Always-On AI Assistant
OpenClaw, configured as an always-on AI assistant for a small dev team, is set up on a Railway server with Claude as the backend and integrates with Google Workspace, GitHub, and more.

Zeude: Self-Hosted Monitoring Dashboard for Claude Code and OpenAI Codex
Zeude is a self-hosted dashboard that tracks Claude Code and OpenAI Codex usage, providing per-prompt token and cost breakdowns, weekly leaderboards, and team skill management. Version 1.0.0 adds Windows support, Codex integration, and per-user skill opt-out.

Socratic Prompt Generator Built as React Artifact Inside Claude
A developer built a Socratic prompt generator as a React artifact that runs inside Claude, featuring auto-detection of input complexity and three-tier prompt generation with failure mode analysis.