Paprwork V2: A Locally-First AI Desktop Assistant Refactored with TypeScript and Mastra

Paprwork V2 is a locally-first AI desktop assistant application, completely rewritten using TypeScript and the Mastra framework. It addresses the technical debt of the V1 version and builds a production-ready architecture. Core features include support for multiple model providers, an extensible tool system, sub-agents, scheduled tasks, an intelligent memory system, etc. It balances privacy and functional scalability, making it suitable for AI interaction scenarios that require local data processing.

The V1 version had severe technical debt: a single file with over 30,000 lines of code, 90% code duplication across processes, tool calls requiring multiple patches, and a lack of type safety. These issues hindered long-term maintenance, so V2 was completely rewritten with the goal of building a maintainable and extensible codebase.

V2 adopts a modular layered architecture. The core directory core/ serves as a shared library to eliminate code duplication, and all components are 100% TypeScript with no any types. The Mastra framework is introduced as the AI orchestration layer to solve the fragility of tool calls, providing a unified Agent abstraction, robust tool calls, and streaming processing capabilities. The architecture is inspired by OpenClaw but adjusted for desktop environments (e.g., Electron+React UI).

V2's core features include: multi-model support (Claude/GPT/Gemini), streaming responses, extensible tools (Bash/file operations, etc.), parallel sessions, scheduled task automation, intelligent memory, Mini-Apps, sub-agents, and secure key management. Regarding privacy: telemetry consists of anonymous rough events (can be turned off), users can customize telemetry endpoints, and API keys are stored encrypted.

V2 uses Rust tools to improve development efficiency: oxlint replaces ESLint (50-100 times faster), and oxfmt replaces Prettier (30 times faster). Development supports hot reloading, and the testing system is comprehensive: unit/integration/E2E tests, type checks, formatting checks, code line limits, all of which can be run with one click via npm run check.

Application scenarios include personal knowledge management, automated workflows, multi-model comparison, and privacy-first AI interactions. Quick start requires Node.js v24+ and npm v10+: clone the repository → install dependencies → configure .env.local → start the application (npm start).

Paprwork V2 has achieved a transition from prototype to product, resolving the technical debt of V1, introducing a modern tech stack (TypeScript/Mastra/Rust tools) and strict standards, and building a maintainable and extensible architecture. It has reference value for both users of locally-first AI assistants and developers involved in refactoring.