Desktop agent platform for customizable localization workflows
stardust, by Tianzeyui, is an open-source desktop AI agent platform for automating model-augmented workflows across local projects. The app combines a chat interface with autonomous agents that can delegate tasks, interact with files and APIs, and produce context-aware text outputs for localization and coding workflows. It exposes plugin hooks and a gateway to external model contexts, enabling custom React/TSX extensions and sandboxed code execution. The tool targets AI developers, localization specialists, and power users who need a programmable desktop environment for project-focused agent work.
Model choice determines the tool's output behavior
The app supports over 60 AI models, including OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Qwen, and Llama. That model variety lets teams match providers to tasks such as translation, summarization, or code generation, and it creates observable differences in tone, factuality, and hallucination risk. Users must therefore evaluate outputs per model and align selections with the accuracy needs of each localization or development task.
Workspace persistence reduces manual context reassembly for long jobs
The tool provides independent workspaces with dedicated file trees and persistent contexts, which keeps project state across sessions for longer localization or coding jobs. Built-in converters turn PDF, DOCX, PPTX, and XLSX into Markdown for model consumption, reducing manual reformatting. Persistent state supports agent chains that reference past files, but final outputs still depend on chosen model behavior and should be validated in downstream review steps.
Local development requires current Node tooling and a desktop environment
The tool is delivered as an Electron application built with React, intended for desktop use on macOS, Windows, and Linux. Building or extending the platform locally requires Node.js 20 or newer, so authors must work in a local Node environment to develop and test extensions. This setup favors developers familiar with JavaScript and TypeScript rather than non-technical, plug-and-play users.
Sync and extension paths support collaborative, source-accessible workflows
The app offers optional Supabase integration to synchronize conversations, project contexts, and persistent memory across devices when enabled. Plugin authors gain access to over 60 host APIs and the React ecosystem, and the interface is built with Tailwind and shadcn/ui. The developer maintains the project openly on GitHub, which supports community contributions and visible extension examples for teams that prefer source-accessible platforms.
Best for teams that can invest in technical setup and verification
For teams able to absorb a developer-focused setup and provide ongoing model oversight, the platform is a practical choice that delivers configurable automation and project continuity. Expect upfront work to instrument and test extensions, and plan human verification for any localization before publication. Tip: run small pilot tasks to evaluate model behavior and tune prompts before broader rollout to reduce output risk.
