昨天说了，接下来想多做一些有意思的事。

今天就来介绍一下新坑：Pero 🚀

一个面向 AI for Science（AI4S）的开源 Agentic Workspace。先从一个简单的场景学习起来，探索 AI Agent 如何参与科研工作流。

另外，还想尝试一种新的开源协作方式：不只是贡献代码，还可以贡献 Agent、Tokens 和算力。

比如，让你的 Agent 自动认领 GitHub Issue，在自己的机器上执行任务，再把结果贡献回来。

也算是给大家提供一种新的方式，来花掉用不完的 Tokens。😂

刚开坑，欢迎一起来折腾！

GitHub：https://github.com/phodal/pero

任务广场：https://app.openpero.com/tasks

phodal/pero

Source: https://github.com/phodal/pero

Pero

Composable research apps and distributed agent collaboration.

Download dev build · Roadmap · Development · Status · Contribute · 简体中文

Development preview. macOS arm64 builds are available. There is no stable release yet.

Why

A research workflow often starts as a script, notebook or model demo. Pero turns it into an App that can run on its own or alongside other tools in a shared workbench. Virtual Cell is the first example, with cell-atlas exploration and reaction simulation.

The longer-term goal is to connect these Apps with agents, collaborators and compute while keeping inputs, methods and results traceable.

Status

How it works

The intended workflow is to define a task with fixed inputs and checks, assign it to a collaborator or approved worker, then review the recorded result. Compute can stay on a personal machine, a lab server or the cloud; coordination can be self-hosted.

Agents propose work; people approve consequential actions and accepted conclusions. Task intake alone grants no execution permission, and runs retain the evidence needed for review.

Solid boxes mark implemented foundations; dashed boxes and links mark planned capabilities or integration. The end-to-end research workflow remains in progress. See the architecture and development guide for details.

Contribute

Bring a research use case, contribute a method or App, or review results. Pick a bounded item from the Roadmap; for substantial changes, open an Issue to agree on scope and acceptance criteria. Include a reproducible check with code changes. AI-assisted contributions follow the same review process. See CONTRIBUTING.