After 30 Days of Work, WeSight Is Now Open Source: One Entry Point for All Your Agents

Hi everyone, I’m Canghe.

After my product WeSight launched on April 8th, the invite codes were being scalped for 999 RMB each.

While it feels a bit silly, it showed just how many people’s pain points WeSight was hitting at the time.

But it wasn’t “above board” — it couldn’t “stand in the light.” So I voluntarily shut down the project.

After a month of careful thought, I decided to restart WeSight on May 1st, giving it the chance to meet everyone properly.

After more than a month of development, today I want to give WeSight to everyone as a Children’s Day gift.

WeSight is now officially open source, with brand new features and a brand new entry point.

The official website is still the same one, but it now has a new identity.

WeSight is now an open-source desktop AI Agent console. It can install or reuse Claude Code, Codex, OpenClaw, Hermes Agent, OpenCode, Qwen Code, DeepSeek-TUI, and the built-in Agent runtime,

unifying them into a visual workspace that covers chat, tools, files, IM channels, skills, model providers, run monitoring, and desktop pet workflows.

One entry point, manage all your agents.

You can hand over your Claude Code, Codex, OpenClaw, Hermes Agent, OpenCode, Qwen Code to WeSight, and it will handle unified scheduling and management for you.

WeSight uses Claude Code, Codex, OpenClaw, Hermes Agent, OpenCode, etc., as built-in engines, fully proxying your local agents.

It can automatically detect your local agent environment, and also install Claude Code, Codex, etc., with one click.

You don’t even need to open Claude Code or Codex in a terminal, and you don’t need to install another cc-switch to configure models. WeSight can monitor and configure local models for you.

You just need to create a new task, select the engine and model you want, and WeSight will handle everything.

Especially the ability to switch models with one click — I find it really smooth. No more opening a new app to switch models every time.

When you hand over the model selection for your local agents to WeSight for unified management, you’ll find it much more seamless. One model can be adapted to multiple different agents, and you don’t have to worry about how to configure third-party models for your Claude Code and Codex.

Previously, I wrote many tutorials about multi-agent solutions with OpenClaw or Hermes Agent. Even though I provided guides, many people still found it troublesome.

A bunch of configurations, a bunch of steps — it’s not something ordinary people can quickly get started with.

So I defined an upper-layer Harness in WeSight, where you can choose different agents or an Agent Team.

Take a development team for example: you can add different agents to this team, each configured with a different engine and model. For instance, an agent powered by the Claude Code engine acts as the product manager, while an agent powered by the Codex engine acts as the development engineer.

You can quickly pull this Agent Team into a task. The Agent Team will collaborate in a way that mimics a human team — they can communicate with each other, but their contexts are isolated.

Now, controlling your local Claude Code, Codex, and other agents from Feishu (Lark) becomes extremely simple. You just need to complete the binding in WeSight.

I designed two logics here. One is to fully hand over management to WeSight — your Claude Code conversations in Feishu will directly show up in WeSight.

However, this is affected by the WeSight lifecycle. For example, when WeSight is closed, you won’t be able to control it via Feishu. This can be inconvenient for agent runtimes like OpenClaw and Hermes.

Therefore, WeSight also fully supports system users managing on their own. What does that mean? It means messages you send to OpenClaw in Feishu

can work normally even if WeSight is not running. WeSight is just your workspace, not a scheduler. Your data and usage rights remain on your local machine.

You can see in Feishu I’ve connected local OpenClaw, Hermes, Claude Code, and Codex simultaneously. They can be centrally scheduled by WeSight or work independently, but WeSight will help you configure everything.

Learning from Codex, I also added a desktop pet to WeSight — you can play with it as you like.

Regarding model tasks, we often wonder how many tokens are used, what the TPS is, how much context is used, and how long it takes.

So WeSight added run monitoring capabilities. It can perform system monitoring on all your local agent engines.

You can also monitor individual tasks:

Of course, for a single run, you can also see the skills, tools, file changes, and output files used.

Learning from QClaw, I added a workspace capability in each conversation — your agent can visually see themselves working in the office, with different agents having different effects.

Skills and MCP are nothing special — basically everything you’d expect is there.

Actually, there are many small, interesting details that you’ll probably need to explore on your own.

Honestly, WeSight right now is like a newborn baby.

It has a complete skeleton and a clear direction, but it’s not perfect.

Some features are still being polished, some experiences need optimization, and some edge cases aren’t covered yet.

But I believe the meaning of open source has always been about “letting more people define what ‘perfect’ means together,” not waiting until everything is flawless before showing up.

This past month of solo development has taught me one thing clearly:

One person can go fast, but a group of people can go far.

So today, I’m handing WeSight over to the community completely.

If you think this direction is valuable, please go to GitHub and give it a Star. For an independent developer, that’s the biggest encouragement.

If you have ideas or skills, you’re more than welcome to contribute — submit Issues, submit PRs, even a small typo fix is a step forward for this project.

Finally, a special thanks to the Youdao Lobster open source project.

The overall framework of WeSight is built on Youdao Lobster. Standing on the shoulders of giants allows us to see farther. The best part of the open source world is this: the previous generation planted trees, the next generation enjoys the shade, and then the next generation can make the trees even bigger.

I always believe in one sentence:

A good tool lets you think about one less thing.

What WeSight wants to do is free you from worrying about which terminal to open, which model to configure, or how to get multiple agents to collaborate.

You just focus on your ideas and creativity. The rest, leave to WeSight.

The road ahead is still long, but at least the first step has been taken.

Thank you to everyone willing to believe in this direction. Here is WeSight’s new logo — I hope you like it too.