WeSight: Desktop AI Agent Workspace for Local CLIs

TL;DR

TL;DR: WeSight is an MIT-licensed Electron desktop workspace that installs, detects, and reuses local AI coding agent CLIs such as Claude Code, Codex, Kimi Code, OpenClaw, Hermes Agent, OpenCode, Qwen Code, and DeepSeek-TUI, and gives them a visual chat, model router, skill marketplace, and runtime metrics dashboard.

Source and Accuracy Notes

• Project page: wesight.ai
• Source repository: github.com/freestylefly/wesight
• License: MIT (verified via GitHub API license.spdx_id)
• Platform: macOS Apple Silicon and Intel builds via GitHub Releases
• Source last checked: 2026-06-16

What Is WeSight?

WeSight is a desktop control console for local AI coding agents. Terminal-native agents are powerful, but their setup, model routing, permissions, IM entry points, file changes, and runtime metrics usually live in separate tools. WeSight pulls those into one workspace: install or detect a local agent CLI, chat with it through a visual surface, route it to a model provider, follow its file changes live, and inspect its runtime metrics.

The current set of supported engines is Claude Code, Codex, Kimi Code, OpenClaw, Hermes Agent, OpenCode, Qwen Code, DeepSeek-TUI, and the WeSight built-in runtime. Engines can be installed through the UI or reused from an existing local CLI config.

Repo-Specific Setup Workflow

Step 1: Download

Download the matching DMG from the latest release:

• WeSight-*-mac-arm64.dmg for Apple Silicon
• WeSight-*-mac-x64.dmg for Intel

Open the DMG and drag WeSight.app into Applications.

Step 2: Handle the unsigned app warning

Preview builds are not signed and notarized yet. macOS may show a “WeSight.app is damaged and cannot be opened” message. The README explicitly explains this is a Gatekeeper quarantine warning on an unsigned app, not a corrupted download.

xattr -cr /Applications/WeSight.app

After the command finishes, open WeSight again.

Step 3: Develop from source

Requirements: Node.js >=24 <25 and npm.

git clone https://github.com/freestylefly/wesight.git
cd wesight
npm install
npm run electron:dev

The Vite dev server runs at http://localhost:5175.

Step 4: Add a model provider

The Model Providers section supports official OpenAI, Anthropic Claude, and Google Gemini providers, plus OpenAI-compatible providers for DeepSeek, Qwen, Moonshot, Ollama, OpenRouter, GitHub Copilot, local gateways, and private endpoints. WeSight can either manage the model settings itself or import an existing local engine configuration and sync it.

Deeper Analysis

The engines table

The README’s engines table is a useful summary of the design. The built-in runtime is the general-purpose cowork session and skill runtime. Claude Code, Codex, Kimi Code, OpenClaw, Hermes Agent, OpenCode, Qwen Code, and DeepSeek-TUI each have a one-click install path and a “use my existing local CLI config” path. WeSight is not trying to replace any of these engines. It is trying to be the desktop front-end for all of them.

The runtime dashboard

The AI Runtime Dashboard surfaces engine, model, tokens, TTFT (time to first token), output-phase TPS (tokens per second), estimated model TPS, cost, status, and per-step tool latency. The Live Workspace view shows file writes, code changes, tool activity, and generated artifacts in real time as the agent works. The combination is what most agent users end up scripting together by hand.

The skill and pet layers

SkillHub is a skills marketplace inside WeSight. Categories include browser, dev tools, productivity, and design, and the marketplace supports install and lifecycle management of skills. The Studio tab gives a visual office-style workspace and a desktop pet that follows active agent tasks. These are the productivity features that the README explicitly positions as the bridge between terminal agents and a more user-friendly desktop experience.

IM channels

WeSight can connect agent tasks to IM channels, with Feishu as the canonical example. Each engine has its own IM configuration, so the same agent task can be picked up from chat, from the desktop workspace, or from a Feishu thread without losing state.

Practical Evaluation Checklist

• [ ] Are you on macOS (Apple Silicon or Intel) and willing to use an unsigned preview build?
• [ ] Do you already use one or more of Claude Code, Codex, Kimi Code, OpenClaw, Hermes Agent, OpenCode, Qwen Code, or DeepSeek-TUI?
• [ ] Do you want a single UI to install, configure, and switch between those agents?
• [ ] Do you need a runtime dashboard with TTFT, TPS, token, and cost metrics?
• [ ] Do you want to follow agent file changes and tool activity live in the workspace?
• [ ] Do you use Feishu or another supported IM channel for agent entry points?
• [ ] Will you use the SkillHub skills marketplace, or do you keep your own skill folder?

Security Notes

The preview builds are not signed and notarized, so macOS will show the “WeSight.app is damaged” warning until you run xattr -cr. Treat the build as an unsigned developer preview. If you store API keys in WeSight’s model provider settings, the keys live in the app’s local data directory; the README does not document cloud sync, so a default install keeps everything on disk.

The agent engines themselves are local CLIs, so the model provider keys are passed through to whichever engine WeSight is front-ending. If you connect Feishu as an IM channel, follow your normal Feishu bot security model and keep the bot scoped to the channels it needs.

FAQ

Q: Is WeSight available on Windows or Linux? A: The README explicitly says early public releases ship macOS Apple Silicon and Intel builds. The project roadmap page is on GitHub Projects, but the README does not promise a Windows or Linux timeline.

Q: Does WeSight replace Claude Code or Codex? A: No. WeSight installs, detects, or reuses those CLIs and gives them a desktop front-end. The engines do the work; WeSight is the workspace.

Q: Why does macOS say WeSight is damaged? A: The README explains this is a Gatekeeper quarantine warning on an unsigned app, not a corrupted download. Run xattr -cr /Applications/WeSight.app to clear the quarantine attribute.

Q: Can I use a local model with WeSight? A: Yes. The Model Providers section accepts any OpenAI-compatible endpoint, including Ollama, local gateways, and private endpoints. You can add the provider once and route any engine through it.

Q: Does WeSight sync with my existing Claude Code or Codex config? A: Yes. The Model Providers section can import and sync an existing local engine configuration when you want WeSight to manage it, or you can keep the engine’s local config untouched and use WeSight as a thin client.

Conclusion

WeSight is a useful desktop workspace for developers who already use a local AI agent CLI and want one place to manage setup, models, runtime metrics, IM entry points, and skills. It does not try to replace the engines. It tries to be the desktop front-end for all of them. If you are happy in the terminal, you do not need it. If you want a visual chat, a model router, and a runtime dashboard over the same engines, it is a real productivity gain.

Related reading: GitHub Trending tools, Developer tools, tinycld, Runtime, FanBox.