Wegent
Desktop drives Codex locally; web adds remote agents, knowledge, and shared team projects.
What it is
Wegent is an open-source AI work system from the wecode-ai team that brings the local desktop, cloud Agent, and remote machines into one shared project space. The desktop side, Wegent Desktop, is a local coding workbench with Codex as its execution core; the web side, Wegent Web, provides remote Agents, knowledge retrieval, automation, and admin management in the browser; the Backend connects both ends to shared project spaces, models, and execution devices.
How to run it
For the desktop, download the installer from the releases page, open it, pick a local project, create a task and describe the work to be done, then review Agent activity, command output, and change diffs in the same workbench. Self-hosted Web and Backend start with a single Docker command:
curl -fsSL https://raw.githubusercontent.com/wecode-ai/Wegent/main/install.sh | bash -s -- --standaloneAfter it starts, open http://localhost:3000 to go through the initialization flow, set an admin password, and configure models. To run the development environment from source, you need Node.js 20+ and pnpm:
pnpm install
pnpm --filter wework devCore mechanics
- Executor drives Codex: Executor is the task execution layer, handling task creation, session follow-up, and event dispatch; the one that actually writes code is Codex, reading and modifying files, commands, and the development environment in the local project directly.
- DSH plugin runtime: The desktop interface is assembled from a set of DeepSeek Harness plugins; DSH itself does not execute coding tasks, it only handles plugin discovery, dependency injection, lifecycle management, and bridging Electron's restricted capabilities and the local terminal.
- Multiple execution targets: The same project can run locally or be handed to a remote work machine or a server-hosted Executor; task boards, shared files, automation run records, and deliverables all converge in the project space.
- Web and desktop share the Backend: Remote Agents in the browser reuse the same models, knowledge, Skills, and tools, while the Backend centrally manages permissions, scheduling, plans, and knowledge services.
Who it is for
For people who write code on a single computer over the long term, the desktop is the familiar Codex coding experience, usable directly without deploying any service. For teams that want to extend coding work from a single machine to multi-person collaboration, self-host Web and Backend to collect tasks, files, discussions, and execution status in one place, and also set up automation queues for repetitive project work.
Limits
- Coding execution uses Codex as its core, so the final capability depends on the configured model and on Codex itself.
- Web and Backend are yours to operate, with Docker as the fastest path; permissions, scheduling, and knowledge services all fall within the self-hosted scope.
- The project is still under active iteration, so interfaces and versions will change; desktop installers are released per platform, as listed on the releases page.
- Open source and free, with no commercial support commitment; the community is on DingTalk and Discord.