neuron-explorer
Qwen3.5‑4B 전용 조향형 뉴런 탐색기, 모든 MLP 뉴런을 밀어 결과를 보관·탐색
Plan, build, and deliver with an open-source, self-hostable AI workspace for coding, collaboration, and automation.
저장소 README에서 가져옴
README 전체를 불러오는 중입니다. 아래는 저장된 발췌입니다.
An open-source AI work system spanning your local desktop, cloud agents, and remote machines.
Wegent includes a desktop workbench and a self-hostable web platform. Wegent Desktop works with local projects, files, commands, tests, and code changes. Wegent Web provides browser-based agents, knowledge, automation, and administration. Wegent Backend connects these applications to shared project spaces, models, and execution devices.
Download the latest Wegent Desktop · Documentation · ContributingREADME 전문 보기 ↗An open-source AI work system spanning your local desktop, cloud agents, and remote machines.
English | 简体中文
Wegent includes a desktop workbench and a self-hostable web platform. Wegent Desktop works with local projects, files, commands, tests, and code changes. Wegent Web provides browser-based agents, knowledge, automation, and administration. Wegent Backend connects these applications to shared project spaces, models, and execution devices.
Download the latest Wegent Desktop · Documentation · Contributing
Wegent Desktop organizes local coding work by project and task. The task view keeps the conversation, tool activity, tests, changed files, and diffs together; the project-space view provides the board, shared files, automation, and execution status.
The local coding experience built on Codex is the foundation of Wegent Desktop. It works directly with the code, files, commands, and development environment in your projects. Connect it to Wegent Backend when coding work needs to move beyond one computer or one person. Once connected, the desktop workbench and Web use the same projects, tasks, and execution devices.
If work always stays on one computer, Wegent Desktop provides a familiar Codex coding experience. Connect it to Wegent to extend that experience when tasks must move across people, devices, or a self-hosted environment.
Wegent Web provides the browser interface for remote agents. Agents can use configured models, knowledge, Skills, and tools while Wegent Backend manages their tasks and execution.
Wework is the Wegent desktop workbench. Electron owns desktop windows, system capabilities, and process management. DeepSeek Harness (DSH) is the embedded application and plugin runtime, and the Wework product UI is composed from a set of DSH plugins. Executor manages tasks and sessions, then drives Codex to perform the actual work in local projects.
flowchart LR
Electron["Electron<br/>Desktop host"]
subgraph DSH["DSH Runtime"]
direction TB
Core["DSH Core<br/>Plugin discovery, dependency injection, and lifecycle"]
App["dsh-app-wework<br/>Wework product shell and UI extension points"]
UI["dsh-ui-*<br/>Tasks, project spaces, settings, apps, and automation"]
ElectronPlugin["dsh-electron-host<br/>Electron capability bridge"]
ExecutorPlugin["dsh-executor-runtime<br/>Task execution and event adapter"]
TerminalPlugin["dsh-terminal-runtime<br/>Local interactive terminals"]
Core --> App
Core --> UI
Core --> ElectronPlugin
Core --> ExecutorPlugin
Core --> TerminalPlugin
App --> UI
end
Executor["Executor<br/>Task and session execution"]
Codex["Codex<br/>Coding agent"]
Workspace["Local project<br/>Files and commands"]
Electron -->|"Starts and hosts"| Core
Electron -->|"Starts and supervises"| Executor
ElectronPlugin <-->|"Restricted desktop capabilities"| Electron
UI <-->|"User actions and UI updates"| ExecutorPlugin
ExecutorPlugin -->|"Create, follow up, approve, cancel"| Executor
Executor -->|"Status, messages, tool events, results"| ExecutorPlugin
Executor <-->|"Launch, control, and events"| Codex
Codex <-->|"Read, modify, and execute"| Workspace
TerminalPlugin <-->|"PTY"| Workspace
DSH does not execute coding tasks itself. dsh-app-wework defines the product shell and extension points, dsh-ui-* provides the product modules, dsh-electron-host bridges restricted Electron capabilities, dsh-executor-runtime communicates bidirectionally with Executor throughout execution, and dsh-terminal-runtime manages local interactive terminals. Executor is the task execution layer, while Codex is the concrete coding agent it drives.
Wework can also connect to Wegent Backend for shared project spaces, models, and remote execution devices. Wegent Web is a separate browser interface built on the same Backend.
To start the self-hosted web application and Backend in standalone mode with Docker:
curl -fsSL https://raw.githubusercontent.com/wecode-ai/Wegent/main/install.sh | bash -s -- --standalone
Open http://localhost:3000 and follow the setup flow to create the administrator password and configure a model. See the Installation Guide and Standalone Mode for other deployment options.
Requires Node.js 22.15+ (22.x) or 23.8+ (Node.js 24 recommended), and pnpm:
pnpm install
pnpm --filter wework dev
To run the macOS desktop app:
pnpm --filter wework dev:mac
See wework/README.md for desktop development, build, and release instructions. wework/ is the current desktop application source directory.
| Directory | Responsibility |
|---|---|
wework/ | Wegent Desktop (Electron, Vite, React) |
wework-mobile/ | Independent Expo application for iOS and Android |
frontend/ | Wegent Web (Next.js, React) |
backend/ | Default public API, business logic, and data models |
backend-rs/ | Rust hybrid gateway and APIs that have completed migration |
executor/ | Rust task executor, local-device runtime, and Codex integration |
executor_manager/ | Server-side executor scheduling, sandboxes, and orchestration |
chat_shell/ | Lightweight chat and agent runtime |
knowledge_runtime/ | Knowledge indexing and retrieval API |
knowledge_doc_converter/ | Document parsing and asynchronous conversion service |
| Directory | Responsibility |
|---|---|
knowledge_engine/ | Reusable Python kernel for parsing, indexing, vector stores, and retrieval |
shared/ | Python models, protocols, utilities, and telemetry shared by services |
packages/ | TypeScript chat and collaboration modules shared by Web and Desktop |
sdk/ | Plugin authentication, build, creation, and DingTalk workspace SDKs |
wegent-cli/ | Python CLI for managing Wegent CRD resources |
| Directory | Responsibility |
|---|---|
docker/ | Service images, standalone resources, and the docker/telemetry/ observability stack |
scripts/ | Installation, release, acceptance, repository-policy, and development automation |
tests/ | Cross-service Playwright integration tests and installation tests |
docs/ | English and Chinese user docs, developer docs, architecture, and historical designs |
patches/ | Third-party dependency patches used by pnpm patchedDependencies |
LICENSES/ | Third-party license texts required for repository distribution |
.github/ | CI, release workflows, composite actions, and repository automation |
| File | Responsibility |
|---|---|
start.sh, install.sh | Local startup and self-hosted installation |
build_image.sh, build_image_mac.sh | Multi-architecture and host Docker image builds |
docker-compose*.yml | Default, build, and E2E Compose orchestration |
package.json, pnpm-workspace.yaml | JavaScript/TypeScript workspace definition |
pyproject.toml, uv.lock | Repository-level Python tooling and lock metadata |
AGENTS.md, CLAUDE.md | Contributor rules and compatibility symlink |
Bug reports, documentation improvements, code contributions, and new ways of using Wegent are all welcome.
190890002451Thanks to everyone who helps Wegent grow.