영어 |

WorkWit - 엔터프라이즈 AI 오피스 어시스턴트

기업 지향적이고 프라이빗 배포 가능한 AI 에이전트 플랫폼입니다. 백엔드는 FastAPI + SQLite로 구축되었으며, 프론트엔드는 의존성이 없는 바닐라 JS 단일 페이지 애플리케이션입니다. 전체 스택은 Docker Compose를 통해 한 명령어로 배포됩니다.

이는 단순한 ‘계약 검토 장치’ 그 이상으로, 다중 모델 추론, 확장 가능한 기술, 도구, MCP 커넥터, 샌드박스 실행, 다중 테넌트 접근 제어, 예약된 자동화, 장기 기억을 통합한 통합된 오피스-에이전트 기반입니다. 계약서 검토 및 이력서 심사 같은 사무실 시나리오는 기본 상태에서 작동하며, 새로운 기능을 "skills + tools"를 통해 지속적으로 확장할 수 있습니다.

---

왜 워크위트

WorkWit은 조직의 사무실 업무를 더 똑똑하게 하면서도 데이터에 대한 통제권을 포기하지 않도록 설계된 자체 호스팅 엔터프라이즈급 AI 에이전트 플랫폼입니다.

• 완전 오픈 소스 및 자체 호스팅(MIT). 구독 없이, 좌석당 수수료 없이, 벤더 고정 없이. 자체 인프라에서 운영하세요; 데이터는 절대 서버를 떠나지 않습니다.
• **개인뿐만 아니라 조직을 위해 만들어졌습니다.

English |

WorkWit - Enterprise AI Office Assistant

An enterprise-oriented, privately deployable AI Agent platform. The backend is built on FastAPI + SQLite, and the frontend is a zero-dependency, vanilla-JS single-page application. The whole stack is deployed with one command via Docker Compose.

It is far more than a “contract-review gadget” - it is a unified office-agent foundation that integrates multi-model inference, extensible Skills, Tools, MCP connectors, sandboxed execution, multi-tenant access control, scheduled automation, and long-term memory. Office scenarios such as contract review and resume screening work out of the box, and new capabilities can be continuously extended through “skills + tools”.

---

Why WorkWit

WorkWit is a self-hosted, enterprise-grade AI Agent platform built to make an organization’s office work smarter - without giving up control of your data.

• Fully open source and self-hosted (MIT). No subscription, no per-seat fees, no vendor lock-in. Run it on your own infrastructure; your data never leaves your servers.
• Built for organizations, not just individuals. Real multi-tenant RBAC (Organization / Department / User) with a single source of truth for permission bits (FEATURE_REGISTRY).
• Office scenarios out of the box. Contract review (with Party A / Party B stance) and resume screening (job-profile matching score) ship as built-in modules - no assembly required.
• Trivial to deploy and audit. One docker compose up, a single backend container, SQLite storage, and a zero-dependency frontend. No build step; easy to read, fork, and harden.
• Unified capability catalog. Skills, tools, and MCP connectors are merged into one catalog and kept in sync automatically, so adding a capability is one registration, not three integrations.
• Safe by default. User-supplied code runs in a restricted sandbox (AST scan + CPU/memory/process/timeout limits + privilege drop); secrets stay in .env / data/ that are git-ignored.
• Extensible by design. 40+ built-in skills, a Skills Marketplace, a Tool Library, and the MCP connector framework let you grow capabilities continuously.

Core Features

• General-purpose Agent core: a ReAct (plan-act) loop supporting tool calls, streaming SSE output, and automatic decomposition/execution of multi-step tasks. The model proactively warns when it “claims a file was generated” but never actually invoked a tool (hallucination guard).
• Skills Marketplace: file-based skills (SKILL.md), with 40+ built-in skills (PPT generation, Word typesetting, WeChat official-account / Xiaohongshu writing, contract/legal, research reports, map compliance, 3D/video, Tencent Docs, and more). Supports versioning, rollback, visibility control, and install/uninstall.