dream-num/univer-workspace

Univer Workspace – AI‑augmented Office suite

What it is – An open‑source, self‑hosted office environment built on the Univer Office SDK. It combines traditional productivity tools (Sheets, Docs, Slides, relational tables, boards, etc.) with AI‑agent capabilities that can read, edit, and review documents through a conversational interface.

Key components

Component	Role
Workspace Browser	Web UI for humans to create, edit, and collaborate on Office‑type files.
Workspace Server	Central service that stores documents, permissions, blobs, and runs the OpenAPI contract.
Workspace Agent	Local desktop app (built on DeepSeek Harness) that lets an AI agent converse about files, make draft edits in an isolated Worktree, and present the changes for human review.
Workspace CLI	Command‑line tool that agents or scripts can call to load, inspect, modify, validate, and render Office content programmatically.

How it works – The server is the single source of truth for all data. Humans interact via the Browser, while agents interact via the CLI or the Agent app. Agents edit inside a Worktree (an isolated draft). Once the draft is ready, a human reviews the changes in the Browser or Agent UI and decides to merge or discard them, keeping the main workspace clean.

Main use‑cases

• Human‑centric collaboration – Teams co‑edit spreadsheets, documents, slides, and boards with fine‑grained ACLs, trash/recent views, and file import/export.
• AI‑driven automation – Agents can generate spreadsheet‑based mini‑apps (dashboards, reports), render PDFs/PNGs, run layout checks, and perform multi‑step edits without touching the live trunk until approved.
• Automation & scripting – The CLI lets existing LLMs (Claude, Codex, Kimi, etc.) invoke a Workspace CLI Skill to manipulate Office files in a reproducible, version‑controlled way.
• Self‑hosted deployment – Operators can run a single Browser + Server instance, integrate custom auth (password, GitHub, Discord, OAuth), and keep product, collaboration, and blob storage under their own control.

Quick start (from the README)

# Prereqs
node >=24, pnpm 11

# Install deps
pnpm install
cp apps/workspace/.env.example apps/workspace/.env

# Run server
pnpm workspace:dev:server

# In another terminal, run the web UI
pnpm workspace:dev:web

# Open http://127.0.0.1:5173 in a browser

The server also exposes an OpenAPI spec at http://127.0.0.1:3020/openapi.yaml.

Adding the AI Agent – Follow apps/agent/README.md to install the desktop Agent, configure OAuth callbacks, and supply model credentials. The Agent talks to the same server, creates Worktrees, and shows document previews alongside a chat window.

CLI usage – After installing the global package (npm i -g univer-workspace-cli), configure the target workspace origin, run univer-workspace-cli login (follow the OAuth flow), then invoke commands or the provided Skill from any LLM‑powered agent.

Architecture highlights

• Authoritative Server – All identity, ACLs, revisions, and blobs are stored centrally; clients are never trusted for authority.
• Separate storage domains – Product data, collaboration state, and raw blobs are managed independently but coordinated via idempotent operations.
• Contract‑first HTTP – OpenAPI definitions generate types for the server, browser, and CLI, ensuring a single source of truth.
• Worktree workflow – Isolated draft edits → Ready → human review → Merge/Reopen → trunk.
• Modular packages – Core client, HTML viewer, markdown viewer, reference provider, and DSH plugins are kept in private packages; the repo only composes them.

License – Apache‑2.0 (see LICENSE).

---

All information above is taken directly from the repository’s README; no additional features have been inferred.