selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

AI WorkDeck logo

Self-host AI WorkDeck

v0.74.0

AI workspace for professional document editing and review

BetaOpen SourceAGPL-3.0
Updated2 hours ago
LanguageJavaScript
Setup difficultyModerate
MobileNative app
Stars
167
CPU
4 vCPU
RAM
8 GB
Storage
200 GB data disk / 400 GB backup
Ports
9696 (Backend API), 5173 (Frontend H5 dev server)
Requires
PostgreSQL
Workspace — project files, document editor, and AI chat side by sideRevision toolbar — tracked changes with accept/reject controlsEvidence jump — clicking a citation opens the highlighted source passageIn-app marketplace for plugins and reusable Skills

What is AI WorkDeck?

AI WorkDeck is an open-source AI workspace designed for professional document-intensive work, beginning with legal use cases. It provides a desktop application (Windows/macOS) that bundles a Spring Boot backend, embedded LibreOffice WASM editor with tracked changes, an agent runtime that proposes structured edits users can accept or reject, source-linked citations that jump to supporting passages, three-way document merge for comparing drafts, and a plugin system for custom skills and web panes. The community edition is free under AGPL-3.0-or-later; a Pro subscription adds AI quota, clipboard, staging area, and cloud storage. Commercial licensing and implementation services are offered for firms. The desktop release requires no external dependencies — it runs a local H2 database and bundled JRE. For full self-hosting, operators can deploy the backend with PostgreSQL 14+, Docker sidecars for MinerU (PDF parsing) and PPTX generation, and connect local or cloud LLMs via OpenRouter, Gemini, or other providers. Optional integrations include enterprise data (Qichacha), financial data (Tushare), legal databases (PKULaw), web search (Bocha), OCR, and speech-to-text. The project is pre-1.0 (v0.67.0) with active development.

Should you self-host AI WorkDeck?

AI WorkDeck scores 44/100 on selfhost.directory (how scoring works) and ranks #71 of 120 in AI Assistant. It has 167 GitHub stars.

It shipped 12 stable releases in the past year. The latest, v0.74.0, came out on 28 October 2026.

Setup is moderate: expect some configuration beyond a single command. It is AGPL-3.0-licensed and written mostly in JavaScript.

Its closest alternative on selfhost.directory is hermes-agent (89/100, 257 957 stars). It scores higher overall, so compare the two before you pick. See AI WorkDeck vs hermes-agent side by side.

Features

  • AI agent with reviewable edits
  • Declarative plugin system
  • Document editor with tracked changes
  • Extensible plugin runtime
  • Git-based version control
  • In-app marketplace
  • Local-First Storage
  • MCP / tool orchestration
  • Office/WPS add-in
  • On-device text-to-speech
  • Source-linked citations
  • Team/firm shared memoryRequires server-side component
  • Three-way document merge
  • Vector retrieval with pgvectorRequires PostgreSQL deployment

AI WorkDeck README

How to configure AI WorkDeck

  • AI_PROVIDER_FORMATgeminiAI provider format (gemini, openai, lazyllm)

    AI provider format (gemini, openai, lazyllm)

    values: gemini, openai, lazyllm

  • BACKEND_PORT5001PPTX service backend port

    PPTX service backend port

  • BAIDU_OCR_API_KEYBaidu OCR API key for editable export

    Baidu OCR API key for editable export

  • BOCHA_API_KEYBocha AI web search API key

    Bocha AI web search API key

  • CORS_ORIGINS*CORS allowed origins

    CORS allowed origins

  • DB_HOSTlocalhostPostgreSQL host

    PostgreSQL host

  • DB_NAMEcheckbaPostgreSQL database name

    PostgreSQL database name

  • DB_PASSWORDcheckba123PostgreSQL password

    PostgreSQL password

  • DB_PORT5432PostgreSQL port

    PostgreSQL port

  • DB_USERNAMEcheckbaPostgreSQL username

    PostgreSQL username

  • FRONTEND_PORT3000PPTX service frontend port

    PPTX service frontend port

  • GEMINI_API_KEYGoogle Gemini API key

    Google Gemini API key

  • GOOGLE_API_BASEhttps://generativelanguage.googleapis.comGoogle GenAI API base URL

    Google GenAI API base URL

  • GOOGLE_API_KEYGoogle GenAI API key

    Google GenAI API key

  • MAX_DESCRIPTION_WORKERS5Max concurrent description workers

    Max concurrent description workers

  • MAX_IMAGE_WORKERS8Max concurrent image workers

    Max concurrent image workers

  • MINERU_FORCE_CLOUD0Force cloud MinerU instead of local

    Force cloud MinerU instead of local

    values: 1/true/yes/on to enable

  • MINERU_LOCAL_URLhttp://mineru-service:8000Local MinerU service URL

    Local MinerU service URL

  • MINERU_TOKENCloud MinerU API token (fallback)

    Cloud MinerU API token (fallback)

  • OPENAI_API_BASEhttps://api.openai.com/v1OpenAI API base URL

    OpenAI API base URL

  • OPENAI_API_KEYOpenAI API key

    OpenAI API key

  • OPENROUTER_API_KEYOpenRouter API key for LLM access

    OpenRouter API key for LLM access

  • OUTPUT_LANGUAGEzhOutput language for generated content

    Output language for generated content

    values: zh, ja, en, auto

  • PKULAW_TOKENPKULaw legal data API token

    PKULaw legal data API token

  • QICHACHA_KEYQichacha enterprise data API key

    Qichacha enterprise data API key

  • QICHACHA_SECRETQichacha enterprise data API secret

    Qichacha enterprise data API secret

  • SERVER_PORT9696Backend server port

    Backend server port

  • TUSHARE_TOKENTushare stock data API token

    Tushare stock data API token

AI WorkDeck releases & version history

FAQ

Which external services are optional for self-hosting?

OpenRouter, Gemini, Qichacha, Tushare, PKULaw, Aliyun OCR/Tingwu, and object storage are optional. The basic workbench runs without them.

What are the server requirements for self-hosting?

Minimum 4 vCPU / 8 GB RAM, 200 GB data disk / 400 GB backup. PostgreSQL 14+, Linux, JRE 21, Nginx/HTTPS. GPU for local models sized separately.

Why does open source still charge money?

The open-source workbench is free to use and modify. Commercial revenue comes from usage-based platform services, commercial licensing, implementation support, and marketplace revenue sharing.

How is pricing structured?

Community edition is free. Pro subscription is 149 CNY/month including AI quota, clipboard, staging area, and 3 GB cloud storage. Enterprise licensing and deployment services are separate.

What should I back up before changing computers or uninstalling?

Back up project files, chat data, and personal/project memories manually. Cloud shared content, website account, and external sync copies are separate operations. Mobile relay is not a backup.

How does mobile file transfer work?

Photos taken on iPhone stay in the app (not iCloud). Files relayed via cloud are temporarily stored: computer-to-phone 7 days, phone-to-computer 30 days, then auto-deleted. The relay does not backup files.

Where are conversations stored?

Desktop conversations and logs stay on your computer. Only cloud features (cloud Office add-in, website Skill chat, team shared memory, mobile relay) store data on servers.

What content leaves my computer when I use AI?

Only the content needed for the specific task: your question, relevant conversation, files you provide or the AI reads for the task, and applicable memories. The entire project is not sent by default.

Can a law firm use its own server and models to keep data completely in-house?

Yes. The community edition supports deployment on the firm's own servers with local models, keeping all storage and processing internal.

Will my case materials be used to train models?

No. AI WorkDeck calls existing large models and does not use your materials for training. 'Memory' is just organized notes passed to the model as context.

Where are my case files and chat history stored?

On your own computer. The desktop app stores project files, chat history, and logs locally; they are not uploaded to the vendor's servers.

More in Document Management

Rate this project

Click to rate — no account needed.

0 comments

Posting as guest
Loading comments…