Editian

Editian is a browser-based editor for .docx, .pptx, and markdown files. Upload a document, describe what you want in plain language, review the before/after diff, and accept only the changes you want.

License

Editian is distributed under the PolyForm Noncommercial 1.0.0 license.

• Noncommercial use, modification, and redistribution are allowed under that license.
• Commercial use is not allowed without separate permission from the copyright holder.

See LICENSE for the full license text. See COMMERCIAL.md for commercial licensing inquiries.

Why Use It?

• Works with Word, PowerPoint, and Markdown files
• Lets you edit with natural-language instructions
• Shows a diff before anything is applied
• Supports local models with Ollama or hosted APIs like OpenAI
• Keeps the workflow non-destructive until you click Accept
• Compares two documents side-by-side with AI-powered entity diff
• Extracts a knowledge graph from any document and uses it to ground AI chat answers

What You Can Do

DOCX

• Revise a whole document or selected paragraphs
• Paraphrase, shorten, formalize, or clean up wording
• Insert new paragraphs
• Delete paragraphs
• Summarize a selected section and place the summary below it
• Merge multiple selected paragraphs into one
• Edit tables

Markdown

• Review rendered markdown while editing the underlying .md source by block
• Revise a whole file or selected blocks with AI
• Manually edit markdown blocks and keep undo/redo support

PPTX

• Revise slide text or selected text boxes
• Edit PPTX tables
• Add slides
• Preview slides with a higher-fidelity renderer when LibreOffice + Poppler are installed

Modes

AI Mode

The main editing mode. Upload a document and use three tabs in the sidebar:

• Edit — give natural-language instructions, review the diff, accept or reject each change
• Chat — ask questions about the document; optionally enable the knowledge graph for more accurate, entity-grounded answers
• Graph — extract a knowledge graph from the document and explore entities and relationships visually

Manual Mode

Click any paragraph, shape, or table cell to edit it directly inline. Undo/redo is fully supported.

Compare Mode

Load two documents (from your workspaces or as new uploads) and compare them side-by-side. Three views are available:

• Documents — scroll both documents in sync
• Entity Diff — extract entities from both documents with AI and see a structured table of what changed, was added, or was removed
• Graph — view both entity graphs overlaid, with color-coded diff status on each node

The compare chat uses both document texts and the entity diff as context, so it can answer questions like "what changed in section 3?" with entity-level precision.

Knowledge Graph

Editian can extract a knowledge graph from any document using your configured LLM. The graph captures:

• Entities — people, organizations, dates, numbers, locations, products, and concepts, each with a value and the paragraph numbers where they appear
• Relationships — directed connections between entities (e.g. Acme Corp → employs → John Smith)

Once extracted, the graph is shown as an interactive force-directed canvas (Obsidian-style). You can zoom, pan, drag nodes, and export the graph as a PNG.

Using the Graph in AI Chat

After extracting the graph, go to the Chat tab. An indigo Knowledge graph active indicator appears above the input bar. When active, every chat message sends the full entity and relationship list to the LLM as structured context — in addition to the document text.

This helps the LLM:

• Answer entity-specific questions without scanning the full document
• Cite paragraph numbers accurately
• Reason about relationships between named entities

You can toggle the graph context on or off at any time without re-extracting.

Workspaces

The left panel shows your workspaces — one per document. You can:

• Create, rename, and delete workspaces
• Organize workspaces into directories
• Branch a workspace to create an independent copy
• Drag workspaces between directories

Background tasks (graph extraction, entity diff) run independently of which workspace is active. A pulsing indicator in the top bar shows when any task is still processing. Each workspace row shows a spinner while its graph is being extracted, even if you have switched away.

Quick Start

Docker

If you want the easiest install path, use Docker Compose. The setup runs separate frontend and backend containers:

• frontend: nginx serving the built React app
• backend: FastAPI + LibreOffice + Poppler for document editing and PPTX rendering

To change the host ports, create a root .env file (next to compose.yml) like this:

EDITIAN_FRONTEND_PORT=3000
EDITIAN_BACKEND_PORT=8000
EDITIAN_LOG_LEVEL=INFO
EDITIAN_LOG_FORMAT=text
EDITIAN_LOG_TO_FILE=true
EDITIAN_LOG_DIR=/root/.editian/logs
EDITIAN_LOG_FILE_NAME=backend.log
EDITIAN_LOG_ROTATION_WHEN=midnight
EDITIAN_LOG_ROTATION_INTERVAL=1
EDITIAN_LOG_BACKUP_COUNT=14
EDITIAN_LOG_ROTATION_UTC=false

docker compose up --build

Then open:

http://localhost:3000

Notes:

• Uploaded files and undo/redo history are stored in the editian_data Docker volume
• The container includes LibreOffice and Poppler, so PPTX rendering uses the higher-fidelity slide image path
• If you want S3-backed storage, add the relevant environment variables in compose.yml
• LLM provider settings are still configured inside the app UI
• If you use Ollama on your host machine while Editian runs in Docker, set the app's Ollama Base URL to http://host.docker.internal:11434/v1
• The LLM settings panel includes a Test connection button so you can verify the provider/model before using chat or revise
• If EDITIAN_FRONTEND_PORT is not set, Docker Compose uses 3000
• If EDITIAN_BACKEND_PORT is not set, Docker Compose uses 8000
• The browser should use the frontend port; nginx proxies /api to the backend container automatically
• If you set EDITIAN_FRONTEND_PORT=8080, open http://localhost:8080 instead
• Set EDITIAN_LOG_FORMAT=json if you want structured container logs
• Backend logs are also written to /root/.editian/logs/backend.log inside the backend container by default

To stop the app:

docker compose down

To also remove the persisted document volume:

docker compose down -v

1. Start the backend

Using uv:

cd backend
uv venv
uv pip install -r requirements.txt
source .venv/bin/activate   # Windows: .venv\Scripts\activate
uvicorn main:app --reload

Or with pip:

cd backend
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn main:app --reload

The backend runs at http://localhost:8000.

2. Start the frontend

cd frontend
npm install
npm run dev

The frontend runs at http://localhost:3000.

3. Choose your AI provider

Editian supports:

• Ollama for local models
• OpenAI
• Any OpenAI-compatible API

You can configure the provider from the settings panel inside the app.

Basic Workflow

1. Open http://localhost:3000
2. Upload a .docx, .pptx, or .md file
3. Select a paragraph, table, shape, or slide, or leave nothing selected to edit a larger scope
4. Enter an instruction such as Paraphrase this paragraph or Make this slide more concise
5. Click Revise
6. Review the before/after output
7. Accept or reject each change
8. Download the updated file

AI Prompt Examples

The app shows sample prompts based on what is selected.

• Whole DOCX: Fix grammar and tone throughout the document
• One DOCX paragraph: Paraphrase this paragraph
• Multiple DOCX paragraphs: Summarize this and put it below
• Multiple DOCX paragraphs: Merge these into one paragraph
• DOCX table: Standardize the wording in this table
• PPTX slide: Make this slide more concise
• PPTX slide: Add a new slide about...
• PPTX text box: Rewrite this text more clearly

Better PowerPoint Rendering

For the closest PPTX preview quality, install:

• LibreOffice
• Poppler (pdftoppm)

When these tools are available, Editian renders PowerPoint files like this:

1. .pptx to PDF with LibreOffice
2. PDF to slide images with pdftoppm
3. The app shows those rendered slides while keeping editable overlays for text and tables

If those tools are not installed, Editian falls back to its built-in HTML/Python preview path. That fallback still works, but complex templates may look less accurate.

Install on macOS

brew install --cask libreoffice
brew install poppler

Install on Ubuntu / Debian

sudo apt update
sudo apt install -y libreoffice poppler-utils

Install on Fedora

sudo dnf install -y libreoffice poppler-utils

Install on Windows

• Install LibreOffice so soffice.exe is available
• Install Poppler for Windows so pdftoppm.exe is available
• Add both to PATH

Windows support for the native PPTX renderer is not fully wired or tested yet.

Storage Options

By default, files are stored locally under ~/.editian/files.

If you want S3-backed storage, set this in backend/.env:

STORAGE_BACKEND=s3
S3_BUCKET=your-bucket-name
AWS_REGION=us-east-1
AWS_ACCESS_KEY_ID=...
AWS_SECRET_ACCESS_KEY=...

Optional:

S3_PREFIX=editian
S3_ENDPOINT_URL=

For local-only storage:

STORAGE_BACKEND=local

Logging

The backend logs to both:

• stdout, so you can watch it in the terminal or with docker compose logs -f backend
• a rotating log file

These settings control the logger:

LOG_LEVEL=INFO
LOG_FORMAT=text
LOG_TO_FILE=true
LOG_DIR=~/.editian/logs
LOG_FILE_NAME=backend.log
LOG_ROTATION_WHEN=midnight
LOG_ROTATION_INTERVAL=1
LOG_BACKUP_COUNT=14
LOG_ROTATION_UTC=false

LOG_FORMAT can be text or json. LOG_ROTATION_WHEN can be s, m, h, d, midnight, or w0 through w6.

Default log file locations:

• local runs: ~/.editian/logs/backend.log
• Docker backend container: /root/.editian/logs/backend.log

For local development, put those variables in backend/.env.

For Docker Compose, set these in the root .env instead:

EDITIAN_LOG_LEVEL=INFO
EDITIAN_LOG_FORMAT=text
EDITIAN_LOG_TO_FILE=true
EDITIAN_LOG_DIR=/root/.editian/logs
EDITIAN_LOG_FILE_NAME=backend.log
EDITIAN_LOG_ROTATION_WHEN=midnight
EDITIAN_LOG_ROTATION_INTERVAL=1
EDITIAN_LOG_BACKUP_COUNT=14
EDITIAN_LOG_ROTATION_UTC=false

Examples:

# rotate every day
LOG_ROTATION_WHEN=midnight
LOG_ROTATION_INTERVAL=1

# rotate every 6 hours
LOG_ROTATION_WHEN=h
LOG_ROTATION_INTERVAL=6

# keep 30 rotated files
LOG_BACKUP_COUNT=30

Requirements

• Python 3.13+
• Node.js 18+
• uv or pip
• An LLM provider such as Ollama or OpenAI
• Docker Desktop or Docker Engine + Docker Compose, if using the container setup

Notes

• macOS and Linux are supported for the higher-fidelity PPTX renderer
• The backend expects soffice and pdftoppm on the machine for the best PPTX output
• All edits remain reviewable before you apply them
• The UI is available in English, Chinese (Simplified), and Korean

Todos

• Admin page
• Skills
• Session management
• Support PDF
• Markdown extraction
• OCR / Overlay translation