TextFlow is an LLM-powered Integrated Writing Environment (IWE) for professional literary editing — just as Windsurf and Cursor bring cascaded AI assistance to code, TextFlow brings the same approach to prose.
- Writers — working on novels and long-form fiction
- Editors — professional literary editing
- Screenwriters — screenplay and dramatic works
- Authors — any structured manuscript requiring consistency
TextFlow is not just a set of LLM tools for text editing. It is a convenient environment for working with manuscripts. It features a natural language chat interface for agent control, similar to Visual Studio Copilot or Cursor.
The system includes Story Bible, which automatically generates and maintains knowledge bases for characters, world-building, and timeline consistency. Every change goes through the Patch System, allowing you to review changes with a clear rationale for each change. Quality Metrics offers a quantitative assessment of the manuscript based on parameters such as clarity, tension, and pacing, while a complete version history with semantic versioning provides full rollback capability.
How it works. Type natural language commands: "Remove passive voice from Chapter 3" or "Make the dialogue more authentic. Agent understands context and manuscript structure. Get instant feedback and execution plans. Review changes before applying them.
Examples:
"Strengthen the opening hook"
"Add more sensory details to the forest scene"
"Check if Emma's character development is consistent"
"Improve pacing in the action sequence"
"Find all instances of telling instead of showing"
The AI agent operates through a chat interface and has access to several core functions: it can retrieve specific manuscript fragments with editing rules and Story Bible data, execute structured edit plans with patch validation, update the Story Bible with new information, and analyze the entire manuscript to build or refine the Story Bible. You simply send a request — for instance, “analyze character development” or “remove passive voice” — and the agent autonomously determines which tools to use, retrieving context, generating edits, and applying them where needed.
There are no mode switches or manual configurations — it’s a single intelligent system that understands your intent and performs the necessary operations. Just like Cursor or Windsurf handle multiple coding tasks through one unified assistant, this agent brings the same seamless workflow to literary editing.
Pre-built prompts for common editing tasks (accessible via chat):
| Category | Commands |
|---|---|
| Foundation | • Build Story Bible • Update timeline consistency • Track character development |
| Structure Analysis | • Analyze character arcs • Identify plot holes • Check continuity errors |
| Story Flow | • Evaluate pacing and tension • Strengthen opening hook • Improve scene transitions |
| Scene Polish | • Review dialogue authenticity • Add sensory details • Enhance character voice |
| Line Editing | • Tighten prose • Fix passive voice • Remove redundant adverbs |
Import:
- Files:
.txt,.docx,.pdf - URLs with AI-powered HTML parsing
Export formats:
PDF— print-readyEPUB— e-readersDOCX— Microsoft WordIDML— Adobe InDesignTXT— plain textJSON— full project backup
Traditional AI editors work like this:
User Request → AI Response → Done
TextFlow's cascade ensures quality:
Chat Command → Analyze → Generate → Validate → Report
- Agent receives editing instruction via chat
- Calls
get_document_contextto retrieve text, rules, and memory - Creates actionable TODO list with specific steps
For each plan item, the agent generates:
{
"before_anchor": "context before target",
"before_excerpt": "text to replace",
"after_anchor": "context after target",
"after_proposed": "new text",
"rationale": "why this change improves the text"
}Calls execute_edit_plan tool which runs:
- ✅ Anchor validation — ensures correct location
- ✅ Style checks — no new passive voice, minimal adverbs
- ✅ POV checks — maintains narrative consistency
- ✅ Metrics calculation — before/after quality scores
Only validated patches are applied.
Agent presents results in chat:
- Natural language summary
- Comprehensive JSON report
- Applied patches with reasons
- Skipped patches with explanations
- Metrics comparison
🔍 Auditability — Every change has rationale and validation trail
🛡️ Safety — Multi-layer checks prevent quality degradation
📖 Transparency — Users see exactly what changed and why
⏮️ Revertibility — All changes tracked in semantic version history
🎯 Precision — Anchor-based patches remain valid even during concurrent edits
- Node.js 18+ (download)
- Google AI API key (get here)
- Safari browser (recommended for best performance and compatibility)
1. Clone the repository
git clone https://github.com/KazKozDev/textflow.git
cd textflow2. Install dependencies
npm install3. Start development server
npm run devApp opens at: http://localhost:5173
Double-click Start.command file — it will automatically:
- Install dependencies
- Start dev server
- Open app in browser
Quick edits without patch review:
- Select text
- Press
Cmd+I(orCtrl+I) - Enter instruction
- AI replaces instantly
Search by meaning, not just keywords:
- Search "betrayal" finds thematically related passages
- AI understands context and concepts
- React 18 + TypeScript — UI with strict typing
- Vite — build tool and dev server
- Google Gemini 2.5 Flash — AI model with function calling
- react-window — virtualization for large documents
- localStorage — persistent state management
No backend required — fully client-side application.
Business Source License 1.1
TextFlow is licensed under the Business Source License 1.1, which allows:
✅ Free use for non-production purposes:
- Personal projects
- Research and education
- Testing and evaluation
- Internal development
❌ Production use requires a commercial license
🔓 Automatic transition to Apache 2.0 on January 1, 2029
For full license terms, see the LICENSE file.
For commercial licensing inquiries, contact: Artem Kazakov Kozlov (KazKozDev)
If you like this project, please give it a star ⭐
For questions, feedback, or support, reach out to: