WRITE

v1.2.0 — Open Source

Your screenplay.
Untouched. Unleaked.
Quietly perfected.

A local-first AI sidecar for screenwriters. Sits beside Fade In, Final Draft, and Highland. Understands Fountain syntax. Never rewrite your voice. Never leaves your machine.

↓ Install GramWrite See how it works

New in v1.2.0

✦

Inline Word Diffs & Confidence

Suggestions now highlight exactly which words changed using color-coded HTML. Every suggestion includes a built-in confidence score (HIGH/MEDIUM/LOW) visually mapped to a color indicator.

⌗

Strict Screenplay Mode

Configurable strict mode guarantees elements like character names, parentheticals, and sluglines are 100% ignored. GramWrite strictly evaluates Dialogue and Action lines.

◎

Present Tense Enforcement

GramWrite heuristically detects past tense action lines and attempts to algorthmically convert them to present tense equivalents instantly.

·

Multilingual Support

The underlying LLM pipeline has been augmented to implicitly detect the original language and correct it directly without translating.

Philosophy

◎

Local-first AI

All inference runs on your machine via Ollama or LM Studio. Your text never leaves localhost. Not even a header. Not even a ping.

⌗

Fountain-aware

GramWrite knows INT. OFFICE - DAY from a dialogue block. It won't touch sluglines, character names, or transitions. Fragments in action lines are features, not bugs.

·

Zero distraction

One small dot in the corner. Grey when idle. Green when it found something. It never interrupts. It waits for you to look.

How it works

It reads only what you write.
Nothing else.

GramWrite uses your OS accessibility API to extract the current paragraph — never the full document. After 2 seconds of inactivity, it passes the text through a Fountain parser, then to a local LLM. The response is grammar-only.

INT. COFFEE SHOP — EARLY MORNING SKIP
Rain against glass. MAYA, 30s, stares at a blinking cursor. ACTION ✦
MAYA SKIP
I don't know what you want from I. DIALOGUE ✦
She close the laptop. ACTION ✦
CUT TO: SKIP

✦ = GramWrite checks this element  ·  SKIP = never modified

Setup

  1. Install a local LLM backend

    GramWrite works with Ollama (recommended, free) or LM Studio. Ollama is one command:

    # macOS / Linux curl -fsSL https://ollama.com/install.sh | sh # Then pull a lightweight model ollama pull qwen3.5:0.8b

    Or install LM Studio and load any small model from its interface.

  2. Install GramWrite

    # macOS / Linux git clone https://github.com/revanthlevaka/GramWrite cd gramwrite && ./install.sh # Windows git clone https://github.com/revanthlevaka/GramWrite cd gramwrite powershell -ExecutionPolicy Bypass -File install.ps1

  3. Grant accessibility access

    GramWrite reads text through your OS accessibility API. On macOS, add Terminal to System Settings → Privacy & Security → Accessibility. On Windows, run as Administrator once.

  4. Launch and write

    ./gramwrite.sh # macOS / Linux .\gramwrite.bat # Windows

    Open Fade In, Final Draft, or Highland. The dot appears in the corner. Write normally. Click when it glows.

The experience

1

Start your model

ollama serve — or open LM Studio

2

Launch GramWrite

A small dot appears. Grey. Waiting.

3

Write normally

Draft. Delete. Rewrite. The dot watches.

4

Click the glow

Green dot = suggestion waiting. One click. One copy.

Your screenplay never leaves your machine.

This is not a feature. It's an architectural constraint. GramWrite has no backend server. No analytics SDK. No network calls except to localhost. The source code is open — verify it yourself.

  • All inference runs on your CPU/GPU
  • No cloud APIs, ever
  • No telemetry or crash reporting
  • No user accounts or authentication
  • Network calls: localhost only
  • Open source — fully auditable