Skip to content
EpicDraft Docs

Setting Up AI

AI features in EpicDraft enhance your writing experience by providing intelligent suggestions, automating tedious tasks, and offering insights into your screenplay. These tools are designed to help you focus on creativity.

EpicDraft never requires AI: every feature on this page is opt-in, and the app works fully without any of it.

Who can use AI

  • Make sure your EpicDraft application is updated to the latest version to access AI features.
  • Log in with an EpicDraft account that has a valid subscription. (Pro bought once counts too — AI doesn't depend on the Cloud Service.)
  • New accounts get a 14-day AI trial: for two weeks after signing up you can use every AI feature without a subscription, so you can decide whether it earns its keep before paying.

Choosing a provider

Open Settings → AI and switch AI on. The page has three tabs:

  • Basics — whether AI is on, which model answers by default, when the assistant asks before changing your project, and what AI has cost. A line at the top says how many tasks have a model and names the ones that don't.
  • Connections — where models come from: models downloaded to this device, servers on your network, and services you have a key for. Each is set up once, whatever it is used for.
  • Which model does what — every task the app hands to AI and the model that does it. See below.

The AI settings panel, signed in, with AI enabled

On this device

Run models inside EpicDraft itself, with nothing installed and nothing leaving your device. Connections → On this device is one list of every model this device can download, grouped by what it does — text, dictation, voices, scanned pages, images, 3D models, video, and sound effects and music. Each row shows its size and licence, and text models show whether they fit this device's memory. Download, enable or delete a model from its row. Models load one at a time, on first use.

On your network

If you already run Ollama or LM Studio, EpicDraft can use it — see Ollama support below for the one bit of configuration the web app needs. A ComfyUI server on your own GPU computer can make 3D models, video takes, restyled footage, sound effects and music: enter its address once, then paste a ComfyUI workflow (saved with Save (API Format)) for each job you want it to do.

A device that has on-device models can also share them with your other devices: enable sharing on the machine that has the model, and it appears as a "Remote device" provider on your phone or laptop. Sharing is opt-in per device, and traffic stays between your own devices. Manage it in Settings → Devices.

Services with your key

Bring your own API key for OpenAI, Anthropic, Google Gemini or xAI Grok, or point OpenAI-Compatible at any service that speaks the OpenAI API. Each takes an API key (and optionally a custom endpoint), links to the provider's page for creating a key, and probes the service so you can pick from the models your key actually has access to. Keys for Meshy (3D models), Runway and fal.ai (video takes) and ElevenLabs (sound effects, music and voices) are in the same list.

Your keys are stored on your device and requests go directly from your device to the provider — EpicDraft never proxies them or sees your key.

The default AI, and creativity

The Default AI card chooses which enabled provider answers when nothing more specific applies, and what to optimize for — faster answers or better ones. The Creativity slider sets how adventurous generated text should be; asking for the same thing repeatedly without editing in between automatically nudges creativity up, so a run of samey results actually diverges.

Graph-focused context (experimental, on by default) lets the AI anchor what it reads on the scene your caret is in, or the page you are editing, instead of always starting from the top of the document — answers get more relevant on long scripts. Switch it off in the Default AI card if you prefer.

Synopses are briefs. A Fountain synopsis (a line starting with =, ⌘Y / Ctrl+Y) describes the scene or passage that follows it and never prints. The AI is handed the synopsis for the scene your caret is in and for the passage it is writing, and the scene list it reads carries each scene's synopsis. Write = what this scene is for above a heading and the suggestions, commands and chat work toward it. See Notes and synopses.

Which model does what

Different jobs suit different models, so routing is per task. The Which model does what tab lists each kind of work in two parts, Writing and Director — Chat, Suggestions, Analysis, Field suggestions, Dictation, Read aloud, PDF & image reading, Storyboards & concept art, Video takes — and lets you pin a specific provider and model to any of them. Everything defaults to Auto, which picks from your enabled providers and shows what it picked and why.

3D models, sound effects and music are chosen each time you generate one, so their rows list what the Generate dialog will offer.

When nothing can do a kind of work yet, its row says Not set up and offers the quickest fixes right there: download a model to this device in one click, or jump to the key or server it needs in Connections.

Two tasks are deliberately never auto-routed to a remote service: Image and Speech only use a remote provider if you pin one yourself. See Generating Images and Read Aloud.

What the assistant may change

The Changes to your project setting controls when AI tools that modify your project ask first:

  • Ask for every change — nothing is written without your approval.
  • Ask for destructive changes (default) — additions go through; overwrites and deletions ask.
  • Don't ask — the assistant works unattended.

A journal keeps track of what the assistant changed, so its work can be undone. A version of the project is also saved before destructive tool calls — it shows in File History as Before the assistant — so a rollback survives even a reload.

Stopping AI

Anything the AI is doing can be stopped at any time:

  • Press ⌘. / Ctrl+. to interrupt every running AI feature at once.
  • A Stop notice appears whenever an AI run takes more than a moment, with a button that does the same.
  • Individual features have their own exits too — Escape dismisses a suggestion or command in progress.

Usage and cost

The AI Usage panel in Settings → AI → Basics shows token usage and estimated cost across all your scripts. The log is stored only on your device and is never shared. Usage also surfaces where you work: in the chat panel and the editor's status bar.

Ollama support

Ollama is a platform that provides access to various AI models. EpicDraft supports Ollama, allowing you to leverage its capabilities for generating text, suggestions, and more. To use Ollama with EpicDraft, ensure you have Ollama installed and running.

Allowing EpicDraft's origin

Ollama answers browser requests only from the origins listed in its OLLAMA_ORIGINS environment variable. Out of the box that list covers localhost, 127.0.0.1 and the app://, file:// and tauri:// schemes — enough for the desktop and mobile apps, but not for EpicDraft on the web. Until you add the web app's origin, every request is answered with 403 Forbidden, which the browser reports as a failed connection.

Set OLLAMA_ORIGINS to the origin shown in the address bar (for the hosted app, https://app.epicdraft.com), then restart Ollama and reload EpicDraft.

On macOS:

launchctl setenv OLLAMA_ORIGINS "https://app.epicdraft.com"

On Windows, open Environment Variables and add a new user variable called OLLAMA_ORIGINS with that value — or from a terminal:

setx OLLAMA_ORIGINS "https://app.epicdraft.com"

On Linux, or any time you start the server yourself:

OLLAMA_ORIGINS="https://app.epicdraft.com" ollama serve

Separate several origins with commas. * allows every site to reach your Ollama server, so prefer listing the origins you actually use.

If EpicDraft still shows no models after this, the AI settings panel tells you which problem it is: whether Ollama refused the page's origin, or whether the server could not be reached at all.