Browse the docs
How-to
Use AI to write and improve steps
Use Thoth AI or connect your own AI provider, then improve text, write step titles from screenshots, turn a video into a guide, write an introduction or translate a guide.
- Plan
- FreePro
- Reading time
- 7 min read
- Checked against Thoth for Windows on
- October 23, 2026
On this page
Thoth’s AI features work two ways. Thoth AI comes with Pro and Team, needs no key, and is paid for with credits: your text goes to Thoth and on to OpenAI, which does the work (writing and voice), and Thoth keeps only how much each request used. Or bring your own provider and key, so you choose the model, the price and where your text goes: requests then go from your PC straight to it, and nothing passes through Thoth. With Ollama, the model runs on your own computer.
Choose how AI works#
- 1
Choose a provider
Open Settings → AI and pick Thoth AI, OpenAI, Anthropic, OpenRouter, Ollama (local) or a custom OpenAI-compatible address.
- 2
With Thoth AI, sign in
Choose Sign in with Thoth. There is no key or model to set up, and the same page shows how many credits you have left.
- 3
With your own provider, add your key and a model
Paste the API key (Ollama needs none) and choose a model. Pick one that can read images if you want it to work from screenshots. The key is stored encrypted on this PC, and Thoth never shows it again.
- 4
Test the connection
Choose Test connection. It sends a tiny request to check the key and model. (Thoth AI’s check costs no credits.)
- 5
Decide about screenshots
Send screenshots is a switch. Off, only text goes to the provider. On, screenshots go with it, with your blurs burned in first.
You can also set the Writing language, the Tone (clear and neutral, friendly or formal) and Custom instructions that go with every request, such as your product names or who the readers are. Providers lists what Thoth works with.
Thoth AI credits#
A credit is one US cent of use. Pro includes 500 credits a month and Team 500 for every seat, shared by the team: admins can limit how much each person uses. Monthly credits reset on the first of the month. You can buy more from the Thoth AI page on your dashboard, in a pack or any amount from 1,000 credits ($10): the more you buy, the less each credit costs, down to 80 cents per 100 credits. Bought credits never expire and they are non-refundable. Each purchase has a Receipt button on that page. Free has no monthly credits, but credits you buy work for the writing tools, the step writer, summaries and translation, for as long as they last.
Thoth AI uses a small, fast model by default. Use more advanced AI (Settings → AI) switches the writing to a more capable one for busy or unusual screens, and it uses about 20 times as many credits for the same work. Thoth warns you in the app, and emails your account’s admins, when credits run low. The dashboard’s Thoth AI page shows what was used: counts of requests, tokens and seconds of sound, never the text.
What you can do#
| Where | What it does | Plan |
|---|---|---|
| AI in the text toolbar | Rewrites the selected text (or all of it): Improve writing, Fix spelling and grammar, Make shorter, Add more detail, Simplify, Friendlier tone, More formal tone. Ctrl+Z in the field restores the original. | Free |
| Microphone button | Dictate a description instead of typing it. Needs Thoth AI, or a provider that offers speech to text (OpenAI or an OpenAI-compatible server). | Free |
| Improve guide | Suggests better step titles and instructions, and shows you a preview before anything changes. | Pro |
| Write step titles and descriptions… | Writes them from each screenshot and the control you clicked. By default it leaves steps you’ve written yourself alone. | Pro |
| Write the introduction… | A guide title and introduction from the steps. | Pro |
| Translate guide… | A translated copy of the whole guide. | Pro |
| Review the guide… | Reads the whole guide and lists what could trip a reader up: unclear steps, anything missing, steps out of order, and a control called two different names. See below. | Pro |
| Guide settings → AI instructions | Extra instructions for one guide, like its audience, on top of your global ones. | Pro |
Have AI read the guide through#
Choose AI → Review the guide… in the editor (or Review the wording with AI… at the bottom of Check this guide). It lists problems a reader could run into, each with the step it is about and a suggestion. Choose a step’s name to go straight to it. It never edits anything: you decide what to change.
- Unclear: a step that is vague or could be done two ways.
- Something missing: a prerequisite, a skipped step, a setting or permission the guide takes for granted.
- Order: steps that depend on something that only happens later, or that repeat.
- Wording: the same button called Save in one step and Submit in another.
What a review sends
Only the guide’s title and each step’s title and description, as plain text. Never screenshots, typed text or notes. Before anything is sent, Thoth swaps email addresses, phone numbers, card and ID numbers, API keys and IP addresses for a label like [email], and cuts the part of a link after the ? (where tokens and ids sit). The window says how many it hid and, if you choose Show exactly what is sent, shows you the text itself. Nothing is sent until you choose Review. With Thoth AI the text goes to Thoth and on to Anthropic, and Thoth keeps only how much each review used. With Ollama it stays on your PC.
Thoth can only recognise the kinds of private detail listed above. A password, a customer’s name or an internal server name written in a step is not recognised, so read Show exactly what is sent if a guide contains things like that, and edit the step first. A review of about forty steps takes a sixth of a credit or so. Reviews of a guide longer than 200 steps read the first 200.
Make a guide from a video#
In beta
Making a guide from a recording works, but it is new. Read through the steps before you share the guide, and send an error report from Settings if something is off.
Drop a screen recording on the library (or choose Import → Word, PDF or screen recording…) and Thoth turns it into a guide: a step for each moment the screen changes and settles, with its screenshot. With Pro and Thoth AI (or your own AI provider) it then writes the steps for you.
- 1
Pick the video
MP4 or WebM works best. Choose how sensitive Thoth should be to changes on screen: more steps, or only the big changes.
- 2
Leave the AI options on
Then write the steps and an introduction with AI is on by default. Use what’s said in the video is also on when your provider can turn speech into text (Thoth AI, OpenAI or an OpenAI-compatible server).
- 3
Create the guide
The guide opens with the AI step writer ready. It first hides sensitive information in the screenshots, then looks at each screen together with the one before it, so it can say what you did, and uses your narration for the wording.
- 4
Review and apply
Each step shows its suggested title and description, and the guide gets a suggested title and introduction. Edit anything, untick what you don’t want, and apply. The original is saved as a version.
- Narration: what you say is matched to the step it belongs to by time. A video with no speech is skipped, and a provider that can’t transcribe (such as Anthropic) simply doesn’t offer the option; the steps are then written from the screens alone.
- Long videos: the sound is sent in pieces of a few minutes each.
- If the sensitive-information check can’t run, nothing is sent. You can turn that option off to continue without it.
Recording in Thoth gives a better guide
A video made somewhere else is just pictures, so Thoth has to work out what changed. If you record with Thoth’s own video recorder and choose Make a guide…, every click becomes a step with the click marked and the control named, and AI only has to polish the wording.
Always review
AI can be wrong about what a screen does. The step writer and Improve guide show you the text before applying it, and save a version first so you can go back.
What leaves your PC
The text you ask it to work on, and screenshots only if Send screenshots is on. Blurs are burned into any screenshot before it is sent. For a guide made from a video, the sound of the video also goes to your provider (with Thoth AI, to Thoth and on to OpenAI), only if Use what’s said in the video is ticked.