QUILL Lite’s free AI: the complete guide

QUILL Lite 2.0. Updated November 1, 2026.

QUILL’s free AI helps with the writing in front of you. It can summarize it, rewrite it, proofread it, explain it, or answer a question about the document you have open. It is built into QUILL Lite and it is free. You do not need an account, a password or a card. If you already pay for ChatGPT or GitHub Copilot, or have your own key, QUILL Lite can use those instead, and you pick in one window, Choose Your AI (Ctrl+4).

This guide walks you through all of it: how to start, what each command does, exactly which text is sent, what happens to the answer, the limits, and your privacy. If you would rather learn by doing, open Help > Tutorials… (Ctrl+Alt+F1) and choose Asking a question about a document. It takes about six minutes.

Contents

• Starting, in about a minute
• The keys
• The AI pad
• The seventeen things it can do
• Exactly what is sent
• Asking about a whole document
• When something is too long
• The answer window
• Having a conversation
• Usage, About and your support ID
• Limits, and when they start again
• When something goes wrong
• Your privacy
• Signing out and changing your mind
• Choose Your AI: which AI is in use
• Using your own AI key: no limits
• Using your ChatGPT subscription
• GitHub Copilot and Codex
• Do a Document Task
• Ask About an Image
• Web search
• Tidy Dictated Text
• The Dictionary: ask about the word you are on
• Questions people ask

Starting, in about a minute

You only do this once on each computer. Here is how.

1. Open the agreement. Press Ctrl+Alt+Shift+K, or choose Tools > AI > Privacy Agreement. You can open it even while AI help is switched off.
2. Read it, then choose I Agree. It tells you what is sent, what is kept, what is not kept, and how to change your mind later. Agreeing also switches AI help on, so there is nothing else to turn on. If you choose No Thanks, nothing is sent and nothing changes.
3. Connect this computer. The Connect window opens on its own. It shows an eight-character code, and it reads the code out to you too. Press Open the Connect Page. Your browser opens with the code already filled in, so all you have to do is press Confirm. The window tells you when you are connected.
   • Not sitting at the computer you are connecting? You can type the code at the address the window gives you, on any device. A phone is fine.
   • Say the Code Again reads the code one character at a time. Copy the Code puts it on the clipboard.
4. Try it. Put your cursor in any paragraph and press Ctrl+Alt+G. Choose Summarize, then press Send.

That is all. There is no account, no password and no email address. The code is the whole thing.

If you reach for an AI command before you have agreed, QUILL Lite shows you the agreement first. If you reach for one before this computer is connected, it takes you to the Connect window.

What you learned, and where to go next

You agreed once, connected this computer with a short code, and sent your first summary. Next, The keys lists every AI command in one place, and The AI pad shows you around the window you just used. The tutorial Asking a question about a document walks through these same steps with the app guiding you.

The keys

All of these are in Tools > AI.

You can change any of these keys in Tools > Keyboard Manager.

Escape closes every AI window. You lose nothing by closing one. Your document only changes when you press a button that says it will change it.

What you learned, and where to go next

Two keys open the pad, Ctrl+4 and Ctrl+8 tell you and change which AI answers, and Escape gets you out of any AI window. The next chapter, The AI pad, explains what you find when you press Ctrl+Alt+G.

The AI pad

Press Ctrl+Alt+G and the pad opens on whatever your cursor is in. Here is what you find, from top to bottom:

1. About to send. One sentence telling you how much will be sent, where it comes from, and whether it fits the free limit. For example, “About to send 84 words from this paragraph.”
2. What will be sent. A read-only box holding exactly the text that will go, and nothing else. Focus starts here, so the first thing you hear is the text itself. Arrow through it at your own pace.
3. Send this much. Which part of your document to send. You only see this when there is a real choice to make. See Exactly what is sent.
4. What do you want done? A list of the seventeen things the AI can do. Press a letter to jump. Press F1 on any of them to hear what it does.
5. Your question. This only appears for the choices where you type: Ask a question about the document, Ask a general question, and Have a conversation (where it is your first message, and you can leave it empty).
6. Send, and Close.
7. Status. What happened to your last request: still working, how many requests it used, or what went wrong.

You can reach every line in the pad with Tab, including the summary at the top and the status at the bottom. So if you miss something, you can always go back and read it.

You never have to wait for the AI. Requests run in the background. Keep typing, save, switch documents, or close the pad while the answer is on its way.

What you learned, and where to go next

The pad shows you exactly what will be sent before anything goes, and you can Tab to every part of it. Next, The seventeen things it can do explains each choice in the list.

The seventeen things it can do

The first four

• Summarize. A few plain sentences saying what the passage says. Handy for a long email, a dense report, or terms and conditions.
• Rewrite. The same meaning, clearer and shorter. Handy for the paragraph you have rewritten four times and still do not like.
• Proofread. Fixes spelling, grammar and punctuation, and leaves your wording alone. The corrected text comes back for you to compare with your own. Your document does not change until you choose to put it there.
• Explain. What the passage means, in plain language. Handy for jargon or a legal clause.

The writing tools

These are new in 1.1. Each one works on the same passage the first four do, and each one uses one request.

• Shorten. About half the length, with every important fact kept. Handy for a message that has to fit a character limit.
• Simplify. Plain, easy-to-read language: short sentences, everyday words, and technical terms explained. Handy for a letter from a bank, a hospital or a lawyer, or when you want to write something anyone can read.
• Make more formal. The same meaning in a professional tone. Handy for an email to a manager or an official complaint.
• Make friendlier. The same meaning in a warmer tone. Handy for the note that came out colder than you meant.
• Turn into a list. Bullet points, or numbered steps when the passage describes a process. A list is much quicker to review by ear than a paragraph.
• Find action items. Every task, who does it and by when, and every date or deadline, one per line. Handy for meeting notes and long email threads. If there are none, it tells you so.
• Suggest headings. Headings that would split a long passage into sections, and where each one goes. Your screen reader can jump between headings, so this is a good way to make a wall of text easy to move around. Nothing is inserted for you. You place the ones you want.
• Continue writing. A next paragraph in the same voice, for when you are stuck. Insert Below puts it after the paragraph you are in.
• Write an email reply. Select an email you received and you get a polite, clear reply to edit. It answers each question the email asks, and anything only you can decide is left in [square brackets] for you to fill in.
• Translate. Choose a language from the Translate into list, which appears when you pick Translate: English, Spanish, French, German, Italian, Portuguese, Dutch, Swedish, Polish, Russian, Ukrainian, Turkish, Arabic, Hebrew, Hindi, Chinese (Simplified), Japanese, Korean, Vietnamese and Tagalog. The meaning, tone, line breaks and lists are kept.

The three where you type

• Ask a question about the document. Type a question, and QUILL Lite finds the parts of the document that answer it. See Asking about a whole document.
• Ask a general question. Type any question. Only your question is sent, nothing from your document. You get one answer, of up to about 750 words.
• Have a conversation. Talk back and forth in a window of its own. Each reply remembers what was said before. See Having a conversation.

Ctrl+Alt+Z opens the pad with the question already chosen, so you can start typing straight away.

What you learned, and where to go next

You have four everyday choices, ten writing tools, and three choices where you type your own question. Next, Exactly what is sent tells you which part of your document goes with each request.

Exactly what is sent

For Summarize, Rewrite, Proofread and Explain, the pad sends one piece of your document. It picks that piece like this:

1. If you have selected text, it sends your selection.
2. If you have not, it sends the paragraph your cursor is in. A paragraph is the text between one blank line and the next. You do not need to select anything. Put your cursor anywhere in the paragraph and press the key.
3. If your document has headings, you can choose This section instead. That is everything from the heading above your cursor down to the next heading.

Send this much lets you switch between whichever of these apply: What I have selected, This paragraph, or This section. The What will be sent box changes the moment you pick a different one, so you always hear what is going. When only one of them applies, you do not see the chooser at all, which saves you a Tab.

A few details worth knowing

• Sections follow Markdown-style headings. That means lines that start with #, or a line underlined with === or ---. If your document has none, This section is not offered.
• A document with no blank lines is one paragraph. In that case “this paragraph” is the whole file. The size check in When something is too long stops one that is very long.
• Nothing else goes with it. Not the rest of the document, not the file’s name, not where it is saved. The box shows you the whole of what is sent.

What you learned, and where to go next

The pad sends your selection, or the paragraph you are in, or the section you are in, and you hear exactly what that is before it goes. Next, Asking about a whole document explains how a question about a long document works.

Asking about a whole document

Ask a question about the document works a little differently, and that is what makes it fine to ask about a long document.

QUILL Lite does not send the whole file. Instead, it reads the document on your own computer and splits it into passages of about 180 words. Each passage is labelled with the heading it sits under. Then it picks the three passages most likely to answer your question and sends only those, with your question.

• You see the passages it chose before you press Send, in the What will be sent box. Each one is named, for example “Excerpt 1, from” followed by the heading. If they are not the right ones, try asking again using the words the document itself would use.
• A passage whose heading matches your question counts for more. A heading is the author’s own label for what a passage is about.
• If your question has no useful words in it, like “what is it?”, you get the opening passages of the document. That is usually a decent guess.
• A very long document is searched up to about the first 375,000 words. The pad tells you when that happens. Put your cursor nearer what you mean, or select a part and use one of the other choices instead.
• The answer comes only from those passages. If the document says something somewhere else, the AI did not see it.

What you learned, and where to go next

For a question about the document, QUILL Lite picks the three best passages on your computer and shows you them before sending. The tutorial Asking a question about a document lets you try this with the app guiding you. Next, When something is too long covers the size limit.

When something is too long

One request can hold about 3,000 tokens. That is roughly 2,250 words or 12,000 characters, counting your text, your question and the instructions together. The answer itself is up to about 500 tokens, which is a few hundred words.

If what you are about to send is bigger than that, the pad tells you before you press Send, and it says it in words, not tokens. For example: “That is about 4,000 words, and the free limit is about 2,250.” Select less, or choose a smaller part in Send this much. Nothing is sent and nothing is used.

What you learned, and where to go next

The free AI takes about 2,250 words at a time, and the pad warns you before you go over. Next, The answer window explains what you can do with an answer once it arrives.

The answer window

The answer arrives in its own window. Its title says what you asked for: Summary, Rewrite, Proofread, Explanation or Answer. Focus lands on the answer, so your screen reader starts reading it straight away. Below it you find these buttons:

• Replace My Selection. Puts the answer where your selection was. You only get this button if you had text selected, and only while that text is still exactly as it was. If you changed it while the answer was on its way, the button is left out, so the answer never lands on words it was not about.
• Insert Below. Puts the answer underneath the paragraph your cursor is in, after a blank line, so it never lands in the middle of a sentence.
• Copy. Puts the answer on the clipboard.
• Try Again. Sends the same text again for a different answer. This uses one more request.
• Follow Up. Carries on from this answer as a conversation, so you can say things like “make it shorter” or “and the second point?”. See Having a conversation.
• Requests. How much of your allowance this answer used.

You can undo every edit. Replace and Insert Below work like your own typing, so Ctrl+Z takes an AI edit back.

What you learned, and where to go next

You can replace, insert, copy, try again or follow up, and Ctrl+Z always takes an AI edit back. Next, Having a conversation shows you how to keep talking about an answer.

Having a conversation

Choose Have a conversation in the pad, or press Follow Up in an answer window. The AI Conversation window opens. Here is what is in it:

1. About this conversation. What goes with each message and what it uses.
2. Conversation. Everything said so far, newest at the end, as “You:” and “AI:” paragraphs. You can read it but not change it.
3. Your message. This is where focus stays. Type and press Enter.
4. Send, New Conversation, Copy Last Reply, Insert Last Reply Below, and Close.
5. Status. Still working, what was used, or what went wrong.

Each reply is read aloud as it arrives and added to the conversation, while you stay in the message box. Press Shift+Tab to reach the conversation and read it again at your own pace.

What is sent

Your message, and the conversation so far. If you are following up on a question about your document, the same passages that question used go with every message too, so you can keep asking about them. A conversation you start from the pad sends nothing from your document.

What it uses on the free AI

Each message is one request. Conversations also have their own monthly share: 40 of your 100. To keep every message the same price as any other request, the conversation so far is sent only as far as the ordinary 3,000-token limit allows, newest first. That means a long conversation slowly forgets how it began. The window tells you the first time that happens, and New Conversation starts you fresh. If a single message is too long to fit on its own, it is refused before anything is sent.

With your own AI key

None of those limits apply. The whole conversation goes with every message, and it is billed to your provider’s account, so a long conversation costs more for each reply. The same is true on your ChatGPT plan or GitHub Copilot, where it counts toward the plan instead. It is only shortened if it grows bigger than the model can read at once, and the window tells you when that happens.

What you learned, and where to go next

A conversation remembers what was said, reads each reply aloud, and on the free AI uses one request per message. Next, Usage, About and your support ID shows you how to check what you have left.

Usage, About and your support ID

Usage (Ctrl+Alt+Shift+F9) is one list, with a sentence for each AI you have set up. For the free AI it says what you have used today or this month and when the count starts again, by QUILL’s count, and it asks the service fresh every time you open it. Your ChatGPT plan’s row is a count kept on this computer, GitHub Copilot’s is GitHub’s own count, and your own key’s is what the provider said in its last answer. Each sentence says whose number it is. Manage Usage opens the selected one’s own page. It is all text: no meter, no chart.

Help > About shows the same allowance and support ID, under QUILL’s free AI, once this computer is connected. Give it a moment after About opens, because the numbers are fetched while you read.

If your own AI key is in use, About shows something different: the model you are using and where to find your provider’s usage page, instead of an allowance. See Using your own AI key: no limits.

Your support ID is a short code like A1B2-C3D4. Support needs it to find your account, because there is no name on it. It is in Help > About, the Connect window tells you it when you connect, and Get Help from Support puts it into your message for you.

What you learned, and where to go next

Usage and Help > About tell you where you stand, and your support ID is how support finds you. Next, Limits, and when they start again has every number. The tutorial Asking a question about a document ends with a step on checking what you have left.

Limits, and when they start again

The service is free because everybody shares it, so there is an allowance:

These are the service’s current numbers, and they can change without you updating anything. Usage always shows the numbers that apply to you right now.

When the counts start again

The service keeps time in UTC. The daily count starts again at midnight UTC, which is 8 PM Eastern or 5 PM Pacific in summer, and an hour earlier in winter. The monthly count starts again at midnight UTC on the 1st.

What counts

Only answers count. If a request does not come back with an answer, it uses nothing. That includes a request that was too long, one sent with no connection, one that hit a problem on the service, and one refused because you were already at a limit. Opening Usage or About uses nothing either.

A new connection starts smaller

For its first 48 hours, a newly connected computer has 15 requests. After that, the full allowance applies on its own. Signing out and connecting again starts a new connection, so it will not get you more.

Computers that share an internet connection

A household, a class or a library may have several computers on one internet connection. Those computers also share a few limits on that connection: how many can be connected at once, how many new connections can be made in an hour, and a monthly total. Signing a computer out gives its place back. If a class runs into these limits, write to support.

Need more?

Choose Get Help from Support and say what it is for: a course, a deadline, a book. Support can raise your allowance, and can end the smaller new-computer allowance early. The change applies from your next request. There is nothing to reinstall and nothing to reconnect.

What you learned, and where to go next

You get 100 requests a month, 20 a day and 8 an hour, and only answers count. Next, When something goes wrong explains the messages you might hear.

When something goes wrong

When a request fails, the pad’s Status line says what happened and what to do. Focus moves there, so your screen reader reads it, and you can read it again with the arrow keys. If the pad is not in front at the time, the message is spoken instead, and focus is waiting on it when you come back.

The message ends with an error code, such as QUILL-AI-GATEWAY-QUOTA for a limit. Include it if you write to support. These are the ones you are most likely to hear:

• A limit was reached. The message names the limit and when it starts again. Nothing was used.
• No connection. Nothing was sent and nothing was used. Try again when you are back online.
• The free AI is paused. The service has one spending ceiling for everyone together. When it is reached, the free AI pauses for everybody until the 1st. This is rare, and nothing is wrong with your computer.
• This computer was signed out. Choose Connect or Sign Out and connect again.
• No AI is in use. You signed out of the one in use, so the next AI command says “No AI is in use. Choose one.” and opens Choose Your AI. Nothing was sent.
• The AI you chose cannot take it. If your own key has gone, or OpenAI or GitHub ended your sign-in, you hear so once, for example “Your ChatGPT plan sign-in has ended. Nothing was sent. Press Ctrl+4 to sign in again or choose another.” QUILL Lite never sends it to the free AI, or anywhere else, instead.
• It needs something the AI in use cannot do. The command’s menu name says what, such as “(needs a connection that sees)”. Pressing it sends nothing and asks whether to switch to one that can, with Yes as the default.

What you learned, and where to go next

When something fails, you hear why and what to do, and focus waits on the message. Next, Your privacy explains what is sent and what is kept.

Your privacy

• What is sent: only the text shown in What will be sent, plus your question if you asked one. On the free AI it goes to QUILL’s service and on to OpenAI, which writes the answer. On any other AI it goes straight from your computer to that company, and About this connection in Choose Your AI says which.
• What is kept: how many requests you made, how big they were, which kind of request each one was, and what it cost.
• What is not kept: what you wrote. What came back. Your name, your email address, or anything else that identifies you. Connecting a computer creates an account with no name on it at all.
• When it is sent: only when you press Send. Nothing is sent as you type, when you save, in the background, or when you open the pad. If you close the pad without pressing Send, nothing is sent.

The service’s own page, https://ai.community-access.org, explains the same things and publishes what the service costs to run.

What you learned, and where to go next

Only the text you see in the box is sent, only when you press Send, and your words are not kept. Next, Signing out and changing your mind shows you how to step back out.

Signing out and changing your mind

• To disconnect this computer from the free AI, open Choose Your AI (Ctrl+4), arrow to QUILL’s free AI and press Delete (or Sign Out when it is the AI in use). It asks first, with No as the default. You can connect again whenever you like.
• It is safe to read the agreement again. Once you have agreed, Privacy Agreement (Ctrl+Alt+Shift+K) opens it with two buttons: Keep Using AI, which is what Enter and Escape do, and Withdraw and Sign Out. Nothing changes unless you choose the second one.
• Withdrawing switches AI help off and signs this computer out, both in one step.
• You can always open Privacy Agreement, whether AI help is on or off, and whether or not you have ever agreed.

What you learned, and where to go next

You can sign out in Choose Your AI, and withdrawing your agreement also signs you out. If you would like more than the free allowance, the next chapters explain the other ways to run AI help, starting with the one window where you choose.

Choose Your AI: which AI is in use

QUILL Lite keeps one record of which AI answers you: QUILL’s free AI, your ChatGPT plan, your GitHub Copilot, Codex, or your own key. It is the same setting QUILL uses, ai_connection, so the two programs agree. Copilot’s and Codex’s models are kept in ai_copilot_model and ai_codex_model; empty means the service picks.

Choose Your AI (Ctrl+4, or Tools > AI > Choose Your AI…) is where you see and change it. Focus lands on the Connections list, on the one in use. Each row says its name and then “in use”, “signed in” or “not set up”, such as “My own key, OpenAI, signed in”. Tab moves to About this connection (what it can do, who gets your text and what it costs), then Model (that connection’s own models, the recommended one first), then one button that follows the row (Use This, Sign In… or Sign Out), then Close. Enter on a row does what the button says; Delete on a signed-in row asks whether to sign out, with No as the default. While your ChatGPT plan’s row or Codex’s is selected and signed in, Plan Options… is there too (see Web search, below).

QUILL Lite does not talk over your screen reader in this window. It speaks only what happened: “Now using My ChatGPT plan.”, “Signed in to My ChatGPT plan. It is now in use.”, or “Signed out. No AI is in use until you choose one.”

Which AI Am I Using (Ctrl+8) says in one line which AI is in use, the model, and one thing about usage, such as “My ChatGPT plan, gpt-6-luna. 3 requests today from QUILL Lite, counted on this computer.” Press it twice quickly to open Choose Your AI. The status bar’s AI cell says the same thing shorter, such as “AI: ChatGPT plan”; press F6, arrow to it, and press Enter to open Choose Your AI.

When the AI in use cannot do something, the command stays in the menu with its key and its name says what it needs, for example Ask About an Image (needs a connection that sees) or Do a Document Task (needs Copilot or Codex). Pressing it sends nothing and asks whether to switch to one that can, with Yes as the default. If nothing you have set up can do it, Yes opens Choose Your AI on a row that can.

Use My Own AI Key (Alt+F2), Use My ChatGPT Subscription (Alt+F5) and Connect or Sign Out (Ctrl+Alt+Shift+F10) keep their keys, and each opens Choose Your AI on its own row, ready to sign in.

When you first start this version, QUILL Lite keeps what you already had: your ChatGPT plan if you were signed in, else your key if you had one saved, else the free AI.

After that, signing in is choosing. Sign in with ChatGPT or GitHub Copilot, save a key, or connect the free AI, and that one is in use. Use This switches to a row you are already signed in to. Sign out of the one in use and no AI is in use until you pick one; QUILL Lite never quietly sends your text somewhere else. The next AI command says “No AI is in use. Choose one.”

If OpenAI ends your ChatGPT sign-in, GitHub ends your Copilot sign-in, or a limit is reached, you hear it once. Nothing is sent, nothing is retried, and nothing switches.

QUILL Lite counts the requests it sends on your ChatGPT plan, today and this week, with the token counts OpenAI reports. The count stays on this computer.

Answers are plain text. QUILL Lite asks every AI not to draw charts or diagrams, and anything like that which arrives anyway is shown as text.

Using your own AI key: no limits

If you have a key from OpenAI, Google Gemini, Anthropic, OpenRouter, Ollama Cloud or another OpenAI-compatible service, or Ollama on your own computer, you can use your own key instead of QUILL’s free service. Then every limit goes away: no monthly, daily or hourly allowance, no 2,250-word ceiling, and no smaller first 48 hours.

Setting it up

Choose Tools > AI > Use My Own AI Key (Alt+F2). One window opens. Here is what is in it:

1. Provider. Choose OpenAI, Google Gemini, Anthropic Claude, OpenRouter, Ollama Cloud, a Custom OpenAI-compatible service, or Ollama on this computer, which needs no key. Everything else in the window follows this choice: where your text goes, which key, which models and whose prices. AI help always uses the key saved for the provider you choose here, and never another one. Your choice is saved with your settings, and QUILL never guesses it from a key or a model name.
2. About this. A read-only box saying exactly where your text goes. Arrow through it once.
3. API key. Paste the key for the provider you chose. An OpenAI key comes from platform.openai.com, under API keys, and starts with sk-. A Gemini key comes from Google AI Studio at aistudio.google.com/apikey. The About box says where to get a key for each of the others.
4. Model. A list of every model your key can use for text. It fills in from your account once the key is checked. For OpenAI, Luna 6 comes first, then the other GPT-6 models. For Gemini, 2.5 Flash comes first, then 2.5 Pro. The rest follow by name. Each row says roughly what it costs, for example “gpt-6-luna, about $0.57 per 100 requests (estimate)”, so you can compare them just by arrowing down the list. Models that cannot answer in text (speech, transcription, images, embeddings) are left out.
5. Cost estimate. The chosen model’s estimated cost per request and per 100 requests, at that provider’s prices. These are estimates to help you compare, not the provider’s prices. A typical request here is about a page in and a paragraph back. The real prices are at openai.com/api/pricing and ai.google.dev/pricing.
6. Status. Whether a key is saved for the provider you chose, what the last test said, and whether the other provider’s key is saved too.
7. Test the Key. Checks the key with your chosen provider and fills the model list. Then it sends one tiny request to the chosen model and tells you whether it answered. That request costs a fraction of a cent.
8. Remove the Saved Key. Forgets the chosen provider’s key straight away. If your own key was the AI in use, no AI is in use until you choose one.

Press OK and your key is saved. QUILL Lite says “AI help uses your own OpenAI key, with no limits”, or “your own Google Gemini key”.

Changing the model, or the key

You can change the model at any time. With a key saved, the window lists your models as soon as it opens. Press Alt+F2, pick another, and press OK.

Saving a key is choosing it. Press OK with a key saved and your own key is the AI in use. Remove it while it is in use and no AI is in use until you pick one in Choose Your AI. Your text never moves to the free AI unless you choose it.

Ollama on this computer needs no key. Install Ollama from ollama.com, pull a model, and choose Ollama on this computer. The About box starts with “Runs on this computer. Nothing is sent.”, and that is exactly what happens: the model runs on your computer, nothing is billed and nothing leaves it. The key box is off, a Server address box says where Ollama listens (usually http://localhost:11434), the Model list is whatever Ollama has, and the test button reads Test the Connection. A custom service has a Server address too. These providers used to be set up in QUILL’s AI Hub; they are the same providers, with the same saved keys, in this one window in both editors.

You can save both keys. The Provider list decides which one is used. If you choose OpenAI, a saved Gemini key just sits there unused. The status line tells you it is there, and choosing Google Gemini is how you use it. Your request never goes to a company you did not choose.

With a Gemini key, Ask About an Image works too. Gemini’s models can read pictures, so Ctrl+F5 works on a Gemini key just as it does on a ChatGPT subscription, and so does a picture-reading model in Ollama on this computer. The free service and the other providers here cannot.

What changes with your own key

• Your text goes straight from this computer to the provider you chose, on your account. Nothing passes through QUILL’s servers, so QUILL records nothing at all, not even the count it keeps for the free service. With Ollama on this computer, nothing leaves the computer.
• That company bills you for each request, under its own terms and privacy policy. Ollama on this computer bills nothing.
• Usage (Ctrl+Alt+Shift+F9) has a row for your key: how many requests are left this minute, from the provider’s last answer, where it reports one. Manage Usage takes you to your provider’s usage page, where your requests and charges are. Help > About also shows the model and that page’s address instead of the free allowance.
• You do not need to connect this computer, and you are not asked for the free service’s agreement. That agreement is about QUILL’s servers, and your own key never uses them.
• The pad, the things it can do, and what comes back are exactly the same, right down to the instructions sent with your text. The pad still warns you before sending something very large, because with your own key a large request costs you money.
• While an answer is on its way, Send becomes Stop (Alt+S). Stopping closes a Gemini stream immediately; for another route it discards that request’s eventual answer. It never stops a request from another AI window.

Where your key is kept

Your key is kept in Windows’ own credential store, not in a settings file. Once saved, it is never shown again. The box stays empty and tells you a key is saved. A portable copy keeps it in an encrypted file instead. QUILL and QUILL Lite share the key: save it in either and it works in both, and remove it in either and it is gone from both.

If a request fails

You hear why, followed by the error code QUILL-AI-OWN-KEY-FAILED, and the sentence names the provider. The usual causes are a mistyped key, a model your account cannot use, an account with no credit, or the wrong provider chosen for the key you saved. Test the Key tells you which one it is.

What you learned, and where to go next

With your own key saved and its provider chosen, AI help has no limits and your text goes straight to that provider on your account, or nowhere at all with Ollama on this computer. Remove the key while it is in use and no AI is in use until you choose one. If you already pay for ChatGPT, Using your ChatGPT subscription is next, and it needs no key at all.

Using your ChatGPT subscription

If you already pay for ChatGPT, QUILL Lite can run its AI commands on your plan, and you never have to handle a key. You sign in once in your browser, the same way you sign in to ChatGPT itself. From then on, the pad, the seventeen things it can do, Ask About This Document and conversations all go to OpenAI on your plan. There is no per-request bill, and nothing goes through QUILL’s servers. Requests draw on your plan’s own usage limits, shared with ChatGPT and any other app you have connected; ChatGPT’s Settings > Usage lets you set a weekly limit for each app. When the limit is reached, QUILL Lite says so and does not switch to an API key. You also get two things the free service cannot do: pictures, described, and web search for an answer about today. The same sign-in also unlocks Codex, for document tasks.

Signing in, step by step

1. Press Alt+F5 (Tools > AI > Use My ChatGPT Subscription). Choose Your AI opens on My ChatGPT plan, with its sign-in step ready.
2. The step says “To use My ChatGPT plan, approve QUILL Lite on OpenAI’s page.” Press Open the Page. Your browser opens on OpenAI’s own sign-in page.
3. Sign in to ChatGPT there. A page asks whether QUILL Lite may use your plan. Allow it. That is the name you will see under Apps in ChatGPT’s settings from now on. Each QuillVille app signs in on its own: QUILL, QUILL Cast and Quill Radio each ask under their own name, and signing out of one leaves the others signed in. If you are already signed in to ChatGPT in your browser, this step is just one Allow, with no password.
4. Come back to QUILL Lite. The step closes by itself and you hear “Signed in to My ChatGPT plan. It is now in use.” You are done.

If the browser did not open, the step shows the address. Paste it into any browser on this computer and finish there. Cancel gives up without changing anything.

Choosing the model

The Model list in Choose Your AI, on the plan’s row, is read from your account, so it holds exactly what your plan offers, in OpenAI’s own order with the first marked “(recommended)”. Models that cannot answer in text are left out. Your choice is saved as soon as you make it, so there is no OK to press. If the list cannot be read, the window says so and the recommended model is used.

What changes

• Every limit goes away. Nothing is refused for size, the answer is not cut short, and Ask a question about the document sends the whole document instead of three passages. Before you send, the About to send line tells you how many words are going and that they count toward your plan.
• What you use counts toward your plan’s usage. OpenAI keeps track of that and shows it on ChatGPT’s own usage page. When you reach a limit, QUILL Lite tells you in words, once, with the code QUILL-AI-CHATGPT-LIMIT, and the usage page shows when it resets.
• Usage (Ctrl+Alt+Shift+F9) has your plan’s row: the requests QUILL Lite sent today and this week, with the token counts OpenAI reported, counted on this computer and always called that. Manage Usage opens ChatGPT’s own usage page. Help > About names your account and model instead of the free allowance.
• You are not asked for the free service’s agreement, and this computer does not need to be connected to it. That agreement is about QUILL’s servers, and your subscription never uses them.
• A saved key does not compete with your plan. Whichever one you chose in Choose Your AI is in use, and the other waits. Sign out of the plan and no AI is in use until you choose one; the key does not take over by itself.

Signing out, and forgetting

Sign Out, in Choose Your AI while the plan is in use (or Delete on its row, which asks first), asks OpenAI to cancel QUILL Lite’s sign-in and forgets it on this computer. Forget on This Computer, in Plan Options…, only forgets it here. Use it on a computer that is offline, or when you have already removed the sign-in in ChatGPT’s settings. Until you remove it there, QUILL Lite stays listed in ChatGPT’s settings. Either way, if the plan was in use, no AI is in use afterwards until you choose one.

Where the sign-in is kept

Only a refresh token is stored, in Windows’ credential store (or an encrypted file in a portable copy), under QUILL Lite’s own name. QUILL Lite never sees your ChatGPT password. The access token that does the work stays in memory and is renewed when needed. Each QUILL app signs in as itself. QUILL Lite, QUILL and Quill Radio each appear under Apps in ChatGPT’s settings, so signing one out leaves the others as they were.

What you learned, and where to go next

Press Alt+F5, sign in once in your browser, and your AI commands run on your ChatGPT plan with no allowance and no key. Next, GitHub Copilot and Codex covers the two connections that can carry out document tasks.

GitHub Copilot and Codex

My GitHub Copilot uses your GitHub Copilot plan, Copilot Free included. Sign In… on its row in Choose Your AI opens the same small sign-in step as the plan, with one more field: a short code from GitHub, spaced out so your screen reader reads it one character at a time, said once, and already on the clipboard. Press Open the Page, paste or type the code on GitHub’s page, and approve QUILL Lite. The step closes by itself and you hear “Signed in to My GitHub Copilot. It is now in use.” On Copilot Free the only model is “Auto (Copilot Free chooses)”. QUILL Lite signs in to GitHub on its own, apart from QUILL and the other apps.

Signing in from inside QUILL Lite needs QUILL Lite’s own registration with GitHub. Until a build carries it, the row says “not available yet”, and About and Sign In… both say “Not available yet: GitHub sign-in for QUILL Lite has not been registered, so it comes in a later update.” Nothing goes to GitHub until then.

With Copilot in use, the pad, Ask, the Prompt Library and Ask About an Image (on a Copilot model that can see) go to Copilot. Which AI Am I Using and Usage include GitHub’s own count, such as “41 of 300 premium requests this month, by GitHub’s count”; the first time, Ctrl+8 waits a moment for it. A Copilot error is said once; nothing is retried or switched.

Codex, for tasks runs on your ChatGPT plan with the same sign-in, so once you are signed in to the plan, Use This on the Codex row is the only key to press. While Codex is in use, single requests (the pad, Ask, Ask About an Image) go straight to your plan, and Ctrl+8 says “Codex, on your ChatGPT plan. Single requests go to the plan directly.” The Codex program itself starts only for a document task. If this copy does not include the Codex engine, the row says “not available yet”.

What you learned, and where to go next

Copilot signs in with a short code from GitHub, and Codex is one keypress once your plan is signed in. Next, Do a Document Task shows what they are for.

Do a Document Task

Do a Document Task (Ctrl+3, in Tools > AI) hands a job with several steps to GitHub Copilot or Codex, whichever is in use: “Fix the heading levels, then build a table of contents”, or “Turn these notes into actions grouped by owner.” On any other AI its name says “(needs Copilot or Codex)”, and pressing it sends nothing and offers to switch.

The window has six things, in Tab order: Conversation, a read-only record of the task; Status, which says “Using GitHub Copilot” or “Using Codex on your ChatGPT plan” and whether it is working; Instruction, where focus starts; Send; Stop; and Close. Type what you want done and press Enter. While the task runs, anything you type goes to it as a follow-up. Stop ends it at once, and Close stops a task that is still running.

Nothing changes without you. Before a step that changes something, Allow This Step? says in one sentence what will happen. Enter is Allow once; Allow for this task stops it asking about that kind of step until the task ends; Escape is Don’t allow. Every edit then opens Review AI Changes, so you accept or reject each change, and each one undoes with Ctrl+Z. The next task asks again.

When the task ends you hear its summary, such as “Done. 1 change applied: Changed the document. Each can be undone with Control Z.” It is built from what really happened and also says how many changes you declined. The AI’s own summary is in the conversation, one Shift+Tab away.

A task can only read your document, propose edits and run a short list of safe QUILL Lite commands. It cannot reach the web or your files, and it answers in plain text, never charts or diagrams. What the engine keeps while it works is deleted when the task ends or you press Stop.

What you learned, and where to go next

Ctrl+3 hands a job to Copilot or Codex, every step that changes something asks first, and every edit is reviewed and undoable. Next, Ask About an Image and Web search cover pictures and looking things up.

Ask About an Image

Tools > AI > Ask About an Image (Ctrl+F5) tells you what is in a picture. That could be a screenshot a colleague sent, a photo of a letter, a chart pasted into a report, or a menu from a restaurant’s website. Choose the file, ask a question or ask nothing, and hear what the model sees.

1. Press Ctrl+F5. Focus is in Image file.
2. Type or paste the file’s path, or press Browse (Alt+B). JPEG, PNG, WebP and GIF files work.
3. Your question (Alt+Q) is optional. Leave it empty and you get a description written for a blind reader: what the picture is, what matters most, and every word of any text in it, copied exactly. Or ask something, like “what does the sign say” or “what is the total on this receipt”, and the answer comes first.
4. Press Enter. You hear “Working.” The picture is checked on your computer first. A file that is not really an image, or one over 20 MB, is refused with a sentence telling you why, and nothing is sent. Otherwise it goes to the AI in use.
5. The Image Description window opens with focus on the text. Insert Below puts it into your document under the paragraph you are in, which is a handy way to caption a picture. Copy puts it on the clipboard, and Ctrl+Z takes an insert back.

You need an AI that can see pictures: your ChatGPT plan or Codex, a GitHub Copilot model that sees, a Gemini key, or a picture-reading model in Ollama on this computer. On any other AI the menu says Ask About an Image (needs a connection that sees), and pressing it sends nothing and offers to switch. Only the picture and your question are sent. Nothing from your document goes with them.

This is the only AI in QUILL Lite that looks at a picture, and only at a file you choose. Nothing looks at your screen, and no AI here makes pictures or charts: every answer is text.

What you learned, and where to go next

Ctrl+F5 describes a picture, or answers your question about it, and Insert Below turns that into a caption. Next, Web search explains how to let the model look things up.

Web search

On your ChatGPT plan, check Allow web search in Plan Options…. Then the model may search the web through OpenAI when your question needs it, such as an event, a price, or what a page says today, and tell you what it found. This works for Ask a general question, Ask a question about the document, and conversations.

It is off until you turn it on. A search sends your question somewhere else as well, so that is your choice to make. Your choice is saved as soon as you make it.

To turn it on:

1. Press Alt+F5. Choose Your AI opens on My ChatGPT plan.
2. Tab to Plan Options… and press it.
3. Press Alt+W, or Tab to Allow web search, and press Space. It is saved straight away. Press Escape to close the options.

Press Space on the same check box to turn it off again. The free AI has its own ten searches a month instead, through the pad’s Search the web choice and Web Searches Left (Ctrl+9); with your own key there is no web search.

What you learned, and where to go next

On your plan, web search is off until you check Allow web search in Plan Options, and then the model can look up current answers. Next, Tidy Dictated Text shows how AI can clean up what you dictated.

Tidy Dictated Text

Dictation writes down what it heard, and sometimes it hears wrong. Tools > AI > Tidy Dictated Text (Ctrl+F3) fixes what was misheard: sound-alike words, names, words run together or split apart, punctuation, capital letters, fillers and false starts. It changes nothing else.

1. Leave your cursor in the dictated paragraph, or select the part you want tidied.
2. Press Ctrl+F3. You hear “Working.”
3. The Tidied Dictation window opens on the corrected text.
4. Press Replace My Selection to put it back where the dictated text was. Ctrl+Z undoes it in one step. Press Escape to keep what you had.

This runs on any AI except the free one: your ChatGPT plan, GitHub Copilot, Codex or your own key. On the free AI it tells you so and opens Choose Your AI on your plan’s row.

What you learned, and where to go next

Ctrl+F3 fixes what dictation misheard and leaves the rest alone. Next, The Dictionary: ask about the word you are on covers questions about a single word.

The Dictionary: ask about the word you are on

Tools > Dictionary is a dictionary that reads the sentence your word is in. You can also reach it from the Dictionary for “word” submenu when you press the Applications key on a word. The thesaurus that comes with QUILL Lite knows every meaning of bank, but it cannot tell which one your sentence means. These commands send the word and the sentence it sits in, never the rest of the document. The answers come back in plain sentences that are easy to listen to, and where it helps, with a list of choices that each say why.

The questions you can ask

• Define in Context (Ctrl+Alt+Shift+’)
• Synonyms That Fit (Ctrl+Shift+;)
• Simpler Word (Ctrl+Shift+-)
• More Formal Word (Ctrl+Shift+=)
• More Vivid Word (Ctrl+Shift+’)
• Opposites (Alt+Shift+-)
• Is This the Right Word? (Ctrl+Alt+Shift+/)
• Use It in a Sentence (Alt+Shift+=)
• Where It Comes From (Alt+Shift+,)
• How to Say It (Alt+Shift+’)
• Rhymes (Ctrl+Alt+Shift+;)
• Word Explorer (Ctrl+F10), which answers all of them at once.

Find the Word For… (Ctrl+Alt+Shift+]) works the other way round. Describe the meaning you are after, hear the words that might fit and how each one differs, and insert the one you choose at your cursor.

The answer window

The answer window opens with focus on the answer, so your screen reader starts reading it. Tab to reach Choices. Press Enter, or Use This Word, to put the selected word in place of yours. Ctrl+Z undoes that in one step. This only works while your word is still where it was. If you type over it while the answer is on its way, the button is turned off and tells you why. You also find Copy Choice, Copy Answer, and Ask Something Else…, which opens the Dictionary submenu again on the same word.

What it needs

It runs on any AI except the free one: your ChatGPT plan, GitHub Copilot, Codex or your own key. People look words up many times a day, and that would use up a free allowance very quickly. On the free AI, each command tells you so and opens Choose Your AI. The user guide’s Dictionary section has the full table, and Look Up in that section is the dictionary without AI.

What you learned, and where to go next

Tools > Dictionary answers questions about one word in its own sentence, and Find the Word For… helps when the word is on the tip of your tongue. Last, Questions people ask answers the things people ask most.

Questions people ask

Do I need an account? No. No name, no email, no password. The eight-character code connects this computer, and that is all.

Does it cost anything? No. It is paid for by Community Access. If you use your own AI key instead, that provider bills your account and there are no limits; Ollama on this computer costs nothing. If you sign in with a ChatGPT subscription or GitHub Copilot, what you use counts toward that plan and nothing else is billed.

I pay for ChatGPT. Do I still need a key? No. Press Alt+F5, press Open the Page, and sign in the way you do on the web. That is the whole setup, and it is the way to have pictures described and the web searched.

Which AI am I using? Press Ctrl+8. Press it twice to change it.

Can it describe a picture? Yes, on an AI that can see: your ChatGPT plan, Codex, a Copilot model that sees, a Gemini key, or a picture-reading Ollama model. Press Ctrl+F5, choose the file, and listen. See Ask About an Image.

Can it do a job with several steps? Yes, on GitHub Copilot or Codex. Press Ctrl+3. See Do a Document Task.

Can it change my document without asking? No. Nothing goes into your document until you press Replace My Selection or Insert Below, and Ctrl+Z takes either one back.

Does it read my whole document? Only when you ask a question about the document, and even then only the three passages it picks are sent. The pad shows you which ones before you press Send.

Why is my allowance 15 when I was told 100? This computer connected less than 48 hours ago. The full allowance applies on its own when that time is up, and Usage tells you exactly when.

Why did my daily requests come back in the evening? The service keeps time in UTC, so in North America the new day starts in the evening.

Who do I ask for help? Choose Get Help from Support in the Help menu, or write to support@community-access.org. A person reads every message.

What you learned, and where to go next

That covers the questions people ask most. To practise with the app guiding you, open Help > Tutorials… (Ctrl+Alt+F1) and choose Asking a question about a document. The QUILL Lite user guide’s AI help chapter covers the same ground in the context of the rest of the editor.