Run prompts in Chat

Chat with the local model or your own AI provider, run your saved prompts in the conversation, and keep, rename or delete your chats.

The Chat screen runs your prompts with an AI model and shows the replies, with your prompts in a panel beside the conversation. It works with the local model on your computer, or with an AI provider you set up with your own API key or your own server. This page shows how to use it; Chat and context explains what each message sends, how long chats are handled, and where chats are kept.

Chats are on every plan and don’t count toward the Free limits.

Before you start

• Chat must be switched on. If Chat & Browse isn’t in the sidebar, click Settings, then Features, and turn on Chat. See Show or hide features.
• You need an AI provider. Download the local model, or add an API key (Set up AI). Until one is set up, the Chat tab shows Set up AI to start chatting. Click Download the free local model to open Settings → AI & connected apps at the local model, where you download it. You can add an API key in the same place.

The Chat screen

Click Chat & Browse in the sidebar, or press Ctrl+9. The screen has three areas:

• Chats on the left: your saved chats, newest first.
• The conversation in the middle, on the Chat tab. With web tabs on, websites open in more tabs beside it.
• Prompts on the right: your prompts and blocks, to run in the chat.

The Chats and Prompts buttons at the top right show or hide the side areas. Drag the line between two areas to resize them. In a narrow window, if both side areas don’t fit, the one you opened last stays open.

The screen stays loaded while you use other screens, so a reply keeps arriving while you’re away.

Start a chat

1. Click Chat & Browse in the sidebar.
2. Click New chat at the top of the chats list, or press Ctrl+N when you’re not typing in the message box.
3. Optional: under Model, choose the provider and model for this chat (see Choose the model).
4. Type your message in the box at the bottom. Enter sends it; Shift+Enter starts a new line.
5. Click Send.

The reply appears under your message as it’s written, with “Writing…” until the first words arrive. The provider and model are shown under the reply.

A new chat is saved when you send its first message. Its name is taken from that message until you rename it.

While a reply is being written, Send becomes Stop. Click Stop to end the reply early; what arrived so far is kept, marked “Stopped.” You can type your next message while a reply is being written, and send it once the reply is done.

The lines under the chat’s header

Under the Model picker, a grey notice starts with a short line that says where this chat’s messages go:

The notice shows only that first line at first, with a small arrow at its right. Click the line to open the rest of the notice, and click it again to close it. PromptMason remembers whether you left it open.

In the open notice, with a cloud provider, the last line shows how many tokens the chat has sent so far, beside How chat memory works, for example “Sent in this chat: 18,000 tokens to OpenAI”. A token is about three quarters of a word. The count covers everything each message sends, including the earlier part of the chat that goes with it (more).

If the chat can use your library, a line in the open notice says so (see Let the chat work in your library).

How chat memory works opens a short explanation of what each message sends, with a link to Chat and context.

Choose the model

A new chat starts with your default provider and its first model, set under Default AI provider in Settings → AI & connected apps.

To use a different model for this chat, click the Model picker and choose a provider, then a model. The picker lists only the providers you’ve set up, with the local model first. The choice applies to this chat only. You can’t change it while a reply is being written.

If the chat already has messages and you choose a different cloud provider, PromptMason asks first, because your next message sends the recent conversation to that provider. This happens when you move from the local model to a cloud provider, and from one cloud provider to another. In the Switch this chat to [Provider]? dialog:

• Switch moves this chat to the new provider. Nothing is sent until your next message.
• New chat starts a new, empty chat on the new provider. This chat is kept as it was.
• Cancel keeps the current provider.

Moving a chat to the local model never asks, because none of the chat leaves the computer. Choosing another model of the same provider doesn’t ask either.

While the local model isn’t downloaded, the Model row also has a Download the free local model button. It opens Settings → AI & connected apps at the local model, where you download it. Once it’s downloaded, the button isn’t shown.

If you remove a provider in Settings, a chat that used it shows “[Provider] (not set up)”. Choose another provider to carry on.

Run a prompt from the Prompts panel

1. In the Prompts area on the right, find a prompt, block or document. Type in Search prompts and blocks to narrow the list.

2. Click it. A preview shows the text that will be sent.

   

3. Click one of these:

   • Run in chat sends it to the chat now. If no chat is open, a new one starts.
   • Copy copies it to the clipboard. To put a prompt in the message box so you can change it first, copy it and paste it there. For a Markdown document, Download .md saves it as a file.

4. If the prompt has variables that need a value, Fill in your values opens first. Type a value in each box, then click the form’s Run in chat button.

A line above your message reads “Ran” followed by the prompt’s name. With no AI provider set up, the panel offers only Copy.

When a web tab is showing instead of the Chat tab, the panel offers Run in this tab. See Web tabs.

Work with replies

• Replies are formatted: headings, lists, tables and code. Links open in your web browser, except links to PromptMason’s guides, which open in Help.
• Copy under any message copies its text.
• Try again appears under the last reply when it failed or was stopped. It asks for the reply again.
• The view follows a reply as it grows. Scroll up to read something earlier and it stops following.

Use a skill

A skill gives the chat instructions to follow from its next message on, such as how to build a prompt with you. The skill is sent with every message in the chat.

1. In the chat’s header, click the Skill picker (it shows No skill at first).
2. Choose one:
   • Ask about PromptMason, built in: “Ask how to do something in PromptMason. The answers come from its user guides.” See Ask how to use PromptMason.
   • Build a prompt, built in: “Write a prompt with the AI, block by block, and save it to your library.”
   • Any of Your skills: your Markdown documents whose Type is Skill. See Markdown documents and Skills.
   • No skill to stop using one.

When a skill saves something to your library, Chat asks first, like any other change (Approve or refuse a change).

In a new, empty chat, the Start with a skill buttons do the same. Choosing a skill sends nothing by itself.

A skill must be under 20,000 characters. If a skill’s document is deleted, or its Type is no longer Skill, the picker shows Skill not found. Choose another skill, or No skill.

To read or change the built-in skills’ wording, edit the Skill: Ask about PromptMason and Skill: Build a prompt entries in the AI instructions.

Ask how to use PromptMason

Chat can answer questions about PromptMason itself, such as “How do I turn on backups?” or “Where do I change the theme?”, from these guides. They’re the same guides as in Help, as they were when your version of PromptMason was made.

• In any chat: ask your question. With a cloud provider, the chat can look in the guides for any message. With the local model, it looks in the guides only when your message asks how, where, what or why about something in PromptMason.
• In a chat just for questions: click Ask about PromptMason at the top of the Help screen, or choose Ask about PromptMason under Start with a skill in a new chat. A new chat starts with that skill, and every message in it can look in the guides.

While the chat looks something up, a line in the reply says “Searched the guides for” followed by your question. The reply gives the guide’s steps, with the names of buttons and settings in bold as the guide writes them, and ends with a link to the guide. Click the link to open that part of the guide in Help. If the guides don’t cover your question, the reply says so.

What goes where:

• With the local model, your question and the parts of the guides it looks up stay on your computer.
• With a cloud provider, your question and the parts of the guides the chat finds are sent to that provider, with the rest of the chat. Nothing from your library goes with them.

Looking things up in the guides reads nothing in your library, so it works whatever What connected apps can do is set to, Off included.

Let the chat work in your library

A chat can find, create and change blocks, prompts and folders in your library when you ask it to, for example “Make a Constraints block that says to keep answers under 200 words.” Chat asks you before it changes anything (see Approve or refuse a change). It uses the same setting as AI agents and the Chrome extension: What connected apps can do in Settings → AI & connected apps. With a cloud provider, whatever the chat reads from your library is sent to that provider, and with Full access (the default) a cloud chat can also change your library, after you approve each change. See Privacy in PromptMason.

Click Change on that line to open the setting. (The line is in the header’s notice, which you open by clicking its first line.) Changing it also changes what AI agents and the Chrome extension can do. See Control what connected apps can do.

While the chat works, a line in the reply shows “Working…”, then what it did, with Open to go to a block or prompt it made or changed. If something couldn’t be done, the line starts with “Couldn’t” and says why; the reply carries on. Anything a chat deletes goes to Recently deleted, where you can restore it (Archive, delete and restore).

Approve or refuse a change

Reading and searching your library never ask. Before Chat changes anything, it stops and shows a card in the reply that asks Allow this change to your library? The card says what it wants to do, for example Create prompt “Cold email” or Delete block “Tone”, with a short look at what will be saved. Nothing changes until you answer.

To answer from the keyboard, press Shift+Tab in the message box to reach the card, then press Enter on a button. While the card waits, the line under the message box says the reply is waiting for your answer. There is no time limit on your answer.

Things to know:

• Allow for this chat lasts only for that conversation, and only until you close PromptMason. A new chat asks again.
• Clicking Stop while a card waits counts as Don’t allow. So does switching to another chat, or deleting the chat.
• This applies to Chat only. AI agents such as Claude Desktop, and the Chrome extension, don’t show these cards: what they can do is set by What connected apps can do alone.

With the local model

The local model is smaller, so it gets only the kinds of tools switched on for it, and only when your message asks it to do something in your library. A question gets an answer, not a search of your library; a question about using PromptMason gets the guides (Ask how to use PromptMason). To choose the kinds of tools:

1. Click Settings in the sidebar, then Chat.

2. Under Local model tools, turn each kind on or off: PromptMason guides, Find, Blocks, Build prompts, Organize, Tags, Archive, Delete, More prompt tools, Browse everything, Variables, Notes and expected outputs and Libraries.

   

The kinds marked “Not recommended for this model” are off at first, because the local model doesn’t handle them reliably. These switches apply once the local model is downloaded. With PromptMason guides off, the local model doesn’t look in the guides, except in a chat started with Ask about PromptMason. Cloud providers can use everything What connected apps can do allows. Either way, Chat asks you before it changes anything in your library.

Save a reply as a prompt

When a reply contains a prompt written as prompt code (for example from the Build a prompt skill), Save as prompt appears under it.

1. Click Save as prompt. Build opens with a new, unsaved prompt on its Code tab.
2. Check the blocks, name the prompt and save it as you would any prompt. See Edit a prompt as code.

Rename or delete a chat

The rename and delete buttons appear when you point to a chat in the Chats list.

Rename:

1. Point to the chat and click the pencil (Rename).
2. Type the new name (up to 120 characters) and press Enter. Esc cancels.

Delete:

1. Point to the chat and click the trash can (Delete).
2. In Delete this chat?, click Delete chat.

Deleting a chat removes it and its messages from this computer. It can’t be undone, and it can’t delete what a cloud provider already received.

To delete every chat at once, click Settings, then Chat, then Delete all chats, and confirm with Delete all chats. This can’t be undone either. Your blocks, prompts and settings aren’t affected.

If something goes wrong