Assistant Chats & History

The Assistant is Pia’s main chat. Everything you talk about is saved automatically and waiting for you whenever you come back. When cloud sync is on, your chats sync across your devices, and with end-to-end encryption on they reach your Pia account encrypted.

Start Chatting

Type in the message box where it says Type a message…
Press Enter

A brand-new chat shows Start a conversation with Pia until you send your first message.

A link in one of Pia’s replies opens when you click it only if it points to a web page (http:// or https://) or an e-mail address (mailto:). A link to a Windows setting, another program or a file on your computer doesn’t open on a click.

Writing more than one line

Enter on its own sends the message. To add a line break without sending, hold a modifier — Shift+Enter is the usual one — and press Enter. The message box grows as you type, so you can write a long, multi-paragraph message before sending it.

How a chat gets its title

Every chat is titled from your first message: Pia takes the opening of what you typed and shortens it into a label. This happens on your device, costs nothing, and happens whether or not auto-titling is switched on. That is why a chat still gets a sensible name with auto-titling off.

Auto-generate chat titles (under Settings, off by default) is an upgrade on top of that. When it’s on, Pia makes one small extra AI request after the first reply and replaces the derived title with a better-written one.

Renaming a chat

A chat can be given a name of your own, and it keeps that name from then on. There are two places to rename one:

Chat history — pick the chat and use Rename, which sits next to Resume.
The chat picker at the top of the Assistant — hover a row, click the pencil, type, and press Enter. The list stays open around you, so you can rename several in a row.

Renaming earns its keep above all with auto-titling switched off, when the title is otherwise just the opening of your first message.

The message bar, button by button

The row along the bottom of the Assistant is the message bar. None of its buttons carry a text label, so here is what each one is, left to right. Hovering any of them shows the same name.

To the left of the message box

Button	What it does
New chat (+)	Opens a fresh empty chat and leaves a running reply to finish in the one it belongs to. Asks first if you have typed a message you haven’t sent.
Delete chat (trash can)	Deletes the chat you are in.
Cancel (stop)	Only visible while a reply is being written. Stops it.
Record (microphone)	Dictate your message instead of typing it.
Attach file (paperclip)	Attach a file to your message as a chip.
Capture screen	Pick a display or an open window and attach a picture of it.
Join a meeting and transcribe	Have Pia sit in a Microsoft Teams meeting and transcribe it.
Live transcription with consent	Transcribe what’s being said in the room around you.
Hand this to your Pia server as a background assignment (rocket)	Run the work on your server instead of in this chat. Only present when your server offers it.

To the right of the message box

Control	What it does
Persona picker	Which persona answers. Shows each persona’s glyph, name and accent dot.
Chat \| Agent lever	Whether what you type is a message or a goal.
Run in background	In Agent mode only — detaches the run so you can carry on elsewhere.
Send	Sends the message. Enter does the same thing.

Record versus the two transcription buttons

Three of those buttons capture audio, and they do quite different things:

Record dictates for you. It turns your speech into text in the message box, which you then send like anything you typed. It is one-way input into a normal chat. See Speech.
Join a meeting and transcribe puts Pia into a Microsoft Teams meeting as a participant. It captures the meeting audio and produces a transcript split by speaker. See Meetings.
Live transcription with consent captures the audio in the room you’re sitting in, through your microphone, for an in-person conversation with no online meeting involved. See Live Transcription.

The short version: Record is you dictating a message, and the other two produce a transcript of a conversation between several people. Meeting transcription needs an online meeting; live transcription does not.

New chat and Delete chat are two different buttons

The message bar keeps starting a chat and destroying one deliberately apart:

New chat (+) opens a fresh, empty chat in the same working folder. It is additive — a reply still being written carries on in the chat it belongs to and tells you when it’s finished, and the chat you left stays in Chat history.
Delete chat (trash can) deletes the conversation you are in. A reply still being written is cancelled, and any approval card still waiting for your answer is withdrawn, so those tool calls never run.

Either button, once it goes through, leaves you with an empty message box, and anything you had typed goes with the chat. Because the + is easy to mistake for attaching a file, New chat checks first: if the message box holds a message you haven’t sent, Pia asks “Start a new chat?” — “Your unsent message will be discarded.”

Yes opens the new chat and throws your draft away.
No keeps you in the chat you’re in, with your text still in the box.

With an empty message box, or a single stray character in it, New chat opens straight away without asking. The chat title chip’s own New chat entry asks the same question. Delete chat has its own confirmation before it deletes anything, and doesn’t ask about your draft on top of that.

To attach a file, use Attach file (paperclip). To empty the message box without touching the conversation, select the text and delete it.

Stopping a reply that’s already generating

While Pia is writing a reply, a Cancel button appears in the message bar, beside the Delete chat button. Click it to stop the reply where it is. Whatever had already been written stays in the chat, and you can send a new message straight away.

The Cancel button is only there while a reply is streaming — it disappears once the reply finishes, which is why you won’t find it on an idle chat.

Referencing your own items with `@`

Typing @ in the message box opens a picker that lets you point Pia at something specific you already have — a memory, a todo, a reminder, or a file in the chat’s folder. Nothing on screen advertises this, so it is worth knowing it exists:

Type this	To reference
`@Files`	A file in the chat’s assistant folder
`@Memory`	Any page or record in your memory vault
`@Todo`	A task
`@Reminder`	A reminder
`@Routine`	A routine

Files sits at the top of the list and is picked by default, so typing @ and pressing Enter goes straight to your files. It is missing only if you turned the file tools off under Settings → Assistant.

Memory lists your whole vault — every topic page and every record it holds, filtered as you type. Tagging one points the assistant straight at that page instead of leaving it to search.

Keep typing after the @ to filter the list, and pick an entry to insert the reference. An @-tagged message also narrows the tools Pia may use on that turn to the matching kind of item, so it is a way of being precise as well as quick. @Files on its own is the exception: the file is context, and every tool stays available.

Tagging a routine with @Routine and asking for it to be run starts that routine. Pia asks you to confirm first, then the routine runs with its own working folder, persona and tool grants, and its answer arrives as a new chat.

Switch Between Chats

There are a couple of quick ways to move around:

Press Ctrl+H to open the quick-switcher overlay. Search for any chat and jump straight to it.
Click the chat title at the top to open a flyout with your recent chats. From here you can also start a fresh one with + New Chat.

Find Past Chats

Open the Chat history page to browse everything you’ve talked about. You can search and filter your past chats by date, provider, and status. Chat history lists every conversation from the moment you open it, however old — pick a start date yourself to narrow it down.

Marking a chat as a favorite

A star sits beside every row in Chat history and in the chat list that drops down from the title. Starred chats collect in a Favorites group above the date groups, however old they are, so a conversation you keep coming back to is one click away rather than a search.

Favorites do two other useful things:

A favorite is never cleared out. “Delete chats not opened for N days” does not apply to a starred chat, so one you marked to keep stays whatever happens. Deleting a favorite by hand still works as normal.
Favorites sync. A chat starred on one machine is starred on the next, end-to-end encryption included.

Opening a long conversation

A long chat opens on its most recent messages rather than loading the whole thread, so it appears at once instead of after a wait. Show older messages above the conversation brings the rest back a block at a time, and the view keeps your place while it does.

The preview pane in Chat history works the same way, and it stays where you put it when the list refreshes behind it.

Chat chip or Chat history — which one to use

The chat title chip and the Chat history page both list your chats, and they are meant for different jobs:

The chat title chip at the top of the Assistant is a switcher. Its tooltip reads “Switch between recent chats”. It shows a short list of recent chats so you can hop between them without leaving the conversation, and it’s where + New Chat and the working-folder picker live. Ctrl+H opens the same idea as a search-as-you-type overlay.
The Chat history page is a browser. It shows everything, not just the recent few, and it gives you search plus filters by date, provider and status. Use it when you’re looking for something from a while back rather than flipping between what you’re working on now.

Importing and exporting chats

The Chat history page can bring conversations in from a file and write them out again:

Import (“Import chats from a Pia archive or an Open WebUI export”) reads a Pia archive or a JSON export from Open WebUI, so chats you had there carry on in Pia. Pia recognises which of the two the file is. The model’s thinking in an Open WebUI chat lands behind the same collapsible heading Pia uses for its own reasoning, instead of in the middle of the answer.
Export all writes every stored chat to one Pia archive you can import again later.
A single chat has Export as Markdown and Export as Pia archive buttons in the preview pane.

A few things to know about importing:

Chat history keeps text only. Attachments in the file are left behind, and Pia tells you how many.
A chat you already have is skipped if the stored copy is the same or newer — so a chat you carried on in Pia is never overwritten by an older file.
Importing the same file again repairs earlier imports. Open WebUI chats you brought over before Pia handled their thinking properly are updated in place.
Imported chats show up straight away. After an import, Chat history clears its search and filters and widens the date range back to the oldest chat it brought in, so old conversations aren’t hidden by the date filter.
A file Pia can’t read is refused as a whole. If it is neither kind, Pia says “This file is neither a Pia chat archive nor an Open WebUI export.” and imports nothing.

Run Several Chats at Once

You don’t have to wait for one chat to finish before starting another. Open a + New Chat while another is still working — the background chat keeps running and lets you know when it’s done.

While a chat is still working

A chat that is still working carries a status, both on its entry in the chat list and as a card in Flow, so you can tell at a glance what it’s doing:

Status	What it means
Still running	Pia is working on it. Nothing is needed from you.
Waiting for your confirmation	A tool approval card is waiting for your answer. The chat cannot continue until you answer it.
Finished — tap to review	It’s done and you haven’t looked at the result yet.
Ended with an error	It stopped without finishing.

You can keep typing while a chat is working, but you can’t send until it finishes. The Send button is covered by a busy indicator for the whole turn — hovering it says “Working on the last message…”. Anything you type sits in the message box and sends as soon as the reply lands. If you don’t want to wait, either stop the reply with Cancel or open a + New Chat and let this one finish in the background. If the message box still holds text you haven’t sent, New chat asks first before it throws that text away.

A chat that is Waiting for your confirmation stays that way until you answer — you can navigate away and come back. Flow will show it as a card with a link straight back to the chat, so you don’t have to hunt for it in Chat history.

Give It a Goal Instead of a Message

At the bottom of the message box is a Chat | Agent lever. Flip it to Agent and what you type becomes a goal rather than a message: Pia writes a plan, works through it step by step, and shows every move in a panel above the chat. You approve the plan, can pause and re-order steps mid-run, and can detach it to keep running in the background.

See Agent Runs. To put one on a timer, see Routines.

Switching between Chat and Agent mid-conversation

You can flip the Chat | Agent lever at any point in a conversation — you don’t have to start a new chat to use Agent mode. The lever sets what your next send does, so an ordinary chat can turn into a goal-driven run, and once that run has settled you can flip back to Chat and keep talking in the same conversation about what it produced.

The lever belongs to each conversation. Flipping it for one chat never changes what the next chat starts on. What a new chat opens in is a setting of its own: Start new chats in Agent mode, under Settings → Assistant → Agent runs → New chats, off by default.

Asking a question with the lever on Agent

Agent mode recognises a question. Ask one with the lever on Agent and Pia answers it straight away rather than drawing up a plan for it, and marks the reply Answered directly — “You sent this in Agent mode, but it needed no plan — Pia answered it in one turn.” Real multi-step work still plans as before, so you can leave the lever on Agent without every passing question becoming a run.

The one thing that disables the lever is the persona: if the persona you have selected has Tool Access: None, Agent mode is greyed out with “Agent mode needs a persona with tools.” Pick a persona with tools and the lever becomes available again, mid-conversation and all.

A conversation that has an agent run in it is still a normal chat — the earlier messages stay, and the run’s plan and result are written into the same thread.

Hand Work to Your Server

If your Pia server offers them, a rocket button sits in the composer’s action row — Hand this to your Pia server as a background assignment. It takes whatever you’ve typed as the prompt, asks which of your own records to send with it, and runs the work on the server instead of in this chat: it survives the app closing and comes back as its own chat. This is the one path where content leaves end-to-end encryption, so it always asks first. See Background Assignments.

If Pia is running as administrator

Started with administrator rights, Pia shows a notice at the top of the chat: “Pia is running as administrator. File tools, agent runs and MCP servers inherit those rights, so restart Pia without elevation unless you need it.”

Everything Pia does for you runs with the rights Pia itself was started with, so a file tool or an MCP server could then reach the whole system, not just your own files. Close Pia and start it normally unless you need those rights. The notice’s close button (“Hide this notice for this session”) hides it for this session only — it comes back the next time Pia starts with administrator rights.

Settings

You’re in control of what gets kept. Under Settings → Assistant → General → Chat history:

Retention (days) — “Delete chats not opened for N days”. Chats you haven’t opened for that long are removed. 180 days by default, and you can keep them for up to two years. A chat you marked as a favorite is never removed by this.
Auto-generate chat titles — off by default. It makes a small extra completion call per chat to produce a short title, so it costs a few tokens.
Delete all chat history now — clears everything immediately.

Cleanup always runs. There is no longer a Save chat history checkbox: unchecking it never stopped Pia storing chats, only the clearing out, so chats piled up on exactly the machines that had asked for the opposite.

Which chats are old enough to clear is settled with Pia Cloud rather than from this device alone, so a chat you read on another computer is kept. A device that cannot reach the server clears nothing at all rather than guessing.

Follow-up Suggestions

Settings → Assistant → General has Show follow-up suggestions — “After each assistant reply, generate 2–4 follow-up question chips.” It’s off by default, because producing the chips costs a small extra request on every turn.

That and Auto-generate chat titles above are the two settings in Pia that spend tokens on something other than your actual question. Both are off out of the box.

The suggestion chips on a brand-new, empty chat are a different thing — a fixed list built into the app that costs nothing and is always shown.

Rate an Answer

An answer from the Pia Cloud provider has two buttons beneath it, Good answer and Bad answer. Answers from a provider you added with your own key don’t have them.

Good answer sends the rating straight away, without any text.
Bad answer opens Report this answer. Under What was wrong? describe the problem, leave Include the answer text ticked if the answer should go along, and tick This concerns data protection if the problem is about personal data. Click Send.

The dialog tells you whether personal data in the text is replaced by placeholders before sending, which follows your privacy settings.

Pia confirms with Thank you. With Pia Server 1.4 or later, a report that was stored on the server but reached nobody gets Report saved instead: “Your report was saved by the server operator, but nobody was notified.”

Concerns about Pia’s AI behaviour that go beyond one answer can be sent to ki@pia-ai.de. The address is also shown in Settings → About under AI feedback and complaints.

Attach Images and Use Your Voice

You can attach images to a message. This needs the Pia Cloud provider. See the File Attachments guide.
You can dictate with your microphone and have replies read aloud. See the Speech guide.

Next Steps

Keep your chats safe and available everywhere by setting up Cloud Sync.