Files

Upload files in chat, browse them on the workspace's Apps and Files tab, share them with people and agents, and roll back with version history.

Updated 22 October 2026

Files live in buckets that belong to a workspace. Agents create buckets as they produce work, such as documents and presentations.

Files that belong to a chat are different, whether an agent made them while working there or you attached them to a message: they live in that chat's own file area and stay visible only to the chat's participants. They do not show up in the workspace's buckets or search for anyone else, including admins. See Files: who can reach what you upload for who can reach what, and for the audited recovery path admins have when access is genuinely needed.

Uploading

Attach a file to any chat message and it lands in that chat's own file area, private to the chat's participants. Agents in the chat can read what you attach, so "summarize this" with a file attached just works. If you want an upload somewhere else, the small badge on the attachment lets you pick any workspace bucket you can reach instead; files sent there are visible to whoever can see that bucket.

The files you attach stay as tiles in your own message. They do not open the artifacts panel on their own and do not appear in it as pages until you open one: that panel is for what the chat has made, and when it is open for a document or an app the files you shared are listed under its search button, in a Files you shared section at the foot of the list, so they are still a click away. The panel shows one artifact at a time, newest first, with arrows to step through the rest; with a single artifact there are no arrows or search, just the card and its close. To make one of them the first thing everyone in the chat sees when they open the panel, press the pin on its bar; pinning another one moves the pin. When an agent writes a document for you on a window wide enough to show the panel beside the conversation, the panel opens on its own with that document at full height. On a phone, or a narrow window, open it from the document's card in the chat.

The message you send with a file is one message: the text you typed and the files ride together. You can edit that text afterwards the same way you edit any message of your own, either from the message's own Edit action or by pressing the up arrow in an empty composer, which picks up the last message you sent in that chat or thread. Editing changes only the words. The files stay attached, and the message is marked as edited.

Pasting markdown

Markdown pasted into the message box keeps its formatting. Lists, quotes, code blocks, bold and links turn into formatted text in the box, the same as when you type them. Tables, headings, horizontal rules, images and footnotes can't be shown formatted while you write, so the box keeps that paste exactly as you copied it, in a monospaced block with a line down its left edge. You can edit it as plain text. When you send, it goes out as written and appears in the chat as a real table, heading or image. Editing a message that holds one of these brings it back as plain text the same way, so nothing is lost on the way through. Very long pastes still become a PASTED card instead, except straight after you type three backticks (```, or ```python and so on): a paste there, however long, becomes a code block holding exactly what you pasted.

The same goes for copying out of a chat and pasting back in. Selecting part of a message and copying it keeps its quotes, bold, lists, tables and images, and the copy button above a table keeps the table. A table or image copied from a spreadsheet, a document or a web page arrives as markdown too, though formatting inside a table's cells is dropped, and an image only comes through when it has a full web address. Pasting a selection from a chat into another app, such as an email or a document, also keeps its formatting.

Viewing images and videos

Click an image in a chat to open it full screen. From there you can step through the chat's other images and videos without closing it: use the chevrons either side of the picture, or the left and right arrow keys. Left goes back to older media, right goes forward to newer, and a counter at the top shows where you are. Videos play full screen as you reach them. On a phone, swipe sideways to step and pull down to close. Opening an image inside a thread steps through that thread's media only.

Reading PDFs, Word documents, text files and web pages

A PDF shared in a chat, by an agent or by you, opens in place: click its name and it opens in the chat's side panel on a computer, so you can read it beside the conversation without downloading it. Word documents (.docx), text files (.txt, .md, .csv, .json and code) and web pages (.html) open the same way, and Preview on a web page opens it there too. The name at the top of the panel takes it full screen. Closing the panel puts a file you shared away again. In a PDF, scroll through the pages, select and copy text, and search it with your browser's find. The download button in the header saves a copy. A PDF that lives on another website, rather than in Files, still opens as a download.

Voice messages

The composer has a microphone where the send button sits when you have not typed anything. Record, stop, and it sends as a voice message with a waveform you can play back. It lands with the chat's other files, so a voice note is findable afterwards the same way an upload is.

Bolter transcribes the note while you are recording it, and you can watch the words appear as you speak. Because the transcription happens live, it picks the language up front rather than detecting it: there is a language pill next to the stop button, defaulted from your device. Changing it mid-recording starts the take again in the new language, so set it before you begin.

The transcript travels with the message, which is what makes a voice note useful in a chat with an agent: the agent reads what you said rather than being handed audio it cannot open. If the live transcription fails, the note still sends and still plays, just without a transcript, and the recorder tells you at the time.

Where files live

The Apps and Files tab of the workspace dashboard is the browser: buckets, folders, version history, and sharing all live there, alongside the workspace's web apps and recent background activity. This is one merged tab in every workspace, personal and team alike, not just a personal one. It appears once there is something to browse: a workspace with no files you can see and no apps has no Apps and Files tab, and the tab shows up on its own as soon as the first file or app lands.

The tab opens on Recent, a single newest-first list of apps and files across the whole workspace. Files people attached in chat are left out of it by default, so the documents, presentations and apps your agents produce stay visible. To bring them back, use the settings cog beside the search box and turn on Uploads. A chat file's row names the chat it belongs to, and "Open in files" on any chat attachment or agent file jumps straight to it in Folders.

Copying a file's text

Open a document or a text file and its header has a copy button next to Download. It puts the file's own text on your clipboard, so a Markdown document arrives as Markdown, headings and tables intact, ready to paste into another tool. Selecting the page with the mouse instead would copy what you see rendered and lose the formatting, which is why the button is there.

You do not have to open the file first. In a chat, every document and text file in the artifacts panel carries the same copy button in its row menu, behind the three dots. It is also in the preview header of the file browser.

The card in the chat itself has one too. A text or Markdown file shared in a message, and a long paste that became a PASTED card, show a copy button beside Show or Download. It copies the whole file, not just the lines the card shows.

Downloading a document or a presentation

Open a document and the Export menu in its header offers Word (.docx), PDF, and the raw Markdown (.md).

Open a presentation and its Export menu offers a self contained web page (.html) you can present offline, PowerPoint (.pptx), and PDF. The PowerPoint and PDF are built on our servers rather than in your browser, so everyone who downloads the same version of a deck gets an identical file that matches what you see on screen. That takes a few seconds the first time: the button reads "Export in progress" while it runs, anyone else in the chat sees the same thing, and the download starts on its own when it is ready. Ask for the same version again later and it arrives immediately.

PowerPoint and PDF export need a paid plan or a free trial of one. The plan that counts is the one of the workspace whose agent made the deck. On the Free plan the web page export still works.

In both of those files each slide is a picture of the slide, so the text in them cannot be selected or re-edited. Edit the deck in Bolter and export again.

Exporting only some slides

You do not have to send the whole deck. Hover a slide in the filmstrip along the bottom and a tick box appears in its corner. Tick the slides you want, hold Shift while ticking to take a whole run at once, and the Export button changes to read how many are selected. All three formats then contain just those slides, in the order they appear in the deck.

On a phone the tick boxes are always visible, in the corner of each slide card. Tapping a slide also offers Select for export among its actions, and the export sheet names the count.

Clear the selection from inside the Export menu, or by unticking, and the next export is the whole deck again. A partial export is stored separately from the full one, so having both does not make either go stale, and it downloads under a name that says how many slides are in it. Selecting every slide is the same as selecting none.

Interactive HTML pages

When you open an HTML page from a file in the browser, it opens on its own address like yourworkspace-bucketname.bolter.page, and scripts run there. Pages get their own separate address so they can run JavaScript, store data locally, and work like any other web page you visit. A page cannot reach your Bolter session cookie or write to your workspace (bucket addresses are read-only), and each bucket gets its own separate address, so a page from one bucket cannot read another bucket's files.

These addresses are stable: the link you share stays the same even if you rename the file, as long as the workspace and bucket names do not change. Old links keep working after a workspace rename, redirecting to the new address.

HTML files shared in chat work the same way. The Preview button on an HTML attachment renders the page with its scripts running, so an interactive page behaves in the preview as it would in a tab, and the open in new tab button beside it takes you to the page's own address. Copy URL in the attachment's menu copies that stable address too. It opens for anyone who already has access to the file, and shows a sign-in prompt to anyone who does not, so pasting it never hands out your own access.

If you open a page while signed in to an account that cannot see it, the page says which account you are using. If you have another account that can, choose Use a different account: Bolter asks you to confirm signing out, and once you sign in with that account it takes you back to the page.

Sharing

Access is controlled per file and per folder. You can share with a specific person, a specific agent, a chat's members, workspace admins, everyone in the workspace, by private link, or publicly, with read, write, or admin permission. Select a file, click the badge in its preview header, and use Add access, or approve the card an agent posts when it asks for access. For exactly who can reach a file, see Files: who can reach what you upload.

Files in a chat have a Copy URL action, in the attachment's three-dots menu and on a file card in the artifacts panel. The file browser's preview header has the same thing as Copy link. It copies the file's own stable address, which opens only for people who already have access to the file. To give someone new access, share the file instead.

A private link is the safest way to share outside your workspace: only people who have that exact link can open the file, the link is read only, and you can revoke it at any time from the file's access panel. You can copy the link again whenever you need it from the same panel.

Once someone opens a private link, their browser bar shows the file's plain address, not the link. Bolter swaps the link for a private, unreadable cookie on the first page load, so the page itself never sees it and a screenshot or a copied address cannot pass the file on by accident. If you want to share it with someone else, copy the link again from the access panel rather than copying what is in the address bar. Public sharing makes the file readable by anyone on the internet. Search engines are told not to index public files, but link previews in other apps may still copy what you share, and revoking later does not remove copies others already made.

Agents and files

Agents can create and edit files, and their edits are versioned like anyone else's. Every file keeps a version history, and you can list versions, compare them, and restore an earlier one.

Files an app keeps

An app can keep files in a bucket of your workspace, for example an archive of what it did or a backup of its data. The app's agent writes them, so they show up in Files like any other bucket, and the usual Files permissions decide who can open them. Ask the agent which bucket an app uses if you want to look.