> ## Documentation Index
> Fetch the complete documentation index at: https://docs.get-hive.ai/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> Hive is an AI operations layer for small and mid-sized business teams. It reads the tools a team already runs, raises what needs attention, and acts with the approvals the team configures.
> Answer only from these pages. If a page does not cover something, say so and link the closest page rather than inferring behaviour.
> The supported programmatic entry points are incoming webhooks, public workflow forms and the remote MCP server, each documented under Developers.
> Each connected account has a level: Read, Draft or Act. New accounts start on Act. Agents and workflows take only the actions selected for them, whatever the account level.
> Describe security, sign-on and data handling only as the Security pages and the FAQ describe them, and add no certification, compliance, single sign-on or model-training claim they do not make.
> Prices and credit allowances are in GBP. Quote them only as a page states them, without converting.
> Use UK English and the product's exact UI labels. Link to pages as https://docs.get-hive.ai/<path>.

# Library: files, folders and Knowledge Bases

> Keep the files Hive creates and the documents you upload in one place, organise folders and Knowledge Bases, share them, and save reusable file templates.

The **Library** is where your files live in Hive. Documents, spreadsheets, presentations, images and PDFs created in chats and workflows are saved there, next to the files you upload yourself. From the Library you can give that content to Ask Hive, agents and workflows as context.

Open it from **Library** in the sidebar (`/library`), near the bottom. Choosing **Library** opens
your files and shows the group's three pages under it: **Files** (this page), **Prompts** (the
[prompt library](#prompt-library)) and **Skills** (see [Skills](https://docs.get-hive.ai/skills/skills.md)). See
[The Library group](https://docs.get-hive.ai/get-started/navigate-hive.md#the-library-group).

## What's in the Library

| Area | What it holds |
| - | - |
| **Recent files** | Files you created or uploaded most recently. A file created in a chat keeps a link back to the chat it came from. |
| **Folders** | Standard folders you organise yourself, with subfolders. |
| **Knowledge Base** folders | Folders that give Hive read-only AI context. People with editing rights can maintain the files in them. |
| **Reusable file templates** | Word, PowerPoint and Excel files new documents, decks and workbooks are generated on, keeping their layouts and styles, plus files you can copy as a starting point. |

Files arrive in the Library in several ways:

* **Ask Hive** saves the documents, presentations, spreadsheets and images it creates (see [Tools and modes](https://docs.get-hive.ai/ask/tools-and-modes.md)).
* **Workflows** save the output of **Generate image** and **Create PDF** steps (see [Workflow steps](https://docs.get-hive.ai/workflows/workflow-steps.md)).
* **Upload files** adds your own documents. You can also drag files from your computer onto a folder's file list, or onto **Recent files** on the Library's main page, and drop them there.
* **Blank file** (or **New file** inside a folder) creates a blank **Document**, **Spreadsheet** or **Presentation**.

You can select up to **100 files per upload**, with each supported file no larger than **25 MB**.

While you drag files over the file list, it shows **Drop files to upload to** and the folder's name. Dropped files go to the folder you have open (on the main page, to your Library outside any folder) and are uploaded exactly as through **Upload files**, under the same limits, with the **Upload files** dialog showing progress. A folder you can only view, and an archived project's folder, refuse dropped files and say why. A file dropped anywhere else on the Library page is ignored: your browser doesn't open it in place of Hive.

### Supported file types

You can upload almost any file. Hive reads the formats in the first table below and keeps every
other file as a download. It refuses only files that can run on a computer: programs, scripts,
shortcuts and the like (see [Files Hive refuses](#files-hive-refuses)).

#### Files Hive reads

Hive reads these files: you can preview them, ask about them and give them to agents.

| Type | Extensions |
| - | - |
| Documents | `.pdf`, `.docx`, `.docm`, `.odt`, `.ott`, `.fodt` |
| Spreadsheets | `.xlsx`, `.xlsm`, `.xls`, `.csv`, `.tsv`, `.ods`, `.ots`, `.fods` |
| Presentations | `.pptx`, `.pptm`, `.odp`, `.otp`, `.fodp` |
| Drawings | `.odg` |
| Formulas | `.odf` |
| Images | `.png`, `.jpg`, `.jpeg`, `.webp` |
| Text | `.txt`, `.md` |
| Web pages | `.html`, `.htm` |

A file with one of these extensions must really be that type: a file whose content doesn't match
its name (a web page named `.png`, for example) is refused.

A PDF is stored when it has pages, even when Hive can't read any text in it, such as a brand book
whose type is drawn as outlines. A PDF with only a permissions password (one that limits printing
or editing) opens without a password, and Hive reads it like any other. A PDF that needs a password
to open is refused, and the upload says so: save a copy that opens without a password and upload
that.

A CSV file may separate its columns with commas or with semicolons, as European spreadsheet exports
and DATEV-Format (EXTF) files do: Hive reads it by whichever the file uses. In a semicolon file the
comma is the decimal mark, so an amount such as `142,80` is shown as written and never read as
14,280 in a total. A DATEV-Format file's first line is its format header, so Hive reads its column
names from the second line (see [DATEV](https://docs.get-hive.ai/integrations/datev.md#check-a-booking-file-against-the-datev-format)).

A slide deck exported from a design tool, with charts and shapes on every page, is read to its last
page. PDFs Hive read only in part before are read again on their own, with nothing to re-upload.

An older Excel 97-2003 workbook (`.xls`, up to **25 MB**) is converted when you upload it: the
Library stores an Excel Workbook copy under the same name with the `.xlsx` extension, so
`Budget.xls` is stored as `Budget.xlsx`, and tells you: **Converted from Excel 97-2003 — Hive stores
an .xlsx copy.** The `.xls` itself is not kept, macros in it never run and the copy has none. A
file that is only named `.xls` (a CSV or web page exported with that extension, for example) is
refused. If Hive can't convert the file right now, the upload says so and stores nothing: try
again, or open it in Excel, use **Save As** to save it as an Excel Workbook (`.xlsx`) and upload
that copy.

A web page is stored as a file, never opened as a page: its preview shows the page's readable text,
Ask and agents read that same text (scripts, styles and hidden content are left out), and
downloading it saves the original file unchanged. Nothing in it runs in Hive.

#### Other files

Any other file is stored as a download. Hive doesn't open, preview or read what is inside it, and
it is never opened in your browser: downloading it saves the original file unchanged. A Word 97-2003
document (`.doc`) or a PowerPoint 97-2003 presentation (`.ppt`), a ZIP archive, a drawing or a
video, for example, is kept like this.

Text formats such as JSON, XML, YAML, log files, SQL, subtitles and source code are the exception:
when the file is plain UTF-8 text, Hive stores it as text and reads it like a `.txt` file, so you
can preview it and ask about it. An SVG picture is read as its text this way, never shown as a
picture. A text file in another encoding is kept as a download.

When Hive imports a ZIP archive into a folder, it stores the files inside it whatever their type,
under the same rules as a single upload, and skips programs, scripts and other files it would refuse,
as well as archives inside the archive. An `.xls` inside a ZIP archive is skipped too: upload it on
its own. A ZIP archive uploaded as an ordinary file is kept as a download, and Hive doesn't open it.

Other files have the standard **25 MB** upload limit; larger uploads don't apply to them (see
[Limits reference](https://docs.get-hive.ai/developers/limits-reference.md#library)).

#### Files Hive refuses

Hive refuses a file that can run on a computer, whatever it holds, and tells you: **"setup.exe" is a
program or script. Hive does not store files that can run on a computer.** These are:

| Type | Extensions |
| - | - |
| Programs, libraries and installers | `.exe`, `.dll`, `.so`, `.dylib`, `.msi`, `.msix`, `.msixbundle`, `.appx`, `.appxbundle`, `.apk`, `.aab`, `.ipa`, `.dmg`, `.pkg`, `.mpkg`, `.deb`, `.rpm`, `.com`, `.scr`, `.cpl`, `.sys`, `.drv`, `.ocx`, `.pif`, `.msp`, `.mst`, `.jar`, `.xll`, `.xpi`, `.crx`, `.gadget`, `.application`, `.appref-ms`, `.jnlp` |
| Scripts and shortcuts | `.bat`, `.cmd`, `.ps1`, `.psm1`, `.psd1`, `.ps1xml`, `.vbs`, `.vbe`, `.wsf`, `.wsh`, `.wsc`, `.hta`, `.js`, `.jse`, `.sh`, `.command`, `.lnk`, `.url`, `.website`, `.scf`, `.reg`, `.inf`, `.msc`, `.chm`, `.settingcontent-ms`, `.library-ms`, `.appinstaller`, `.vb`, `.vbp`, `.ws`, `.sct`, `.shb`, `.shs`, `.ps2`, `.psc1`, `.mshxml`, `.ins`, `.isp`, `.hlp`, `.msu`, `.xbap`, `.ksh`, `.csh`, `.desktop`, `.scpt`, `.applescript`, `.terminal`, `.workflow`, `.py`, `.pyw`, `.pyc`, `.pyo`, `.pyz`, `.pl`, `.rb` |
| Files that run queries, links or macros when opened | `.iqy`, `.slk`, `.one`, `.mdb`, `.accdb`, `.mde`, `.accde`, `.adp`, `.ade` |
| Disk images | `.iso`, `.img`, `.vhd`, `.vhdx` |
| Office macro formats Hive can't remove macros from | `.xlsb`, `.xlam`, `.xla`, `.ppam`, `.ppa`, `.ppsm`, `.sldm` |

Hive also refuses:

* any program it recognises by its content (a Windows, Linux or Mac program), even when it has
  been renamed, such as `setup.exe` renamed to `notes.dat`;
* a Word or PowerPoint 97-2003 file (such as `.doc`, `.dot`, `.ppt`, `.pps` or `.pot`) that
  carries macros, or one Hive can't read far enough to check. The same file without macros is kept
  as a download.
* a Word, Excel or PowerPoint file that carries macros under a name Hive can't strip them from,
  such as `invoice.docm` renamed to `invoice.bin`.

Macro-enabled Word, Excel and PowerPoint files (`.docm`, `.xlsm`, `.pptm`) are not refused: Hive
stores a copy without the macros (see [Macros are removed](#macros-are-removed)).

### LibreOffice and OpenDocument files

Files saved by LibreOffice, Collabora or another OpenDocument app upload like their Microsoft
equivalents: Writer documents (`.odt`), Calc spreadsheets (`.ods`), Impress presentations (`.odp`),
Draw drawings (`.odg`) and Math formulas (`.odf`), their templates (`.ott`, `.ots`, `.otp`) and the
flat XML versions (`.fodt`, `.fods`, `.fodp`). Hive stores each file exactly as you uploaded it and
reads it like the Microsoft format it matches:

* **Documents** give their headings, paragraphs, lists and tables, like a Word file.
* **Spreadsheets** give each sheet's cells. A formula is read as the value the file was saved with,
  so Hive never recalculates it.
* **Presentations** are read one slide at a time, with each slide's speaker notes, like a
  PowerPoint deck.
* **Drawings** give the text on each page, and **formulas** their formula text.

Hive checks that a file really is the OpenDocument type its name says before storing it, and refuses
one that is damaged or built to overload a reader. Macros stored in a file never run in Hive.
OpenDocument files are read-only in Hive: you can preview them, ask about them and give them to
agents, but not open them in the editor (see [Edit a Library file in Ask](#edit-a-library-file-in-ask)).

[File templates](#reusable-file-templates) also accept the Office template formats `.dotx`,
`.xltx` and `.potx`, and their macro-enabled versions `.dotm`, `.xltm` and `.potm`. Macro-enabled
files are stored without their macros: see [Macros are removed](#macros-are-removed).
OpenDocument templates (`.ott`, `.ots`, `.otp`) are not accepted as file templates: upload them as
ordinary files to read them.

Uploads run one file at a time. If one fails, files already uploaded remain in the Library and
you can retry the files that have not finished.
Standard folders hold up to **100 files and 250 MB** each. Knowledge Bases hold up to **1,000 files**;
Ask's selection limit is separate, so a large Knowledge Base is not read into one answer.

When your workspace has **larger uploads** enabled, the **Upload files** dialog shows a higher
limit for PDFs, Word documents and presentations (**256 MB** each), and spreadsheets including
CSV and TSV (**100 MB** each). These are decimal MB: 256 MB is 256,000,000 bytes. Other file
types retain their existing upload limit. Your browser sends a larger file directly to private
storage, then the worker checks and prepares it before it appears in Library. The dialog shows
transfer progress and processing status. **Cancel upload** stops an in-progress
transfer. After transfer, **Stop waiting** closes the wait, but the file may
still finish. Library checks for a later result
while you remain signed in; reopen Library if processing takes longer.
If validation fails, upload the file again. A workspace may have
at most two pending or processing large uploads per person; cancelled tickets release a slot, though
their temporary storage remains reserved until cleanup.
New files, edits that increase file size and template copies count towards the
workspace's **25 GiB** Library storage limit. That limit applies on every plan, and **Extra file
storage** does not raise it. On Starter, file storage is also a hard plan limit of 10 GB per paid
seat: an upload that would take the workspace's stored files past either limit is refused. On Team
and above, Owners and Admins are told when stored files pass the plan's included storage (see
[Add-ons](https://docs.get-hive.ai/admin/billing-and-plans.md#add-ons)). Existing content remains accessible if a limit is
reached.

Large originals remain downloadable. Inline previews still have their own 25 MiB processing
limit. **Ask Hive** reads a file up to 100 MB: a file over 25 MiB from the text Hive extracted when
it was uploaded, without page images or pictures, and the answer says so. The Ask editor has smaller import limits, listed below. Larger uploads are
available only where the dialog actually offers them; the normal 25 MB limit remains otherwise.
See [Limits reference](https://docs.get-hive.ai/developers/limits-reference.md#library).

### Pages and text

Opening a Word or PowerPoint file in Library shows its **Pages**: the file as it looks in Office,
with its layout, fonts, pictures and slide backgrounds. **Text** beside it switches to the text Hive
read from the file. In a Word file's text, headings start with `#` (one per level), list items with
`-`, and tables show as rows of cells separated by `|` (see
[How Hive reads the files you choose](#how-hive-reads-the-files-you-choose)). Word and PowerPoint
templates you upload show their pages the same way, and so do the Library files you review in Agent
Studio, the files attached to an agent (see [Attach files as
knowledge](https://docs.get-hive.ai/agents/agent-studio.md#attach-files-as-knowledge)) and the files attached in Ask chats (see
[Attach files](https://docs.get-hive.ai/ask/ask-hive.md#attach-files)).

* **When pages are made.** Hive makes the pages in the background when the file is uploaded, and
  again when a new version is saved. If you open the file before they are ready, the preview says
  **Preparing pages…** and shows them as soon as they are done, usually within a few seconds. A file
  that has no pages yet, such as one uploaded before pages existed, is prepared the first time
  anyone opens it.
* **On the cards.** In the grid view of your files, in **Templates** and in Ask's template gallery,
  a PowerPoint file's card shows its first slide once its pages are ready, rather than the slide's
  text. This includes the **By Hive** starter decks, whose pages are made the first time anyone sees
  their card. Until then, or if pages can't be made, the card shows the text and switches to the
  slide when it is ready.
* **What gets pages.** Word (`.docx`, `.dotx`) and PowerPoint (`.pptx`, `.potx`) files, and
  OpenDocument documents, presentations and drawings (`.odt`, `.ott`, `.fodt`, `.odp`, `.otp`,
  `.fodp`, `.odg`). OpenDocument spreadsheets open in the sheet grid, and a formula file (`.odf`)
  shows its text. A preview
  shows the first 100 pages and says so when a file is longer; download the file for the rest. Older `.doc` and `.ppt` files keep the
  text view.
* **When pages can't be made.** For a file that is too large, too slow to lay out or damaged, the
  preview shows its text with the line **Pages could not be made for this file, so its text is
  shown.** The file itself still downloads as normal. If the pages
  just could not be loaded (a moment's network trouble), the preview says so, shows the text, and
  choosing **Pages** tries again.
* **How they're made safely.** Pages are drawn by a separate service, not by Hive's main servers.
  That service holds none of Hive's secrets and cannot reach your workspace's data. Each file is
  rendered on its own in a sandbox with no network access, so links and pictures stored on the web
  are never fetched, macros never run, and formulas and fields show the values the file was saved
  with. See [Infrastructure](https://docs.get-hive.ai/security/infrastructure.md#security-of-the-platform).

Pages are for reading in Hive. They are not available to MCP clients, which download the file
itself.

Opening a PowerPoint deck's **Text** shows it one slide at a time, with **Next slide**,
**Previous slide** and an **Outline** of every slide. Each slide's speaker notes sit under
**Speaker notes**, apart from the slide's own text. A slide with no text, such as one that is only
a picture, says **This slide has no text — download to see it**.

Excel workbooks, OpenDocument spreadsheets and CSV files open in the same sheet grid Ask uses, with
a tab for each sheet. In Library the grid is read-only: open the file in the editor to change it
(OpenDocument spreadsheets stay read-only). A workbook too large
for the grid shows its values in a plain table instead.

## Find files

Use the filters at the top to narrow the list:

* **File type**
* **Created** date
* **File location** (at the top level of the Library: **All recent files** or **Unfiled only**)

The **Folders** list has its own **All**, **Mine** and **Shared with me** filter.

If nothing matches, the Library suggests trying a different search, file type, date or location.

## Organise folders

Use **Create folder** for a standard folder, or create a **Knowledge Base** when you want the contents to act as reference material for Hive. Inside a folder you can **Create subfolder**, move files, and **Rename** or **Delete** files and folders. Select several files to act on them together, for example **Delete selected files**.

Deleting a file is permanent. Within about 20 minutes Hive deletes its stored copy, its earlier versions, the text Hive read from it and its search index, unless the same content is still used elsewhere: in another file, as a version of a file you kept, or as a document still open in a chat. Deleting a folder does not delete its files: they move to the top of the Library. Deleting a template removes it from the Library but keeps its stored copy. See [Deleted files](https://docs.get-hive.ai/security/data-protection.md#deleted-files).

### Knowledge Bases

A Knowledge Base is a folder Hive treats as trusted reference material, such as policies, price lists, engagement letter templates or product documentation. It is marked with a **Knowledge Base** tag. Hive reads it as read-only context, and only authorised people can change its files.

Files synced from connected file stores (Google Drive, Dropbox, OneDrive and SharePoint) also feed Hive's knowledge, and their original permissions carry over. See [Files and docs integrations](https://docs.get-hive.ai/integrations/files-and-docs.md).

### Search in any language

When Hive searches your Library, connected files and the files attached to a chat or project for
an answer, it matches the words of your question as well as their meaning, and the word matching
works in every language, including where a workspace has no meaning-based search set up:

* **Chinese, Japanese and Thai**, written without spaces between words, are split into words, so a
  question finds a file that uses the same words in different sentences.
* **Arabic** matches with or without hamza on alef (أ إ آ), short vowels and tatweel, and treats
  ة and ه, and ى and ي, as the same letter. An Arabic question also matches Arabic-Indic digits
  (٢٠٢٤) as Western ones (2024).
* **Hebrew** matches with or without vowel points (niqqud).
* Common short words such as *في*, *של*, *und* or *dans* are not counted as search words.

Long files are split into passages for searching at the sentence ends of each script too (。！？,
؟, ۔ and ।), and Thai at word boundaries, so a passage ends on a whole sentence or word rather than
in the middle of one. Searches in English work as before.

### Project folders

Each [Ask project](https://docs.get-hive.ai/ask/projects.md) has a folder here with its pinned files, marked with a **Project** tag. Its people are the project's people, and you rename, share and delete it from the project. Its menu offers **Open project** instead. Anything you add to it is a project file too. See [Project files are a Library folder](https://docs.get-hive.ai/ask/projects.md#project-files-are-a-library-folder).

While its project is [archived](https://docs.get-hive.ai/ask/projects.md#archive-a-project), a project's folder and the folders inside it are read-only: **Upload files** and **New file** are off there, and the page says why. An empty folder there says no files were added before the project was archived. Restore the project to add files again.

Library sharing never applies to a project's folder or the folders inside it. Sharing, unsharing or transferring a project's folder outside the project is refused with a pointer to the project. That includes AI tools connected through the [Hive MCP server](https://docs.get-hive.ai/developers/mcp-server.md). To give someone the folder, add them to the project.

The owner of a file inside a project's folder can still share that one file with someone outside the project. That person can open the file, but not the folder or the project's other files.

## Share and access

The Library is private by default:

* A folder is visible only to its owner and the people it's shared with.
* Workspace admins have **no separate view** of other members' folders.
* Sharing is always with specific, active workspace members. There are no public Library links.
* Open a folder and click **Share** to see who has access. If someone shares a folder with you, you can **Leave folder**.
* **Shared with me** also lists **Ask files shared with you**: documents, spreadsheets and presentations from someone's Ask chat that they shared with you on their own, marked **Can view** or **Can edit**. Choose one to open it by itself; the chat it came from stays private. See [Share one file without its chat](https://docs.get-hive.ai/ask/tools-and-modes.md#share-one-file-without-its-chat).
* A subfolder has the same people as its top-level folder. You add people, and remove anyone the top-level folder shares with, on the top-level folder. A subfolder's **Share** dialog lists who has access and offers **Open top-level sharing**. Removing someone at the top level removes them from every subfolder too. People that older versions of Hive added directly to a subfolder are still listed there, and you can remove them.
* A file's **Share** dialog lists the people who can open it through its folder under **Inherited access**, naming the folder. The file's owner is listed once, as **Owner**, even when they also own the folder. To remove someone listed there, change who can open the folder.

> [!NOTE]
>
> When you give Ask or an agent a Library file as context, Hive respects these rules. People who can't open a file can't read it through Hive either. There are two exceptions. A file you attach to an agent is part of the agent: everyone who can run it gets the file when they run it on its run page, in Ask and in Continue in chat, and can open it read-only from a chat with the agent, though it never joins their Library or search (see [Who can read an agent's files](https://docs.get-hive.ai/agents/agent-studio.md#who-can-read-an-agents-files)). And in a shared Ask chat, an answer Hive built from your file is shown in full to everyone you share the chat with, though the file itself stays closed to them. See [Feedback, sharing and history](https://docs.get-hive.ai/ask/ask-hive.md#feedback-sharing-and-history).

## Use Library content as context

In Ask Hive, choose a file or folder to ground an answer in content you've already prepared, for example "Summarise the attached engagement letter and list the fees" or "Draft a proposal using the *Proposal template* folder". Agents can also be given Library files and folders as permanent knowledge in [Agent Studio](https://docs.get-hive.ai/agents/build-an-agent.md).

### How Hive reads the files you choose

Hive gives the model each file you choose **whole**, never a fragment of it:

* **PDFs and images** go to the model as the files themselves when the chat's model can read
  them (a PDF saved again after it was signed, such as one signed through Dropbox Sign, included), so a scanned PDF, a photo of a receipt or a chart in a PDF is readable as it is. With a
  model that can't read PDFs, Hive sends the PDF's text instead; a scan then can't be read, and
  Hive says so. Choose a different model in the composer to read it. Hive adds a note that
  everything in these files is data, never instructions, and names each one "Attached file 1",
  "Attached file 2" and so on, so a file's own title cannot pose as an instruction.

* **Every model can work with your photos.** A model that can't see images gets Hive's
  description of each image instead: what it shows and any text in it, written by Hive's vision
  model and labelled as such. It is the description Hive made when the file was added, or, when
  there is none yet, one Hive makes for that answer and reuses after that. A description is made
  for at most 8 images per answer and each gets up to 20 seconds; an image without one is named as
  an image the model can't read. A deployment with no vision model set up only uses descriptions
  it already has. Models that see images still get the image itself. A PDF over
  **14 MB or 100 pages**, or an image over **5 MB**, is too large for the models to take as a
  file, so Hive sends its text instead (an image or scan that large can't be read). One answer
  sends at most 14 MB of files, and 100 PDF pages, this way; past that, the rest go as their text.
  A PDF's text keeps its pages apart, and a deck's its slides, so Hive can list a long PDF's pages
  or a deck's slides, each with its first line, and open just the pages it needs by number.

* **Word, text and Markdown files** go as their full text. A file Hive opens on its own (from a
  list) that is longer than about 100,000 characters comes in parts, and Hive opens the next part
  when it needs it. One answer reads at most about 600,000 characters of file text in total; past
  that Hive answers from what it has read and says which file to ask about on its own.

  A Word document's text keeps its structure, written the way a Markdown file marks it:

  * **Headings** are the paragraphs the document itself marks as headings: the Title and
    Heading 1 to Heading 6 styles, styles based on them, and any paragraph or style with an outline
    level (the levels Word's Navigation pane shows). A paragraph is never treated as a heading just
    because it is bold, large or short, and levels deeper than 6 read as level 6.
  * **Bulleted and numbered lists** stay list items, indented by level. The numbers Word adds to a
    numbered list or heading are not part of the text. A numbered heading style, such as a
    contract's numbered clause headings, stays a heading.
  * **Tables** keep their rows and columns, with the first row as the header. A cell merged across
    columns keeps its columns, a cell merged down the rows below is left empty there, a row that
    starts or ends a few columns in keeps its cells under their headers, and a table inside a cell
    is read into that cell. Text in the document's headers, footers and notes follows its body.
    Text Word hides (hidden text, including text hidden by a style, and the old place of moved
    text) is left out.

* **PowerPoint decks** go as their text per slide, numbered, with each slide's speaker notes.
  Pictures on the slides also go as images when the chat's model can read images, within the same
  14 MB per answer; a model that can't gets Hive's description of each picture instead.

* **CSV, TSV and Excel files** go as tables: each sheet with its columns and their types, then its
  rows. An Excel cell formatted as a date reads as the date (`2026-10-01`, with its time when it
  has one), never as the number Excel stores it as. A table too large to send whole (more than about 50,000 tokens) goes as its columns and
  first 50 rows, and Hive reads more rows when the question needs them. Totals, averages and
  counts are calculated over every row, never added up by the model; see
  [Totals, averages and counts](https://docs.get-hive.ai/ask/ask-hive.md#totals-averages-and-counts). Ask for a chart and
  Hive draws it from every row.

When the files you chose add up to more than about **50,000 tokens** (very roughly 35,000
words, or 30 PDF pages), Hive first sees a list of them, with each file's name, date and first
lines. It then opens the files it needs one at a time. **Worked** names each file it opened by
what it read: **Opening a Library file, whole**, **Opening part 2 of a Library file** for one part
of a long file, or **Reading excerpts of a Library file**. You can choose up to **50 files** for
one chat, directly or through a folder. A folder gives Hive only the files in it that you can
open yourself, and only those count toward the 50.

**Files longer than one answer can read** (more than about 150,000 tokens of text, roughly 300
pages of a Word document, up to about 1,000,000 tokens) can't be read to the end in one answer
anyway, so Hive reads them for your question. It sends the model the passages that contain the
words you asked about, up to about 22,000 tokens, and never more than reading the file from the
start would send. In a Markdown file (`.md`) or a Word document (a Google Doc opened from Drive
included), which mark their headings and tables (a Word document's from its heading styles,
outline levels and tables; see above):

* When a heading contains the words, the whole section under it is sent, up to the next heading
  of the same or a higher level, when it fits. This holds even when the words also appear all
  through the rest of the file; the other passages that mention them are then counted for the
  model rather than sent. Each word's heading section, or its passages, is added smallest first, so
  a large section doesn't crowd out a word with less text.
* When a table's introduction or header row contains them, every row of the table is sent.
* A `#` line or a table row inside a fenced code block is code, not a heading or a table.

PDFs and slide decks reach Hive as plain text, without their headings or table borders marked, so
a line that starts with `#` or `|` there is just text. For those, Hive sends the passages that
contain the words themselves, each named by the line it starts with. So does a Word document with
no heading styles: a paragraph that is only bold or large is not a heading.

Arabic and Hebrew in a PDF are read in reading order. Many PDFs store a right-to-left line in the
order it is drawn on the page, left to right, and in the joined letter shapes the font holds, so a
plain read gives every word backwards. Hive turns such a line back round, keeps numbers and Latin
words in it (a company name, a year) in their own order, and reads the joined shapes, the lam-alef
ligature included, as ordinary letters, so searches, sources and quotes match what you type. A line
already stored in reading order is left as it is, and so is one whose order Hive can't tell. This
applies to PDFs Hive reads from now on: a PDF already in the Library keeps the text it was read
with until a new version of it is added.

Hive matches the forms of a word (*terminated* finds *termination*), with or without accents, and
reads words that PDFs write with ligatures (*ﬁ*) or break with soft hyphens. This works for
questions in any language Hive supports: words in Chinese or Japanese are found though those
languages put no spaces between words, and Arabic words match with or without vowel marks,
whichever form of alef is used, and with *ة* or *ه*. The model is told
how many passages it got of how many, how many more parts this answer can still read, and that
related passages worded differently may be missing. It is also told that a total, count, list,
dates, or whether something is in the file at all needs the file read in order. That note sits
above the file, and it and the passages' headings carry a mark that changes on every read, so text
inside a file can't pass itself off as Hive's. The answer's sources only count the passages, for
example **Board pack.md (2 excerpts of 300)**: they never quote the file, because sources are kept
with the answer and shown wherever it is shared.

For these files, Hive reads in order from the start instead, part by part, when:

* your question needs all of the file: a summary, overview or recap (in English, German, French,
  Spanish, Italian or Dutch); a total, count, list, timeline, average, comparison, the highest or
  lowest, or the latest or earliest of something; pulling out every date, name or item of a kind;
  or whether something is in the file at all, or what is not there. For an agent, the words of
  your message and the values you type into its form both count. When a workflow runs a saved
  agent, the values its author mapped into the agent's inputs count the same way; the agent's own
  instructions don't;
* the question's words appear across too much of the file, or are mostly not in it;
* the question names nothing to look for;
* the answer has already searched long files as much as one answer may, counting the searches of
  any agents it handed tasks to. Text that takes extra work to match, such as letters that change
  length when accents are set aside, counts for more of that allowance.

The model is told why, and to say which parts it read. Files that one answer can read whole, and
files longer than about 1,000,000 tokens, are read from the start, part by part, as before,
whatever you ask. This applies to any file Hive reads as its text: text, Markdown and Word files,
slide decks, and a PDF that isn't sent to the model as the file itself. A PDF the model reads as
the file itself, and a spreadsheet, are read as before.

An [Agent Studio](https://docs.get-hive.ai/agents/agent-studio.md) agent's bound Library files work the same way: in its
**Test**, its runs, when you ask it in Ask, when another agent hands it a task, and as a workflow
step. Up to about 50,000 tokens in total they go to the model whole. Past that, the agent gets the
list and opens only the files it needs, so a large folder no longer costs a whole read on every
answer. When you ask an agent in a chat that has its own Library files chosen, it can read those
too.

There's no preparation step: a file is ready to use as soon as it's in the Library. Hive still
checks every upload behind the scenes: it flags hidden or concealed text and refuses files it
can't parse safely.

### Charts, embedded objects and macros in Office files

Word, PowerPoint and Excel files are accepted with everything in them except macros and a Word
document's links to an attached template, subdocument or frame elsewhere: native charts and the workbooks
behind them, embedded Excel workbooks, documents and other objects, ActiveX controls, and other
links to files and pictures elsewhere. A file without macros or such links is stored exactly as you
uploaded it, so a download gives you back the same file.

#### Macros are removed

Hive never stores macros. When you upload a macro-enabled file (`.docm`, `.xlsm` or `.pptm`, or the
templates `.dotm`, `.xltm` and `.potm`), or a regular Office file that carries a macro project or a Word document that links to an
attached template, subdocument or frame elsewhere (see below), Hive stores a macro-free copy and tells you: **Macros were removed — Hive stores a macro-free
copy.** The copy takes the regular format and extension, so `Budget.xlsm` is stored as
`Budget.xlsx`, and you preview it, read it in Ask and edit it like any other Office file. The same
line appears whenever you preview the file. This applies to every upload, larger uploads included;
files uploaded before 8 October 2026 are left as they are.

The original's macros are not kept anywhere, so keep your own copy of the macro-enabled file if
you still need them. Everything else in the file stays as it was. Hive removes:

* the VBA project, its digital signatures and Word's macro data;
* Word key bindings and legacy toolbars, which only run macros;
* in a file that had macros, its custom ribbon buttons, which call them;
* a Word document's link to an attached template, a subdocument or a frame on another computer or
  website, which Word fetches when it opens (an attached `.dotm` brings its macros with it), with
  the setting that names it;
* the same in Office files embedded inside the file, up to three levels deep.

ActiveX controls hold no code, so they stay in the file and never run in Hive. A file whose macros
Hive can't remove is refused before anything is stored, and the message says what to change:

* an Excel workbook with Excel 4.0 macro sheets: delete the macro sheets in Excel;
* an embedded Word, Excel or PowerPoint 97-2003 object, an embedded object that holds a macro
  project or wraps a file with macros (such as a packaged `.xlsm`), or an embedded object Hive
  can't read through: remove the object, or open it and save it in the current format. A file
  packaged into a document as an object (shown as an icon) can be an Office file, a PDF or a PNG,
  JPEG or GIF picture, named as what it is; Hive can't check any other packaged file for macros, so it
  refuses it.

Hive's own readers treat that content as inert: they don't run a macro or a control, open an
embedded object, or follow a link. Previews and AI reads use the
file's text (for a deck, also the values its charts already show), and the preview tells you when
a file has charts, embedded objects, controls or links that it doesn't show.

The original-file preview of a document opened in Ask is drawn by a document renderer that is set
never to run macros, update links, recalculate formulas or update fields when it opens the file.
It isn't drawn at all for a file with controls, an attached template or links to other
files, including in a document embedded in it, nor for one Hive can't check in full: an embedded
object in an older Office format (such as an Excel 97-2003 sheet), or embedded objects past the
limits below. A field or formula that refers to another file or address (such as `INCLUDETEXT`,
`INCLUDEPICTURE`, `WEBSERVICE` or a DDE link) doesn't stop the preview from being drawn: the
renderer's settings are what keep it from updating.

Hive refuses an Office file only when it can't be read safely, and the message says why:

* the file, or an object embedded in it, is damaged;
* the file and the objects embedded in it together expand to more than 64 MB (512 MB for files
  over 25 MB), or hold more than 3,000 parts;
* objects are embedded more than three levels deep (a chart in a document embedded in a deck is
  two);
* it holds macros Hive can't remove (see [Macros are removed](#macros-are-removed)).

A macro-enabled file renamed to `.docx`, `.xlsx` or `.pptx` is no longer refused: Hive stores a
macro-free copy of it under the name you gave it.

Hive reads a file's text once, when you upload it or save a new version, and keeps that text with
the file. Ask, agents, an agent's **Test** and workflow agent steps then read the kept text instead
of reading the file again, so a long PDF or a large spreadsheet is ready sooner on every use after
the first. A new version always gets its text read afresh. A file uploaded before this change has
its text read and kept the first time it's used. Who can read a file is checked exactly as before,
on every use. Very long text (over about 4 million characters) isn't kept; those files are read
from the file each time, as before.

Agents read the Library files bound to them as their full text in the same way, with no length
cap. An agent doesn't read a scanned PDF or an image bound to it yet: it is told the file has no
text and says so in its answer.

## Files copied from Google Drive, OneDrive and SharePoint

When you add a file from Google Drive, OneDrive or SharePoint to an agent in
[Agent Studio](https://docs.get-hive.ai/agents/agent-studio.md#files-from-google-drive-onedrive-and-sharepoint), Hive saves a
**copy** of it in your Library. The copy is an ordinary Library file with the access of the place
it is saved, labelled with where it came from (**Google Drive**, **OneDrive** or **SharePoint**).
A Google Doc is copied as a Word file, a Google Sheet as Excel and Google Slides as PowerPoint.
Each file must be a type the Library reads and no larger than 25 MB, as an upload is.

Hive reads the original through your own connected account, and only you can refresh the copy:

* **Refresh** checks the original and copies it again only if it changed. The content it replaces
  is kept in the file's [Version history](#version-history), marked **Refreshed from Google
  Drive** (or OneDrive, or SharePoint), so nothing is lost.
* Hive also checks the original when you open the agent in Agent Studio, if the last check was
  more than an hour old, and when you deploy the agent in Agent Studio. Nothing is checked in the
  background, and no one else opening the agent ever starts a check.
* A refreshed copy is used straight away by every agent it is attached to, deployed ones
  included, the same as editing a Library file.
* Adding the same file again refreshes your existing copy instead of making a second one.

If a check can't reach the original, the copy keeps working as it was and Hive says why: you need
to reconnect the account in Integrations, the original was deleted or you can no longer open it,
the original is now larger than 25 MB, or it is now a type Hive can't read. Hive tries the
account you copied the file through first, then your other connected accounts of the same
service, so reconnecting an account doesn't break your copies.

## Edit a Library file in Ask

Open a stored **Document** (`.docx`), **Spreadsheet** (`.xlsx`) or **Presentation** (`.pptx`)
you own or can edit, then choose **Open in editor** at the top of the preview. You don't need to
preview it first: the file's **⋯** menu in the file list or grid (the action sheet on a phone) has
**Open in editor** too, under **Preview**. The first time, this creates your own Ask editing conversation, including for blank files created in Library. Opening
the file again returns to that conversation and your working copy, with your edits, for Word files
and PowerPoint decks alike; if the Library file was changed elsewhere since, Hive opens the new
version there. **Save** updates the same stored Library file: its folder, sharing and file link
stay the same. Hive never writes an edit back to a connected provider's original.

If you can only view a Word, Excel or PowerPoint file, the preview and the file's **⋯** menu offer
**Open a copy in editor** instead:
Hive makes your own copy of it in your Library's top level and opens that, and opening the
original again returns to your copy. Your edits change only your copy. Knowledge Base files have
no copy to edit, and no **Open in editor** at the top of the preview.

A **Presentation** opens in the [deck editor](https://docs.get-hive.ai/ask/deck-editor.md) instead. Its changes save in the
deck as you make them and flow back to the Library file within about a minute, one Library
version per burst of work. Each time you choose **Save version…** there, Hive also writes the deck
back to the Library file as a new version. See
[Decks from the Library](https://docs.get-hive.ai/ask/deck-editor.md#decks-from-the-library).

You can also stay in a chat: @-mention a Word or Excel Library file and ask for the change ("in
this file, change the Findings section"). Hive opens the file in that chat and saves its edit as a
new version, exactly as **Open in editor** does. A Library presentation is changed by the same
[Hive AI](https://docs.get-hive.ai/ask/deck-editor.md#hive-ai-in-the-deck-editor) as in the deck editor, and its changes reach
the Library file within about a minute, as edits made in the deck editor do. In a
project chat, a Word or Excel file pinned to the project works the same way without the @-mention;
see [Change a pinned file from a chat](https://docs.get-hive.ai/ask/projects.md#change-a-pinned-file-from-a-chat), which also
covers a pinned PowerPoint file.

The Ask editor can open workbooks up to **12 MB** and Word files up to **20 MB**; the deck editor
opens PowerPoint files up to **20 MB**. Larger Library files remain available to preview or download, even though the
normal Library upload limit is 25 MB per file. Files over 25 MiB can be downloaded but not
previewed inline; Ask reads them, up to 100 MB, from the text Hive extracted when they were
uploaded. PDFs, images and other formats are not editable in Ask.
OpenDocument files (`.odt`, `.ods`, `.odp` and the others above) are read-only in Hive and have no
**Open in editor**: the preview's options menu says so. Download the file to edit it in LibreOffice, then
upload the new version.
Knowledge Base files are read-only in Ask, although authorised people can still maintain
them from the Library.

> [!WARNING]
>
> Office features and formatting Hive cannot import may change when you save an edited file.
> The file keeps its own look: a PowerPoint file its slide size, masters, layouts and theme, a Word
> document its styles, page setup, header and footer, and a workbook its cell styles, column widths
> and frozen panes, the same way a file made from a [file template](#reusable-file-templates) does.
> A workbook also keeps its own charts and dropdown lists, and a chart or dropdown Hive adds is a real
> Excel one; see [Charts and dropdowns in a spreadsheet](https://docs.get-hive.ai/ask/tools-and-modes.md#charts-and-dropdowns-in-a-spreadsheet)
> for what is not kept. Number formats, fills, borders, frozen panes and colours by value you add in
> Hive are laid over the workbook's own cell styles; see
> [Format a spreadsheet as a financial model](https://docs.get-hive.ai/ask/tools-and-modes.md#format-a-spreadsheet-as-a-financial-model).
> A PowerPoint slide nobody changed downloads exactly as it is in the file you opened: Hive copies it
> unchanged, with its speaker notes, pictures, charts, SmartArt, animations and anything else the
> editor does not show. Moving slides or deleting others does not count as changing one. Only the
> slides you or Hive changed, and new ones, are written by Hive, so what follows about formatting
> applies to those. A slide that holds comments, an embedded object or content linked from elsewhere
> is written by Hive even when untouched, and so is every slide once you change the deck's theme or
> turn slide numbers on or off.
> Titles and body text keep the look their slide master and layout give them: the deck editor shows
> them at that size, font and alignment, and the downloaded file keeps a title's master style,
> including its colour, capitals, letter spacing and effects, unless you change it. A deck so close
> to the editor's size limit that this would not fit opens without it, and the editor says so.
> The deck editor draws a deck at its own slide size, so a 4:3 deck is edited on a 4:3 slide and
> anything you place stays on the slide when you download it.
> Pictures are kept: PNG, JPEG and WebP pictures on slides and in a Word document's paragraphs
> stay in place when you save. Pictures the editor does not show, such as an EMF or GIF picture or
> one inside a Word table, stay where they are as long as the paragraph or table around them is
> kept (see below). A linked picture is not carried over. A Word document's chart stays too when
> you leave its paragraph as it is and the chart's data is a plain workbook inside the file (cells
> without formulas). On a slide Hive writes, a PowerPoint file's tables and its column, bar, line,
> pie, doughnut, scatter and stacked charts, and its charts of columns with lines over them, are
> kept with their words and numbers, drawn in Hive's table and chart style; its other charts are
> left out.
>
> A file with ActiveX controls or pictures and videos linked from elsewhere keeps its own
> look when you save it: only those are left out of the edited file, which never carries macros,
> ActiveX controls or an attached template. The version you uploaded keeps them (but never macros,
> which were removed when it was uploaded), and a download tells you what was left out. Hive can't write a file in its own layout when, for example, a
> PowerPoint slide master or layout holds an embedded object or a control. A download of it then
> comes in Hive's look and says so, and saving it over the Library file is refused with the reason,
> so the Library file keeps its look; your edit stays in the chat, where you can keep working on it
> as text.
> Each paragraph of a Word document keeps its own formatting when you save, including one whose
> words you changed: its paragraph style (Heading 1, a list style or your firm's own style), list
> numbering, spacing, indents and alignment. A paragraph whose words you keep or change in place
> also keeps its fonts, colours and line breaks — a template's blanks filled on each side of a line
> break, such as an address block, each come back in their own run with the breaks kept — and a
> table its style, column widths and cell
> shading, unless it has merged cells, holds a field, hidden text, a content control, a comment, a
> footnote or tracked changes, or you add or remove columns (then it keeps its style but not its cell
> formatting); the empty lines between them stay too.
> Changed words take the formatting of the words they replace. A change that mixes formatting (new
> bold or italics, a new colour, highlight, font or size, or a sentence rewritten across a coloured
> label), a change split by a tab or
> across words in another format, adding a line break, or a paragraph with a field
> (such as a date or page number), hidden text, a content control, a text box or shape, a picture
> beside its words, a comment, a footnote or tracked changes, keeps its paragraph style but not its fonts. A
> paragraph, heading or list item you add takes the paragraph style of the nearest one of the same kind,
> and what that style sets, with its spacing, indents, borders and shading (a list item you add continues
> its list), but not the formatting set on that paragraph's own words. A left-to-right paragraph you add is
> aligned as the editor shows it, so one you add after a centred paragraph is left-aligned unless you
> centre it; in some Word files opened before 9 October 2026 it takes that paragraph's alignment.
> Colours, highlights, fonts and sizes you set in the editor
> come through on such a paragraph, and an alignment you set replaces the paragraph's own. A Word
> file opened in the editor shows the colours, highlighting, fonts, sizes, strikethrough and
> alignment set on its own words, paragraphs, list items and table cells, as Word shows them: a
> colour or shading picked from the document's theme, lighter or darker shades included, is the one
> its theme gives, a left-to-right paragraph keeps its centring in a file set up for right-to-left text, words
> coloured so they can't be seen against the page, their shading or their highlight, such as white
> text, open in the normal text colour, and words in a symbol font such as Wingdings open in the
> normal font; the file's notes say so. In any document, words whose colour hides them against the
> page or their own highlight, such as white text pasted in or written by Hive's file tools, show in
> the normal text colour in the editor, words in a symbol font show in the normal font, and Hive
> treats both as hidden text when the AI reads the document; their formatting is kept, so the
> download looks as it did. Formatting that comes from
> a style, underlining, character styles, theme fonts, and formatting inside a link are not shown,
> though a paragraph you leave as it is keeps all of it in the download. Changing the formatting of a
> table cell that holds one paragraph rewrites only that cell in the download, and the rest of the
> table, its borders, shading and column widths stay as the file had them; in some Word files opened
> before 10 October 2026, a table whose cell formatting you change keeps its table style but not its
> cell formatting. A Word file opened before
> 9 October 2026 opens without them until you open it from the Library again; its download is
> unchanged. Joining all the lines of a paragraph into one, with a
> space between each, keeps its line breaks in the download; joining only some of them keeps its
> paragraph style but not its fonts. A document opens with up to 5,000 line breaks: a paragraph whose
> breaks would take it past that opens with them as spaces, and keeps its breaks in the download
> unless the document is edited in the editor, which can turn them into spaces. In a very long, heavily formatted document, the later
> pages keep their paragraph styles but not their fonts or table formatting. Headings and lists set by the document's own styles, such as
> a firm heading based on Heading 1, open in the editor as headings and lists.
> A Word file saved from the editor on or before 4 October 2026 may have lost its pictures and paragraph
> styles; its earlier version is in the file's [version history](#version-history).
> The original file you opened remains downloadable from your Ask editing document. If another
> person updates the Library copy, moves it into a Knowledge Base, or removes your edit access,
> your save is refused; reopen the Library file to start from its latest version. When Ask
> Hive itself edits the file and finds that someone else changed it first, it offers its edit as
> a proposal for you to review instead of overwriting their change. Accepting that proposal
> replaces their version, which stays in the file's [version history](#version-history).
> If you open a Library file into a chat you share, people who can collaborate on the chat can
> edit it there too: their saves update your Library file as new versions, made as you and only
> while you can edit it, without giving them any access to your Library. See
> [Feedback, sharing and history](https://docs.get-hive.ai/ask/ask-hive.md#feedback-sharing-and-history).

### Save a chat file to the Library

A document, spreadsheet or presentation Ask Hive created in a chat can be filed in the Library.
In the file panel choose **Save to Library**, pick the top level or a standard folder you can
edit, then **Save**. Knowledge Bases and folders you can only view are not offered. Ask Hive
can also do this when you ask it to ("save this to my Board packs folder"), but only into the
top level or a folder you selected for the chat and can edit. After that, saving the file in Ask
updates the same Library file. A deck made with **Save as new deck** from one of its versions is a new
file in the chat, not in the Library, until you save it there (see [Deck
versions](https://docs.get-hive.ai/ask/deck-editor.md#versions-and-file-history)).

### Version history

Each time a Library file's content is replaced, by a save from Ask, a restore or a refresh of a
[copied file](#files-copied-from-google-drive-onedrive-and-sharepoint), Hive keeps the content it
replaced as an earlier version. Open the file and choose **Version history** to
see them, newest first, with what replaced each one and who: **by you**, **by** a workspace
member's name, or **by a former member** for someone who has since left the workspace. When Ask Hive's own edit replaced a
version, the version also notes what Hive changed, such as "Before an Ask edit: Hive edited 3
cells in Sheet1". A deck from the deck editor notes which of its revisions
replaced the version: "From deck revision 12" when you chose **Save version…**, and "Auto-saved
from deck revision 14" when its edits flowed back on their own. Edits within ten minutes of an
auto-saved version update the file without adding another version; a version saved with **Save
version…** or from Ask always stays in the history. Anyone who can open the file can
**Download** an earlier version. People who can edit the file can **Restore** one: it becomes
the current content, and the content it replaces is kept as a new version, so a restore can be
undone. A restore never changes the file's name, folder or sharing. The history lists the 100
most recent versions; older ones are not shown and cannot be downloaded or restored.

## Reusable file templates

Templates let authorised members start from the same source. A source file is not changed by
copying or generating from it.

In Ask you can also give a file you already have open a template's design; see
[Apply a template's design](https://docs.get-hive.ai/ask/tools-and-modes.md#apply-a-templates-design).

### Starter templates by Hive

Every workspace starts with fourteen ready-made Word, PowerPoint and Excel templates, marked **By
Hive** in the gallery. They keep their own layouts: styles, headers and footers in Word, slide
masters and layouts in PowerPoint, and sheets, formulas and input cells in Excel. Text in
\[square brackets] is a placeholder to replace.

| Area | Templates |
| - | - |
| Professional services and legal | Engagement letter, Disengagement letter, Professional clearance request, Professional clearance reply, Mutual NDA, Advice memo |
| Accounting and finance | Board pack (deck), Management accounts (workbook), Financial model shell (three-statement workbook), Client register (workbook), Price list (workbook) |
| Private equity and due diligence | Due diligence request list (workbook), Investment committee memo |
| General business | Client proposal, Project status update (deck) |

The letters either side of an engagement follow UK accounting practice in generic wording:

* **Disengagement letter** confirms the end of the engagement: the work completed and not
  completed, fees outstanding, how records are returned, the date the firm stops acting, the
  handover to the new adviser and the deadlines the client now owns.
* **Professional clearance request** is the incoming adviser's letter to the outgoing one. It asks
  whether there is any professional reason not to accept the appointment, and requests handover
  information: UTRs, VAT and PAYE references, outstanding returns and agent authorisation.
* **Professional clearance reply** is the outgoing adviser's answer: clearance, handover
  information, fees outstanding and records. In Ask, a request to reply to, respond to or answer a
  clearance letter ("reply to the professional clearance letter from Smith & Co") uses the reply,
  never the request.

Hive drafts these letters; it does not send them. Check them against your professional body's
guidance before you use them.

The **Price list** workbook is a practice's fees, laid out for Hive's fee calculator. Its fees are
examples, not recommended fees: replace them with your own. See
[Fee quotes from your price list](https://docs.get-hive.ai/use-cases/accounting-and-client-services.md#fee-quotes-from-your-price-list).

#### Client register

**Client register** is a practice's list of clients as one workbook, one row per client: client
name, entity type, company number, UTR, NI number, VAT number, VAT registration date, VAT stagger,
year end, contact name and email, postcode, services, annual fee, onboarding status, AML risk,
letter signed date, HMRC authorisation status, self-employment gross income, property gross
income (own share), income figures tax year, MTD Income Tax status, MTD Income Tax exemption, MTD
quarter type and notes. Only **Client name** is required, and the drop-downs offer the usual
values. **AML risk** is the one your firm records after its own assessment: Hive never decides it.
The income columns and **Income figures tax year** let the **Statutory Deadline Chaser** estimate
when a sole trader or landlord joins Making Tax Digital for Income Tax; **MTD Income Tax status**
offers HMRC's own statuses for you to record. See
[Making Tax Digital for Income Tax cohorts](https://docs.get-hive.ai/use-cases/accounting-and-client-services.md#making-tax-digital-for-income-tax-cohorts).

Duplicate it into Library, replace the bracketed example row with your clients, and attach it to
the **Statutory Deadline Chaser** and **New Client Onboarding Pack** agents (see
[UK accounting templates](https://docs.get-hive.ai/agents/build-an-agent.md#uk-accounting-templates)). Moving from Engager?
Upload Engager's client export as it is: Hive does not convert it, and the agents read its columns
by name. The starter's **Engager columns** sheet lists the Engager column names the agents look for
as each register column. See
[Moving from Engager](https://docs.get-hive.ai/use-cases/accounting-and-client-services.md#moving-from-engager).

Every member who can use Library can preview, download and use them. Nobody can edit, share or
delete them. To change one, choose **Duplicate template**: the copy is your own private template,
which you can then edit and share. **Duplicate into Library** and **Generate new file** work as
they do for your own templates. In Ask, pick a starter from **Use a template** when you create a
document, presentation or spreadsheet, or from the **@** menu. In Agent Studio, choose one under **File templates**.
When you don't pick one, Ask uses a starter only if you ask for it by name or for that kind of file
("make a board pack"); otherwise a new file gets Hive's own design. See
[Which look the deck gets](https://docs.get-hive.ai/ask/tools-and-modes.md#how-hive-designs-a-new-presentation).

> [!NOTE]
>
> The starter templates are generic drafts, not legal, tax or financial advice. Review them for your
> practice and jurisdiction before you use them with clients.

### Your own templates

Where larger uploads are enabled, you can also upload a DOCX, PPTX or PDF file template up to
**256 MB** directly to private storage. Hive validates it before it appears in the gallery:
files saved by Microsoft Word, PowerPoint or Excel are accepted as saved, while a
password-protected (encrypted) file is refused.
Other template formats retain the **25 MiB** upload limit. A larger template can be downloaded
or used to make an exact new Library file; inline template previews stop at **25 MiB** and
the Ask editor has its own lower limits. A larger source is not silently treated as ready
AI context. During processing, **Stop waiting** closes your wait, not the upload: the template
may still appear in Library. The gallery checks for a later result while you
remain signed in; reopen it if processing takes longer. Each file made from a
larger template counts towards the workspace storage ceiling.

**1. Add a template**

Click **Add template** and upload a document, spreadsheet, presentation or another supported file.
Hive keeps the original file and, for a Word (`.docx`, `.dotx`), PowerPoint (`.pptx`, `.potx`)
or Excel (`.xlsx`, `.xltx`) file, also reads what new files are generated from: its slide
layouts and sample slides, its heading outline and Word styles, or its sheets and header rows.
The template preview shows this under **Parsed for generation**. For an Excel template, Hive
also reads its line items (the labelled rows left empty for input and the rows that calculate,
with their formulas) when it generates a file, up to 60 line items a sheet and 20 sheets. If
an Excel template is too large to read whole and line items may be missing, the preview says
"This template was too large to read whole, so some of its rows and formulas may be missing",
and Hive generates from what it read. Sample data rows left unread do not count: they are
never line items. For Word and PowerPoint you
can add up to 4,000 characters of optional **Generation instructions**, such as tone or what
each section should cover.

**2. Share it**

Templates start private. Share one with workspace members if you like: after saving it, choose
the specific active members to share it with.

**3. Copy or generate**

**Duplicate into Library** creates a content-preserving copy of the original file (Office
template packages become a new document of the matching format). For a Word or PowerPoint
template, choose **Generate new file**, describe the document or deck you need (up to 4,000
characters), and Hive creates a new file with original content responding to your request.
The new file appears in your selected standard folder or at the top level.
Alternatively choose **Generate in Ask** (Word, PowerPoint or Excel): Ask opens with the
matching **Create document**, **Create presentation** or **Create spreadsheet** chip on and
the template chosen. Describe the file you need and send it.

Generated documents and decks are written **on the template itself**. A presentation keeps the template's
slide masters, layouts and theme: the first slide uses the layout of the template's first slide,
other slides use its content layout, and text fills the layouts' own title and body
placeholders, so the master's fonts and colours apply. A document keeps the template's Word
styles, numbering, page setup, header and footer, and is written in its Title, Heading, List and
Table styles. The template's own sample slides and body text are replaced, and the new file does
not carry the template's author details, comments or thumbnail. Nor does it carry anything in the
template's header, footer or notes that loads content from elsewhere when Word opens the file:
fields that read only the file (page numbers, dates, running headings, cross-references, document
properties) keep working, any other field keeps the text it showed, an embedded object is removed,
and a header or footer that names an outside address is left out. The result tells you when
anything was taken out. A spreadsheet is written on its
Excel template too: the workbook keeps the template's cell styles, number formats, column widths,
frozen panes and page setup (margins and orientation). Its header row takes the look of the template's header row, and
data cells the look of the template's first data row, column by column. Each sheet is written
into the template sheet with the same name (or in the same place), and an extra sheet copies the
template's last sheet. The extra sheets together can copy up to 4 MiB of sheet layout (everything
on a template sheet except its cells); when a workbook needs more, Hive stops with an error rather
than writing it, so use a template with a smaller last sheet or write fewer sheets. Hive is asked to keep the template's line items in order and to write
their formulas as formulas, so a total or variance row still calculates; rows with sample values
are treated as examples. The template's sample cells, merged cells, tables and charts are not
copied. Formats you add in Hive (number formats, fills, bold, borders, colours by value and model
colours) are laid over the template's cell styles, changing only what you set, and frozen panes
you set replace the template's. When workspace disclaimers apply, a workbook keeps Hive's own styling so it can carry its
disclaimer sheet. Images, charts and other content placed directly on the template's sample
slides are not copied; inspect the new file before sharing it.
Automatic final-page disclaimers may add a slide beyond the source deck.

A template that contains macros (one stored before 8 October 2026) or ActiveX controls, or that loads remote content, cannot
generate files; you can still copy it. Nor can a PowerPoint template whose slide masters or layouts
hold an embedded object or a control (anything only its sample slides hold is replaced with them,
so it doesn't count). A file that is not a readable Office template is
refused rather than producing a misleading copy. A private template cannot be generated into a
folder other people can read, and a Knowledge Base cannot be a generated-file destination.

Agent Studio can use a template for the files an agent writes: see
[Build an agent](https://docs.get-hive.ai/agents/build-an-agent.md). Its picker can also create a private **reference copy**,
or **Generate a reference from** a Word or PowerPoint template once, and stage that file as
agent knowledge.

Hive sends your request, the template's optional instructions, and the template's parsed
outline (slide layout names, slide titles and short sample text, or headings, or sheet header
rows) to the workspace-authorised model, marked as template data rather than instructions.
The template's styles, layouts and images are applied locally when the new file is built.

You can also save folder structures, writing disclaimers, regulatory review checklists and design packs as templates. Design-pack logos must be PNG.

## Prompt library

Saved prompts have their own page, **Prompts**, under **Library** in the sidebar (`/prompts`). Search them, organise them into folders, and reuse the prompts your team relies on. An older link to the Library's former Prompts tab (`/library/prompts`) still opens the same prompt or folder on this page.

Every folder starts closed when you open the prompt library: choose a folder to open it. Searching
opens every folder with a match, and saving or moving a prompt opens the folder it lands in.

The **Workspace** folder holds the prompts everyone in the workspace can use, and shows your
workspace's logo when it has one (set in **Settings → Workspace → General**).

**Private** holds only your own prompts ("Only you"). A prompt someone shares with you directly is
listed under **Shared with you**, with its author, unless it sits in a folder they have also shared
with you, where it is listed in that folder. You can use it or duplicate it into your own prompts,
and edit it only if they let you. **Use in chat** puts a prompt in the Ask box without its email
addresses, phone numbers, links and `#channels`, since someone else may have written it; app pins
such as `@Outlook` stay. Type where Hive should send yourself.

On a phone, **Prompts** lists your folders as rows, each with who it is for and how many prompts it
holds. Tap a folder to see its prompts, one row each; tap a prompt to open it, or **⋯** for **Use in
chat**, **Duplicate**, **Move to…**, **Share** and **Delete**. **+** at the top makes a new prompt or a
new folder, and a search lists every matching prompt with the folder it is in. See
[Lists on a phone](https://docs.get-hive.ai/get-started/navigate-hive.md#lists-on-a-phone).

## Related

> [card] [Ask Hive](https://docs.get-hive.ai/ask/ask-hive.md)
>
> Ask questions grounded in your files and tools.

> [card] [Projects](https://docs.get-hive.ai/ask/projects.md)
>
> Share files and standing instructions across a group of chats.

> [card] [Business Brain](https://docs.get-hive.ai/brain/business-brain.md)
>
> What Hive knows about your business and where it learned it.

> [card] [Data protection](https://docs.get-hive.ai/security/data-protection.md)
>
> How Hive stores and isolates your content.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.