The simple guide.

What each tool does, where to find it and what to try first. No technical knowledge needed.

Open Desk

Start with a useful task

Choose an example, edit the details and decide when to send it.

Reply to a customer

Turn a customer’s message and your facts into a reply you can review.

Use this example

Understand a spreadsheet

Connect Google, name the sheet and ask a question about its contents.

Use this example

Create an image

Start with a clear visual brief, then refine the image in Studio.

Use this example

These links stage a draft. They never send a request or change a connected account.

Your first five minutes

Brandpatch brings AI conversations, ready-made tasks, creative work and saved results into one place. You do not need to know how to code to get started.

  1. Sign in. Open Desk and use your email or username and password. New here? Request early access. Accounts are currently approved by a person.
  2. Ask for one useful thing. Start a new chat and describe what you need. You can use the example below.
  3. Make it yours. Read the answer, then follow up with “Make it shorter” or “Use a warmer tone.” Download any saved files you need.

Password or access trouble?

Ask the person who manages your account for a password reset/setup link or help with permissions. A request for early access does not create an account immediately.

Just looking around? This guide and Status are public. Opening a tool does not start a paid action or submit a job. Sending a prompt is a separate action.

Try a simple request

Write a friendly email reminding a customer about their appointment tomorrow at 10am. Keep it under 80 words. Do not invent a name or address.
Open Desk

Where to go

Choose the place that matches what you want to do. Your account permissions determine which actions are available.

Desk
Ask questions, write, code and continue a conversation.
Open Desk
Library
Find your private saved notes, whiteboards and recorded images. Download a saved note as Markdown, PDF or Word.
Open Library
Apps
Fill in a short form for a specific task, such as a customer reply or flyer.
Open Apps
Studio
Make images, refine saved images and edit videos you upload.
Open Studio
Work
Check running tasks, find completed results, manage schedules and workspace files.
Open Work
Notes
Write a blank note, start a report outline or edit a completed chat reply.
Open Notes
Whiteboards
Draw a process, organize ideas and save your board.
Open Whiteboards
Schedules
Plan a task for later or manage a repeating brief.
Open Schedules
Projects & files
Organize related work and browse its saved files.
Open Projects & files
Admin
Manage people, permissions, connections, keys and usage, if your role allows it.
Read about Admin
API & IDE
Connect a script or coding editor. This is optional and more technical.
Open API
Connections
Connect Google services, add your personal Notion token or check Microsoft mail setup; see which capabilities are available.
Open Connections
Status
Check whether the AI engines are available.
Open Status

Chat, write and code

  1. Open Desk and choose New chat.
  2. Choose Chat for everyday questions or Code for work that needs a coding workspace. Pick an available engine in the model menu: ChatGPT, Claude or Grok.
  3. Type your request. Attach relevant images, documents or ZIP archives using the attachment control, then choose Send.
  4. Keep follow-ups in the same thread when they relate to that work. Use search to find an older thread; pin important threads and archive ones you have finished.

While a reply is running: Steer adds an instruction to the current turn. Stop interrupts it. Regenerate asks for another response; it starts another turn.

Fix a message you already sent

For a text message in your own chat, choose Edit + resend. Make your changes in the message box, then press Send when ready. The edited request opens a new copy of the chat from that message, and the original stays available.

Understand the progress label

The live reply can show Preparing context, Requesting a reply, Using tools or Replying when that activity is observed. Not every engine reports every step. Checking status means the app is checking the connection or current job; it does not mean your request failed. Stop ends the current request. Steer adds guidance to an eligible running ChatGPT turn.

Tell us when a reply needs work

Under your latest finished reply, choose Useful or Needs work. Add an optional reason, or choose Undo to remove your rating. Feedback is private to your account.

Refresh a changed saved reply

If a saved reply changes, choose Refresh this reply to load the verified saved answer. Your draft, attachments and open note stay in place. If the current attempt is running, failed or stopped, Check reply status updates its status while keeping the displayed answer.

Come back to an unfinished request

Desk keeps a separate draft for each of your own chats in this browser tab. Words, model choices and original attached files can return after refresh or a visit to another tool. A restored draft is never sent automatically. Saved drafts last up to 24 hours in the same login; signing out clears them. They are not synced to another device. If browser storage is unavailable, keep the page open and follow the notice beside the composer.

Working with code? Give the goal, the relevant files and what a successful result looks like. A coding workspace may allow the agent to change files, so review the changes and outputs before using them.

Pick by task. ChatGPT handles image generation here. Claude and Grok can help with text and code. Engine/model choices depend on what is connected and available; the menu shows the current options.

Give the answer a clear shape

Turn these notes into a short checklist for a new employee. Use plain English. Put anything that is unclear under “Questions to ask.” Notes: [paste your notes here]
Open Desk

Get a second opinion

You have two ways to use more than one AI engine.

Together: draft, review, execute

  1. In Desk’s model menu, choose Together.
  2. Choose the engine and model for each step. Draft writes the first response, Review checks and revises it, and Execute carries out the reviewed answer.
  3. Send your request. The steps run in order in the same thread. Read the final result.

Review an answer you already have

Use Ask … to review under the latest eligible saved answer. Review checks that specific answer. If the review proposes work to carry out, the available Execute action performs it. A review is not a promise that the answer is correct.

A simple greeting usually only needs a normal chat. Together and explicit reviews start additional turns, so use them when another pass would help.

Use a ready-made App

  1. Open Apps and choose the task you need.
  2. Fill in the required fields first. Optional fields help tailor the result. Enter your own business name, colors and facts where the form asks for them.
  3. Check the exact prompt preview and the available engine choice before running it.
  4. Run once, then open the saved outcome or its Desk thread to make changes.

Build a campaign step by step

Choose Campaign brief. Add your offer and audience, choose Email, Social post or WhatsApp, then add the next step and confirmed facts. The channel controls which questions appear. Back keeps your answers. Review campaign opens the exact prompt; Create draft runs the App in its Desk thread. This App uses your brief and attached files; it does not read or change connected Google services.

Repeating the same task? Save a named input set and apply it next time. Favorites and named input sets sync to your Brandpatch account across devices. Applying a set fills the form without running it.

Keep up to 12 input sets per App and 96 per account. Sets save form text only; attached files and unsent drafts stay in this browser tab.

If older saved inputs or favorites are found in this browser, choose Import browser inputs or Import browser favorites to add them to your account. Opening Apps does not upload them automatically.

Run history belongs to your account. Reopening a saved receipt checks the original task rather than starting it again. See When something gets stuck for uncertain submissions.

Attach optional files, including ZIP, before running an App. Connected Google services work automatically for ordinary text Apps. Image Apps use the host image tool.

Apps save draft words and original files in the same browser tab for up to 24 hours when storage is available. Return to an App to restore its draft; nothing runs automatically. If storage is unavailable, keep the page open and follow the notice. An uncertain send can be checked without sending it again.

All available Apps

There are currently 21 task Apps, including a guided campaign brief. Text Apps use the engine selected on the form; image Apps use ChatGPT. Compinche Apps use its own support guide and are written for Compinche tasks.

Compinche is not Amazon. General Apps can be used for your own business; they do not automatically add Brandpatch branding.

Find, review and edit saved work

Find a saved version and download it

  1. Open Library. Choose Notes, Whiteboards or Images, then choose an item. Load more adds another page. Search loaded items searches the titles already loaded, not every file in your account.
  2. For a note, check the version shown. Choose Download Markdown, Download PDF or Download Word to keep that saved version. Open saved note returns to the same version in Notes.
  3. Downloads do not ask an AI to rewrite your work. If PDF or Word cannot represent a note, no file is downloaded; choose Markdown instead. Original attached files are separate.
  4. On a phone, Back to library returns to the list. Workspace files are in Work → Projects & files; they are separate from the saved versions in Library.
Open Library

Start a private note or report outline

Start with a blank note or report outline, then edit and save your own text.

  1. Open Notes, or choose Documents in Desk. Open My notes and find New note type.
  2. Choose Blank note, Executive summary, Research brief or Data analysis, then New note.
  3. Replace bracketed placeholders with your own text and evidence. Save records a new version.
  4. Download a saved version as Markdown, PDF or Word. Use in chat adds that saved text to your next request.

Report starters create private outlines in the current interface language. They do not read sources, call an AI or add findings.

Creating a note keeps your previous unsaved draft in this tab. Return to unsaved draft reopens it while you still have access.

If its saved base is no longer retained, the recovered draft stays read-only. Download unsaved text keeps a plain-text copy.

Edit a reply and use its saved text in chat

  1. In Work, choose Saved work beside the search box to find conversations and your private saved note and whiteboard titles. Choose Notes or Whiteboards to search only those titles. Open a match directly in its conversation or editor.
  2. Choose Review PDF on a saved PDF in Desk. Move between pages, zoom or search the text already in the file.
  3. Choose Edit as document below a completed reply in a Chat conversation you own. On a large screen, edit beside the chat. On a phone, Back to chat keeps your draft and running task. Choose Notes to start a blank document.
  4. Use headings, lists and emphasis, then Save. In Desk on a phone or tablet, open Downloads & details for saved versions, comparison, downloads and Use in chat. Choose a saved version, pick Markdown, PDF or Word, and Download. Unsaved edits are not included. Your latest ten versions are retained.
  5. Choose Use in chat to add the selected saved version to your next request. A chip shows the version being used. You can select up to four documents with 96 KiB of saved text in total.

Saving a document stays private. Use in chat sends the selected saved text to the chosen AI and records a copy in that conversation, visible to its collaborators. It does not share the private Note itself. The selection adds context for that request. Its accepted copy can remain in saved or provider conversation history and be used in later turns.

Keep up to 25 private documents per account. Editing a document does not change the original reply. PDF review shows the current workspace file, up to 32 MiB. Search covers the first 200 pages and shows up to 100 matching pages; scanned pages need OCR, which is not available here.

Build a simple report table

  1. Open a note, place the cursor where the table belongs, then choose Table. Start with three rows and three columns; the first row is the header.
  2. Choose a cell to add or remove a row or column. Type text or numbers. Formula-like text stays as text; these tables do not calculate totals.
  3. On a phone, scroll the table sideways to reach the other columns. Save, choose a format and Download right here. Library also offers downloads of that saved version.

A note can hold three tables, with up to 30 rows and six columns each, and 180 cells in total. An export that cannot represent the text explains the problem and keeps your saved note. Word supports literal tabs; PDF may require you to replace tabs with spaces.

Open Notes

Add a saved bar chart to a report

  1. In an editable note, place the cursor outside a table and choose Chart. Enter a title, category names and literal decimal values. Open Labels and source to describe units, period and the source or coverage you provide.
  2. A note supports two horizontal bar charts, with up to eight categories each. Use a decimal point and up to six decimal places; negative and zero values are supported. Charts do not run formulas, retrieve sources or calculate findings.
  3. Save the note, then download its saved version as PDF, Word or Markdown. On a phone in Desk, open Downloads & details. Invalid chart fields must be corrected before Save; a last-valid preview does not mean those edits are saved.

PDF includes vector bars. Word includes a chart image and an editable table of the exact values; editing that table in Word does not redraw the image. Edit the chart in Notes and download again to update both. Markdown includes the literal values and metadata, without a chart image.

Unsaved chart fields stay in this browser tab when you return to the same note and version. Downloads use the displayed saved version, not those unsaved fields.

Explore a saved report

  1. In Desk, open Downloads & details and choose Explore saved report. In Library, open a saved note with charts and choose Explore saved report.
  2. Select categories or change their order to inspect that saved version. The chart scale stays fixed. Filters change the view only; downloads include the complete saved version.
  3. Use the Back button to return to the same item. In Desk, your unsaved text and chart fields stay in place. This preview does not save changes or run calculations.

Find an earlier attachment

Open Work, choose Projects & files, and select a conversation. Expand Retained attachments to find your uploads in that project or conversation. Choose a file to jump to the message where you attached it.

Run a saved calculator

  1. Open Calculator from Library. Pick a saved calculator or create the private example. Change the quantity or price and choose Run calculator. On a phone, choose View results to jump to the output.
  2. Return to the same saved version in the same tab within 30 minutes to recover the values you entered. Choose Run calculator again to see the results.
  3. If a run fails, Copy error details identifies the version and the problem. Open saved note to inspect the code.
Open Calculator

Draw and save a whiteboard

  1. Choose Whiteboards in navigation. Choose Blank board, Simple process or Plan ideas, then New board.
  2. Add shapes, text, arrows or freehand marks. Give your board a name and choose Save.
  3. Open Saved versions to restore an earlier copy as a new version. Your latest ten versions are kept.
  4. Use Export saved version to download an editable .excalidraw board, a PNG image or an SVG image.

Keep up to 25 private boards per account. On a phone, choose Edit board to draw and Done editing to return to the saved-work controls. Unsaved changes stay in this tab; save before leaving. Whiteboards do not use an AI or connected service. Your board library searches board names.

Open Whiteboards

Make or refine an image

  1. Open Studio. Choose an existing workspace, or use New workspace and select Images.
  2. Describe the subject, style and any text you want. Mention a shape such as square or wide if it matters.
  3. Generate once and watch Studio history. You can return to an accepted job’s saved receipt later.
  4. Open the finished image in the library. Use its available download or edit action. For an edit, select the saved image and describe the change.

Use Library search, image/video filters and sorting to find saved media. Uploads belong to the selected workspace. You can also ask ChatGPT for an image in Desk, or use an image App for a logo, flyer or post. Branding and text should come from your request. Image generation can take longer than a text answer.

Ask for exactly the image you want

Generate a realistic wide image of a bright indoor bouldering gym with colorful climbing holds. No text, logos or branding.
Open Studio

Save finished work to Google Drive: Choose Save to Drive beside a recorded image in Desk, Work, Apps results or Studio history, or beside a current file on Studio’s bench or library. Recorded image saves check the original from that reply; workspace file saves use the current version. Files save to My Drive, up to 32 MB. Choose Change beside the destination to search for a writable folder instead. Files inherit that folder’s sharing, so check the destination you choose. The Save to Drive button saves immediately; there is no extra review. After saving, choose Open in Drive. If a response is lost, Check status checks the same save without uploading another copy. Connect or reconnect Google in Connections if Drive access is missing.

Reuse an image in a follow-up

When Images from this reply appears in Desk, Work, Apps results or Studio history, choose Download or Use in follow-up. The service checks that the file still matches that reply’s image. Use in follow-up adds it to the original Desk chat’s draft and keeps that draft’s existing text and attachments; it does not send a message. Apps restore separately prepared files when you return, if the same login and browser storage are still available. Follow any missing-file notice. If the original is missing or changed, choose a current file separately.

Retry a Studio edit: Edit and retry restores saved settings only when the selected file versions still match. If a source changed or its version was not saved, your draft stays in place. Choose current library files to prepare a new edit. A restored draft is not submitted automatically.

Edit a video you already have

Studio edits uploaded video. It does not currently generate a brand-new video from a text prompt.

  1. In Studio, create or choose a writable workspace that supports Images and video edits.
  2. Upload your video and select it in the library.
  3. Open the video tools. Choose an available edit or describe the change, then review the settings before starting.
  4. Read the result, preview the output and download the saved file. The original and outputs remain in that workspace.

Edits run through ffmpeg, a video-processing tool. The receipt shows what ran; you do not need to write the command yourself.

Describe a practical edit

Trim this uploaded video to its first 10 seconds, remove the audio and save an MP4 copy. Keep the original unchanged.
Open Studio

Find finished work

  1. Choose Work in navigation, then Results.
  2. Use Unread for results you have not acknowledged, or All for the full available history.
  3. Open a result to read its reply and view saved media. Follow its thread or Studio link to continue in the original workspace.
  4. Mark it read when you have reviewed it. Opening it does not automatically mark it read.

Check what a run completed

Open Run results on a completed Desk turn, an App result, a Work run or Studio history. It shows saved execution and reply status, recorded connected actions and any recorded failure category. Refresh results checks saved records without running the task again.

Draft saved means the action was prepared; it has not been sent or applied. Outcome uncertain means delivery could not be confirmed. Check the original service before repeating it. A partial list or a finished run does not prove every requested task was completed.

Failure labels describe recorded causes such as provider limits, lost access or interrupted replies. Older runs may have no category. A label does not make retrying an email or file change safe.

Is it still working? Activity shows waiting, running and finished tasks. A task waiting for capacity has not necessarily failed. Results only shows work you own and can still access.

A note was not sent? If a run finishes while you are writing a note, choose Use note as follow-up. Your exact words open as an editable draft in its original Desk chat. Undo restores the draft loaded in Desk. The note stays in this browser tab for that login, including a Work reload; it does not sync to another device. Press Send when ready.

Keep projects and files together

A thread is a conversation. A workspace is the folder for its files. A project groups related work.

  1. Use the Desk’s project controls to organize related threads.
  2. Open Projects & files and select the workspace.
  3. Browse its folders and open supported text or image previews. Use Upload text for the supported text-file types.
  4. Download saved files from their links in a Desk reply or from the relevant file/library view.

Desk accepts document and ZIP attachments up to 32 MB each; Studio supports image/video uploads. A filename in an answer does not guarantee the file was saved. If the UI says “not saved” or the preview is unavailable, check the original workspace rather than copying a host path into your browser.

Reuse a file in a follow-up

  1. Attach a supported file in Desk and ask about it. Then open Source files on the message you sent to see how much text was read.
  2. Use in follow-ups keeps the same file selected in that chat. Open Files for follow-ups above the composer to see or remove selections.
  3. Download original retrieves the attached copy. Copy file citation gives you the filename, a link to the source message, and the checksum that identifies the exact file.

Read current repository changes

  1. In Desk, open Files from the chat menu. In Work, open Projects & files and choose a workspace.
  2. Open the folder containing its own .git directory, then choose Repository. Parent repositories are not searched.
  3. Choose a changed file and Staged, Unstaged or Untracked. Download selected patch checks that the current repository still matches what you read.

This view is read-only and shows the current shared folder. It does not identify a past task’s changes or create an isolated worktree. It does not edit or stage files, run a task or call a model.

The inventory covers at most 200 changed Git-visible paths or refuses the complete read. Ignored files are not listed; secret paths are excluded and submodules are not inspected. A selected UTF-8 patch is complete up to 256 KiB or unavailable; binary and unsupported paths cannot be downloaded as patches. Nothing is silently truncated.

Back to files keeps your selected folder, preview and unsent draft while the same login remains valid. The browser Repository view needs current workspace access, read permission and a supported Linux sandbox. Coding workspaces also need coding permission. If inspection is unavailable, continue using Files.

Ask about a ZIP archive

Attach a ZIP and ask: “List the files in this archive and summarize the README and project notes.” The preview lists files and reads bounded UTF-8 text or code. Binary files and nested archives are listed without inspecting their contents. The preview does not unpack files into the workspace or execute them.

A ZIP must be 32 MB or smaller, with at most 2,000 entries. Files inside may total up to 64 MB when expanded, with no individual file above 8 MB. Password-protected, unsafe or excessively compressed archives are rejected; use a standard unencrypted ZIP with fewer or smaller files.

The preview lists up to 100 entries and reads up to 12 text files, prioritizing README files, manifests and documentation. Each text excerpt is limited to 4,096 bytes; the complete preview is limited to 20,000 bytes. Attach an important file separately if you need more detail. Requested Drive uploads keep the exact original ZIP bytes.

Schedule a task for later

  1. Open Work → Schedules.
  2. Write the prompt and select the thread where the task should run.
  3. Choose Once for your local date and time, or Repeats for daily, weekday or weekly tasks. Review the selected timezone and choose Preview next runs to check the timing. Custom cron timing is under Advanced.
  4. Review the next runs, timezone and prompt before choosing Schedule.
  5. Return to edit, pause or resume it. Check the resulting turn and any file changes after it runs.

Ordinary schedules run prompts in their selected thread. Google briefs read Gmail, your primary Calendar and one selected Tasks list into a private chat. They can also create one explicitly configured follow-up task per run. They do not send mail or change Calendar events. Pausing blocks further connected steps; tasks already created remain in Google.

Save a Google morning brief

  1. In Work → Schedules, expand Save a morning brief under Google briefs and choose a private chat. Select Gmail, Primary Calendar and/or Tasks. For Tasks, choose one task list; the brief reads its incomplete tasks. Set up to 20 items per source.
  2. Choose a daily or weekday local time and IANA timezone, then choose Save brief once. This authorizes the saved plan for 90 days using the original Google connection and the chat’s saved AI model.

Turn each brief into a follow-up task

  1. Select Gmail or Calendar and Tasks, then load and choose the exact Tasks list.
  2. Enable Create one follow-up task after each brief. Enter its title and choose a due-date offset from 0 to 7 days. The task title adds the scheduled local date; a due date is not a timed reminder.
  3. Save brief once. Each successful run with every selected source read creates at most one task in that list. Its notes contain the bounded brief and source coverage; partial reads and shortened notes are labeled.
  4. Open Recent runs to see created, failed, skipped or uncertain outcomes. Open Google Tasks to review the original. An uncertain action is never automatically repeated.

If a read or the brief fails, the task is skipped. Existing read-only briefs remain read-only. Changing the connection or saved plan requires a new brief; a resumed brief never silently switches accounts.

Keep up to five briefs. Open Recent runs → Run details to see the saved sources, limits, calendar window and recorded coverage. Emails listed and message details read are separate counts. Open run lets you review the execution; a completed run does not prove a saved reply or full coverage. Missed runs over 15 minutes are skipped; interrupted runs do not replay automatically.

If setup has an unknown response, keep this tab and choose Check setup. Reload checks saved state without creating another brief. Retry only the same saved setup when it is not found.

Pause, resume or remove a saved brief here. Expiry, lost private-chat access or a changed Google connection pauses future runs. A new connection needs a new brief; resume never silently switches accounts.

Start with a low-risk scheduled task

Draft a three-bullet summary of the notes already in this thread. Include open questions. Save the answer here; do not send or publish it.
Open Schedules

See why a scheduled turn is waiting

Waiting for capacity means the turn could not enter the queue yet. Retrying means Work is trying the same scheduled occurrence again. Leave the schedule enabled if you want it to continue.

If you see Review timing, correct the date, time or timezone, then save. If Work says access changed, ask the person who manages your account to restore it before resuming. A Google brief that misses its 15-minute start window is skipped.

A schedule paused because of its settings? Open Edit in Work. A saved model keeps its engine even when the chat switches engines. Choose Use current thread settings to follow the chat’s current model and reasoning settings, then review the next run and save. Invalid saved settings pause once with a visible reason; temporary capacity limits leave the task due for a later attempt.

Share and personalize

Work with a teammate

Use Share from a Desk thread’s menu. Choose read or write access for an eligible coworker. Readers must sign in, and current sharing stays within the company. It is not an anonymous public link. Read access does not grant permission to change the owner’s workspace files.

Choose your language and appearance

Desk Settings offers English and Spanish. Apps, Work and Studio follow the account/Desk language. Open the account menu to choose Light, Dark or System for the workspace, sign-in and password recovery. The choice stays in this browser. The public landing and policy pages use a light appearance. This guide is available in English and Spanish.

Ask for the language you want in the prompt. Interface language and company branding should not decide the language or branding of your AI output.

Move around quickly

Use the shared search/command control to find a page or task App. On a phone, choose More to see every tool. The Guide link is in navigation on every screen.

Work with Gmail, Drive, Calendar and Tasks

Open Connections in navigation and enable the Google capabilities you need. Google asks you to approve access to your own account. New write or Tasks features can require another Google approval.

Google access is still in testing. Only approved Google test accounts can connect. Missing scopes, a disabled API or verification restrictions can prevent an action even when the account is connected.

Ask the assistant to work with Google

  1. Connect your Google account once in Connections and approve the services you need.
  2. Open Desk, choose one AI engine and ask for your Google task. For example: Upload this attachment to my Drive. A connected request goes directly to the assistant.
  3. The assistant reads or completes the requested action and shows sources or a result link. You do not need to approve the same task again.
  4. To review first, ask for a draft or say Prepare this for review; do not send yet. You can edit that draft before confirming it.

Continue a saved draft in chat. Ask Show that draft to read it, then Send that email draft or Save that document draft when you want it completed. The assistant uses the exact saved version and its original action ID. If several drafts match, name the one you mean. Edits you have not submitted in the browser are not part of the saved draft.

Your connected Google services are available automatically in chat, including follow-ups, reopened chats and engine changes. There is no extra activation switch. Access is checked before each request. Disconnect Google in Connections to stop future access. Google tools use one engine at a time. Missing recipients, dates or target files may need clarification. If a result is uncertain, check its status and the destination before submitting it again; the platform does not automatically replay an uncertain write. Manual browsing remains available without sending previews to AI.

Examples you can ask

  • Email: “Find unread emails from my client and draft a reply to the latest one.”
  • Drive: “Find the monthly statement PDF and summarize the main expenses.”
  • Calendar: “Check my upcoming events and prepare a meeting for Friday at 2pm Eastern for 30 minutes.” Include the timezone and attendees if needed; creating the meeting can send invitations.
  • Tasks: “Create a follow-up task for this client due October 12.” Or: “Show my tasks and mark the completed proposal task done.”
  • Google Docs: “Create a Google Doc with this project summary.” Or: “Append these meeting notes to the Notes tab of my project document.”
  • Google Sheets: “Create a spreadsheet with name, email and follow-up status columns.” Or: “Append these contacts to the Clients tab of my tracker.” Name the exact tab and rows. To prepare a draft of changes to existing cells: “Read cells B3:C4 in Sheet12, then prepare these replacement values.” The review shows the current and proposed cells.
  • Understand an image: Attach an image and ask: “Read the details in this screenshot, then prepare a task for the next step.” Say “create the task” to save it directly, or “prepare a draft” to review first.
  • Upload an original file: Attach a WEBP, image or supported document, then ask: “Upload this original file to my Google Drive Marketing folder.” The assistant uploads the original file and returns the result link.
  • Drive text files: “Prepare a plain-text project summary to save in Drive.”

Save an attachment from Gmail to Drive

Ask: “Read the latest invoice email and save its named PDF attachment to my Drive Invoices folder.” The assistant uses the original attachment without asking you to download or attach it again.

This Gmail transfer supports named files up to 5 MB (5,000,000 bytes). Give the attachment name when an email has several files. The file keeps its original bytes and type; you can request a new destination name. Gmail reading and Drive writing permissions are required. Inline images and larger Gmail attachments are not supported by this transfer.

Forward an email and follow up

Ask: “Read the latest invoice email, forward it with invoice.pdf to finance@example.com, then create a follow-up task for Monday.” The assistant reports each action separately. If a later step fails, earlier completed actions remain completed; an uncertain email is not sent again automatically.

What the assistant can read and change

Drive reads bounded text from supported text files, Google Docs, Sheets, Slides, PDFs and Office documents. Scanned PDFs, encrypted files and oversized files may not yield text. Drive writing creates or updates plain-text files. You can create a folder in My Drive or an exact writable folder. Moving a file currently requires an owned, unshared My Drive file and destination folder; folders, shortcuts and shared drives cannot be moved here. You can also create native Google Docs and Sheets. Docs append adds a new paragraph at the end of the selected tab and checks that the document has not changed since preparation. Sheets append adds rows of text, numbers or true/false values after the selected tab’s last row with data, starting in column A. Creation and append do not replace existing native content. Requested Sheets replacement changes only the displayed cell range; empty values clear those cells. Cell styles stay, but rich-text links and smart chips in replaced cells may be removed. A preflight check detects earlier changes, but simultaneous edits may still occur. Specify a number or true/false when that is the value you want. Text such as 0012 stays text; a blank clears a cell in a replacement. Text beginning with = stays literal and does not run a formula. Calendar reads your primary calendar. Ask for a date range, including past dates, such as “Find my meetings from March 1 to March 7.” Each search supports up to one year; without dates, it shows the next 30 days. It also supports event creation and updates. Tasks deadlines are dates, not timed alarms.

For an all-day event, give the first and last dates, such as “Create an all-day event from October 12 through October 14, 2026.” Ask for calendar defaults, no reminders, or a popup or email reminder at a specific number of minutes before the event. If you request a draft, its Last day field includes that day. Updating other details keeps existing reminders unless you ask to change them.

Google-enabled requests can include attachments. Attach up to eight files. Supported types include PNG, JPEG, GIF, WEBP, PDF, DOCX, TXT, Markdown, CSV, JSON, HTML and ZIP. Documents and ZIP archives must be 32 MB or smaller; images use the service’s configured image limits. The assistant receives file details and bounded extracted text from supported documents. Edited-message resends still support text only; attach files to a new turn.

When you ask about an image, or your task needs its contents, ChatGPT, Claude and Grok can inspect a preview. Inspection supports up to four distinct images per turn. Brandpatch corrects orientation, removes metadata and sends the selected provider a JPEG preview of the first frame, resized to at most 2,048 pixels on its longest side. The provider may retain that preview in its conversation history.

Inspection requires images of no more than 20 megapixels and 10,000 pixels on either side. Ask about a smaller copy if an image exceeds those limits. Drive uploads keep the original file bytes and create a new file; image inspection does not change the upload.

Email sending supports plain text, threaded replies and forwarding with selected original attachments. Ask to forward an email with its named original files; name the files when there are several. Forwarding supports up to four selected files totaling 8 MiB, keeping their exact original bytes, names and types. Unselected attachments and inline images are omitted. A plain-text forward without selected files still omits attachments. You can also send up to four original chat attachments totaling 8 MiB. Saving an unsent Gmail-native draft requires Gmail draft permission. A platform draft remains available in chat. If a send or change is uncertain, check the saved receipt and Google before preparing another action. Brandpatch does not automatically repeat uncertain writes.

Stop the connection

Choose Disconnect in Connections and confirm. This removes the current local connection and attempts Google revocation. Original Google content and completed actions remain. Remove Brandpatch access in your Google account too if revocation cannot be confirmed.

Use your Notion pages

  1. Save your personal Notion token in Connections. Existing Read content access is needed for reads; requested insertions also need existing Insert content access and a shared destination. Saving a token does not verify those capabilities.
  2. Choose ChatGPT, Claude or Grok in a private Desk chat. Ask to find pages by title, read an exact page, create a note under a selected page or append structured text to a selected page or supported text block.
  3. For a write, include the exact destination link or ID. For example: “Read this Notion page and create a summary note under this exact parent page: [paste its link].” If the destination is missing or ambiguous, the assistant asks for it.

An explicit request to create or append runs without another confirmation. Asking for a draft or proposal saves an encrypted local draft and makes no Notion write. Ask to execute the saved draft when ready; if several exist, use the action ID shown in its receipt.

The saved action card shows completion, refusal or uncertainty. Check saved status reads only the local receipt. If the outcome is uncertain, Notion may already have changed: inspect the destination before requesting more work. The action is never automatically repeated, including after a worker restart.

This scope supports at most two requested insertions per turn, twenty flat plain-text blocks per insertion and thirty-two kilobytes of structured payload. It does not replace or delete content, edit databases, use templates or mentions, upload files or provide Notion OAuth. Title search and page reads can remain partial.

Notion conversations stay private, outside projects and sharing, with one AI and no attachments. Saved history, local drafts and receipts remain readable after disconnecting; fresh reads and requested writes need the current personal connection. Creating a note or appending text can still appear in Notion history, audit logs, webhooks or existing automations.

Open Connections

Use your Microsoft mail

If Connections shows Setup required, the owner must first configure the protected Microsoft app registration. This feature is unavailable until that setup exists.

  1. In Connections, connect your own Microsoft account for mail reads. Allow mail sending only if you need it; Microsoft asks for that delegated permission.
  2. Choose ChatGPT, Claude or Grok in a private Desk chat outside projects, with no attachments. Ask explicitly: “Read my Outlook inbox and draft an email to selected@example.com.” Replace the example address with your exact chosen recipient.
  3. Ask for a draft to review first. It stays encrypted here and does not create a Microsoft draft or send mail. A clear request to send authorizes the exact email without another confirmation. Ask to send an unchanged saved draft when ready; use its saved action ID if several exist.

The inline receipt distinguishes a local draft, refusal, acceptance for processing and an unknown outcome. Accepted does not prove delivery. Check saved status only reads the local record. An uncertain send is never automatically repeated: check Outlook Sent before requesting more mail.

Mail lists cover at most twenty messages and may be partial. You can filter by exact subject and a bounded date window. Exact message reads show plain text, with shortening labeled; attachments are not opened. Sending supports up to five selected recipient addresses and plain text only. Shared mailboxes, send-as, attachments, Calendar, OneDrive and Slack are outside this connection.

Mail content used in Desk goes to your selected AI. Saved mail conversations, local drafts and receipts stay private. After disconnecting, you can review saved history without fresh provider access; new reads or sends need the current connection.

Open Connections

Read Microsoft Learn documentation

Check Connections for Microsoft Learn setup. An admin must configure the supported public documentation service before it is available.

  1. Open a private Desk chat outside a project and choose ChatGPT, Claude or Grok.
  2. Ask explicitly: “Search Microsoft Learn for Azure App Service and read the overview.” You can also ask about an exact Microsoft Learn article link.
  3. The assistant searches documentation, reads selected articles and shows original source links. Follow up or review the saved answer in the same private chat.

This tool reads public documentation only. It does not connect to your Outlook email, OneDrive files, Teams or Microsoft 365 account. Use one engine without attachments or another connected service in that turn. Partial or shortened results are labeled.

Open Connections

How your Google data is handled

Your connection belongs to your signed-in Brandpatch account. Other users cannot browse its credentials or connection through their accounts. Manual preview content passes through the server to your browser without automatically becoming AI context.

When you submit a chat request, your connected Google services are available to the selected assistant. If it reads data for that request, fetched content goes to the selected provider and can appear in saved replies, job events and provider history. Saved chat content can be included in later requests or visible to authorized collaborators if you share the chat. The Google panel and privacy page explain this sharing.

Credentials and prepared action drafts are encrypted. This does not mean every saved chat, job, file or backup is encrypted. Pending action reviews expire after one day. A receipt records the action result and prevents automatic replay after an uncertain write.

Local daily database backups retain the newest seven copies. Off-host application backups include the database and workspace files and are configured to expire after 7 days for daily copies, 28 days for weekly copies and 90 days for monthly copies. Older object versions are configured to expire after one day. Lifecycle cleanup does not guarantee deletion at an exact time. Manual and release snapshots have no unified expiry schedule. These backup periods do not set a deletion schedule for live saved work.

Disconnecting removes live credentials and attempts Google revocation. It does not erase saved work, provider copies, completed changes or historical backups. Contact info@mybrandpatch.com for privacy and account-removal requests.

Optional: skills, connections and API

You can skip this section and still use the main tools.

  • Skills are reusable instructions that help an agent with a task. Available skills can be used from the Desk’s supported controls; they are not separate accounts or a guarantee of an output.
  • Connectors connect external tools to the platform when configured and authorized. The connected-tools control shows available connections. An admin may need to set them up first.
  • API lets scripts submit work and read results. See the API reference for supported endpoints and examples.
  • IDE setup connects coding editors such as Cursor or Roo Code. Follow Connect IDE and use a platform bg_ key if your account has permission to create one. Keep keys private.

Connections and external actions depend on account permissions and configuration. This guide does not grant access or connect anything automatically.

For owners and admins

The Admin screen appears only for permitted roles. Some accounts can open their own Keys section without full admin access.

  • People and roles: create approved accounts, manage access and issue password setup/reset links.
  • Companies and teams: organize users and their team visibility.
  • Keys: manage platform API keys within your permissions.
  • Host and connectors: manage connected AI logins and supported integrations.
  • Usage and quotas: inspect activity and limits. Some token values are provider measurements and some are estimates; read the labels.

Review an access request

  1. In Admin, open Team and scroll to Access requests. Search by name, email or what the person wants to do.
  2. Use New for unread requests, Reviewing while deciding and Closed when you are finished. These labels organize your work; they do not grant access or send email.
  3. Choose Prepare account to fill in a suggested username and email. Review the company, role and limits, choose a temporary password, then use Create user. A filled form is preserved; use Clear form before preparing another request. Reopen a closed request first.
  4. After creating the account, choose Review request status to return to its status. If the email already belongs to an account, Find account opens the matching account instead.

If a status save cannot be confirmed, use Refresh requests before changing it again.

Open Admin

Public launch is coming. Free signup and paid subscriptions are planned. Account approval is still manual; plan prices, included usage and checkout are not available yet. An access request is not a subscription.

When something gets stuck

“Sign in again” or an expired session
Sign in with the same account. Private content is cleared when the session expires. Retry the read or reopen the original thread after signing in.
The page could not load
Use the visible Retry action. This checks access or reloads saved information; it does not automatically resend your task.
“Waiting” or an unavailable engine
Check Status and Work. Capacity may be full. If the Desk offers another engine, choose it deliberately; results can differ.
The model is at capacity
For a failed ChatGPT reply in your own Desk chat, use Choose another model. If the saved originals are still available, Brandpatch restores your request and all its files as an editable draft. Choose a different model, then press Send. Undo brings back the draft you had before recovery. Older requests or changed files may need to be reattached; the notice explains this and keeps your current draft. Opening recovery and choosing a model do not send work or change the chat’s default model.
A submission was not confirmed
In Desk, choose Check status beside the composer. This reads the original send receipt without sending again. If accepted, continue in the same thread. If not found, the explicit retry keeps the same request key and original files; it never runs automatically. After 24 hours, an unconfirmed send may still have run. Check the original thread, then choose “Start a new request” to close the old recovery and write a new request. Expiry does not mean the original was rejected. Studio’s Check status action checks a known job. For a job with no ID, use “I reviewed the thread” only after checking the original thread. That acknowledgement does not prove a job failed.
An upload was not confirmed
Check the destination workspace or library for the file before uploading again. Retry reads the file list; it does not automatically replay the upload.
An old link says unavailable or access denied
Check that you are using the correct account and still have access to that workspace. Retry authorized history. A rejected job link should not hide other results you are allowed to see.
The answer has the wrong tone, language or detail
Follow up with the exact correction: “Reply in English,” “Remove all branding” or “Keep only the three main points.” Check important facts before using the output.

A simple recipe for a better request

Tell the AI four things: the task, the context, the format and the limits. Paste the facts it needs. Say what it should not invent.

A prompt you can adapt

Task: [what you want done] Context: [the facts or source material] Format: [email, checklist, table, image, code] Limits: [length, tone, language, what to avoid] If an important detail is missing, ask me instead of inventing it.
Open Desk

Writing an example here, copying it or opening a linked tool does not send it. You stay in control of when to run it.