Strata PlatformDocs
↑↓ navigate ↵ open esc close

File upload and analysis

You can attach files directly to a chat and ask Strata to read them, summarize them, transform them, or join their contents with database results.

Supported file types

Strata accepts these formats out of the box:

  • Documents — .pdf, .docx, .txt, .md
  • Spreadsheets — .xlsx, .xls, .xlsm, .xlsb, .csv
  • Slides — .pptx
  • Structured data — .json, .xml, .sql, .log
  • Images — .png, .jpg, .jpeg, .gif, .webp
  • Archives — .zip, kept as one file that the AI opens with Analysis (see Zip files)

The file picker (the + button) only lets you select files with an extension in this list. Some additional types — source code files (.py, .js, .ts, .html, .yaml, .sh) and image formats like .bmp and .svg — can be dragged in and the server will extract or pass through their content. The older .doc and .ppt formats aren't read; save them as .docx or .pptx first. Extensions not recognized by the server at all are rejected with an upload error rather than silently failing.

Zip files

A .zip stays one file, the way Claude handles it: Strata stores the archive as it is and doesn't unpack it. Its chip reads Archive with the zip's size, and it counts as one of the 20 attachments. Nothing inside it is read when you attach it, so the AI knows only that the archive is there; when you ask about it, the AI opens it with Analysis, lists what is inside, and reads only what your question needs. Each time it opens the archive is an Analysis run.

  • Size: the 20 MB upload limit applies to the zip itself. An archive whose contents would inflate past the decompression limits is refused when you attach it, without being unpacked.
  • Password-protected zips are accepted. The AI asks you for the password when it opens one.
  • Files it unpacks stay for the conversation: a later question in the same chat can read them again until the chat has been idle for about 30 minutes. When that Analysis session ends, its files are deleted.
  • When a zip can't be opened, the chip fails and says why: Zip files are opened with Analysis, which is off for you. Upload the files inside it instead., Your organization checks attachments before the AI reads them, and it can't check inside a zip. Upload the files inside it instead., or reports.zip isn't a zip archive.
  • A zip can still be attached to an email draft or saved to OneDrive whole, like any other attachment.

An Office file that was renamed to .zip is read as the document it really is.

A zip that arrives as an Outlook attachment or sits in OneDrive (up to 10 MB) is not unpacked when it is read either. In a chat where Analysis is on, the AI is handed the archive to open the same way; otherwise it tells you it can't open the zip in that chat and asks for the files inside. Voice mode can't open zips, and knowledge base syncs don't unpack them.

Attaching a file

Four ways:

  1. Plus button. Click the + (Add files and more) button next to the input box and choose Add file or photo, then pick the file. You can also type / in the input box to open the slash menu and choose Attach file or photo.
  2. Drag and drop. Drag a file onto the chat area — a "Drop files here" overlay appears. Release to upload.
  3. Paste an image. With an image on your clipboard, focus the chat input and paste (Ctrl/Cmd+V).
  4. From OneDrive. With OneDrive connected, type @ in the input box and pick the file under Files, or choose Add from OneDrive in the plus menu. The file is not uploaded: it joins your message as a OneDrive chip, the AI reads it from OneDrive before answering, and it counts toward the 20-file limit.

You can attach up to 20 files at once and mix types — for example, an Excel workbook plus a PDF plus a reference image. Strata tells you the limit at the moment you reach it rather than labelling the menu.

What happens during upload

Each attached file appears as a chip above the composer with the file name. While the file is uploading and being read on the server, a spinner sits where the file's icon goes and the line under the name reads Processing…. When it finishes, the spinner is replaced by the type icon (or a thumbnail, for a photo) and the line shows the type and size, for example PDF · 240 KB. The chip stays the same size throughout, so nothing around it moves. A small remove (x) button on the chip takes the file off. If the upload fails, the line reads Failed.

The type is taken from the file's extension, such as PDF, Spreadsheet, Document, Presentation, CSV, Image or Text. A name with no extension reads File (or Text for a plain-text file), and an extension Strata doesn't recognize is shown as it is, for example PARQUET.

Strata extracts text from documents and slides on the server side so the AI can read them. For images, the binary is passed through to the AI in vision mode — the model reads the image directly, including text within it. For Excel and CSV, the AI receives the tabular content directly. For PDFs, Strata does both: it extracts the text layer and renders the first pages to images so the model can see the page — see Reading PDFs visually below.

Uploads run one batch at a time. If you drag, paste, or pick more files while a batch is still uploading, Strata tells you to wait for the current files to finish rather than dropping them silently.

Each uploaded file is stored with the conversation, so you can re-open it from the chat later — it does not need to be re-uploaded for follow-up questions.

When you send the message, each file shows above your message as a slim chip: the type icon (or a small thumbnail for an image), the file name, and the same type and size line, such as PDF · 240 KB. Pasted text shows its length instead, such as Text · 1,204 chars. Click a file chip to open a small window with the same name and line and a Download button; click a pasted-text chip to read the text.

Asking questions about attached files

Once a file is attached, the AI sees it on your next message. Try:

  • "Summarize this PDF."
  • "What are the top 5 risks called out in this document?"
  • "Compare this Excel to last quarter's data in the database."
  • "Pull the totals from this CSV and chart them."
  • "What does this image show?"

You can ask several follow-up questions and the file stays in context until you remove it or start a new chat.

Reading PDFs visually

When you attach a PDF, Strata does not only read its text — it also renders the first pages to images and shows them to the model, so the AI can see the page the way you do. That means it can read:

  • Charts, diagrams, and figures that have no text layer.
  • Scanned or photographed pages, where the text layer is often empty.
  • Layout, tables, stamps, signatures, and handwriting.

This runs automatically on any model that supports vision, alongside the normal text extraction. Because only the first pages are rendered, Strata will tell you which pages it was shown if the answer isn't in them — for a long PDF, point it at the section you care about. There is nothing to turn on.

Tip

For a scanned contract or a photo of a document, ask directly — "What are the signature dates on this?" or "Read the handwritten note in the margin." Strata reads the image even when the text layer is blank.

Editing uploaded Office and PDF files

Beyond reading a file, Strata can modify one you uploaded and hand back a new version. Ask in plain language:

  • Word, Excel, PowerPoint — "Add a summary section to this doc," "Update the totals in this workbook," "Replace slide 3." Strata inspects the file, applies the change in place, and returns the edited file as a download card.
  • PDF — "Delete pages 4-6," "Merge these two PDFs," "Add page numbers," "Fill in this form," "Stamp our logo on every page," "Redact every occurrence of this account number." Strata works on the real PDF and returns the result.
Warning

Redaction and drawing a box are not the same thing. When you ask Strata to redact text, the content is removed from the file — it cannot be recovered by copy-paste or text extraction. A drawn rectangle only covers content visually; the text underneath stays in the file. Ask for redaction, not a box, when the information must truly be gone.

Files from earlier in the conversation

Files stay usable for as long as the conversation exists: files you uploaded, and files Strata or a sub-agent made (workbooks, reports, documents, decks, analysis outputs). It doesn't matter whether the file came from your last message, from this morning, or from a session days ago, or whether you have reloaded the page since. Refer to it in plain words, such as "the Excel you made this morning" or "the contract I attached yesterday", and Strata can:

  • Save it to OneDrive — the stored file is uploaded as it is, with its own file type.
  • Attach it to an email.
  • Change it — for example add a column to a workbook or rewrite a section of a report. The edit comes back as a new version, and later edits build on the newest one. This works for older files too.
  • Read or analyze it again.
  • Give it to a sub-agent, which can open, edit, save or attach it. A save or an email by a sub-agent still goes through the chat's Approvals setting (see Sub-agents).

Strata uses the file it already has. It does not make the file again to do any of these.

A few limits:

  • This conversation only. A file from a different chat has to be attached again.
  • The most recent 25 files. In a long conversation, Strata keeps track of the 25 most recent files, with documents and uploads ahead of analysis outputs such as charts. Two different files with the same name are both kept. If you mean a file older than that, Strata says so and asks you to attach it again.
  • Turns that were not saved. A file made or attached in a turn that was not saved to your history is not available to later messages. That happens when Save AI responses to history is off, when Keep my extension-data chats in history is off and that turn used an extension, or when your organization's data policy withheld the answer.
  • Incognito. Nothing is saved, so Strata does not keep track of an incognito chat's files for later messages, and nothing carries over once you leave. If Strata can't find a file from an earlier message, attach it again or ask for it again.

Combining files with other sources

Strata can mix attached files with everything else it has access to:

  • File + database — "Join the customer IDs in this CSV with our Orders table and show me the total revenue per row." Requires a SQL Server extension. See /docs/integrations/sql-server.
  • File + email — "Compare this contract draft to the version Alice sent me last week." Requires Outlook. See /docs/integrations/outlook.
  • File + OneDrive — "Find a similar file in my OneDrive Reports folder." Requires OneDrive. See /docs/integrations/onedrive.
  • File + Analysis — "Run a regression on the columns in this CSV." Strata runs the work as an analysis in its isolated sandbox. See /docs/chat/python.
  • File + Knowledge Base — "How does this draft compare to our standard contract template?" Strata can search your organization's indexed documents alongside the file you attached. See /docs/chat/knowledge-base.

Removing an attached file

Click the small x on the file chip above the composer to remove it before sending. After you send a message with a file, the file stays attached to that message in the transcript — it does not get re-uploaded for follow-ups, and Strata can keep using it later in the conversation (see Files from earlier in the conversation).

Limits and best practices

  • 20 files per message. A message carries at most 20 attachments, pasted-text chips included. Drop more than that and Strata keeps the first 20 and says so in a line above the chips; add one to a full tray and it is refused. Send the rest in a follow-up message — Strata can still use the earlier ones (see Files from earlier in the conversation).
  • 20 MB per file. Each attachment is capped at 20 MB. A larger file is rejected at upload.
  • One copy per chat. If you attach a file with the same name and size as one already attached, the duplicate is rejected with a flash animation on the existing chip.
  • Server-side text extraction runs after the upload completes. For very large .docx or .pptx files this can take several seconds.
  • Long documents are trimmed. Extracted text is capped at roughly 80,000 characters per file. For a very long document, ask about the specific section you care about rather than relying on the whole file being in context.
  • Keep files focused. Strata gets more useful answers when you attach the specific file relevant to the question, not a whole folder dump.
  • Sensitive content. Files you attach are sent to the AI model the same way your messages are. All models run through your organization's own Microsoft Azure AI Foundry, so that traffic stays in your own Azure tenant. See /docs/admin/ai-engine.

What can go wrong

  • File chip shows an error. The upload or text extraction failed (network glitch, or the file is corrupt/encrypted). Remove the chip and re-attach.
  • The AI says it can't read the file. For most formats this means text extraction returned nothing. Scanned PDFs are the exception — Strata renders their pages as images so a vision-capable model can still read them (see Reading PDFs visually). If a scanned PDF still can't be read, the current model may not support vision; switch to a vision-capable model in the model picker. For other formats, re-save in a supported type.
  • Drag-and-drop does nothing. Make sure you drop the file onto the chat area (the "Drop files here" overlay must appear). Files dropped on other pages are ignored. If a file uploads but the AI can't use it, it may be an unsupported type the server couldn't read — re-save in a supported format.

Related