File upload
File upload lets you upload and analyze local files directly in Glean, Public Knowledge, and Apps. You can query, summarize, and generate content from uploaded files without adding them to a connector. Upload limits vary based on the model your organization uses.
File upload must be enabled by your admin. If you don't see the upload option, contact your admin. For setup details, see File upload setup.
File content and metadata are stored in your chat sessions and are retained while the chat session is present in your history.
Supported file formats
What you can do
Upload files
Upload up to 5 files directly from your local computer, with each file up to 64 MB. Limits vary by model — see Upload limits.
Query in real time
Query the text content of uploaded files immediately after upload for instant analysis.
See document metadata
The chat displays document metadata, including title and file type, for easy reference.
Manage files
Delete before querying
You can delete uploaded files before submitting your first query. After you submit a query, files can only be removed by deleting the chat session history.
Retention
File content and metadata are retained while the chat session is present in your history.
Upload limits
The maximum file size and number of files per session depend on the model your organization uses:
| Model | File limit | Size limit |
|---|---|---|
| 128K models | 5 files | 64 MB |
| 32K models | 4 files | 32 MB |
| 8K models | 2 files | 16 MB |
Minimum file size for upload is 1 KB.
Analyze archive files
You can upload and analyze .zip archives directly in Glean. This is useful for working with bundled documents, code packages, or compressed datasets. Drag and drop a .zip file into the composer, then ask questions about its contents. Other archive formats are also accepted: .tar, .tar.gz / .tgz, .tar.bz2, .gz, and .bz2.
Unlike other uploads, archives aren't parsed and indexed up front. Instead, Glean opens them in the Agent Sandbox — a virtual computer with a file system, shell, and code interpreter — and inspects or extracts only the files needed to answer your question.
Requirements
Archive analysis runs in the Agent Sandbox, so both of the following must be true:
- Agent Sandbox is turned on for your organization. If it isn't, contact your admin.
- The chat is in Thinking mode. You can't add archives in Fast mode, and switching a chat to Fast mode removes any staged archive files.
Archive analysis runs in the Agent Sandbox and may be subject to usage-based pricing.
Archive limits
| Limit | Default |
|---|---|
| Archive (compressed) file size | 64 MB |
| Total uncompressed contents | 256 MB |
Archives count toward the per-session file limit for your model, shown in Upload limits. To protect against zip bombs, Glean also rejects archives with an unusually high compression ratio.
Security restrictions
Glean validates every archive before processing and rejects it if it contains any of the following:
- Executable or script files (for example,
.exe,.dll,.so,.sh,.bat, or.msi) - Nested archives, that is, another archive inside the archive
- Password-protected or encrypted entries
- Symbolic or hard links
- Entries with unsafe paths, such as a path that points outside the archive
Glean scans archives for malware, the same as for other file uploads.
Security and privacy
Security
- Files are parsed and scanned for malware before storage.
- Malware-infected files trigger upload errors.
Access controls
- Uploaded files can only be downloaded by the original uploader or by users who have access to the associated shared chat session.
- Files are not publicly accessible, even if the chat session was previously marked as public.
Limitations
- Video and audio files are not supported. Files outside the supported types above cannot be uploaded.
- For scanned PDFs, Optical Character Recognition (OCR) must be enabled to work properly. If you experience issues, contact your admin or Glean support.
- For analytical queries (row counts, filtered counts, aggregations) on large spreadsheets referenced from a connected source such as SharePoint or OneDrive, uploading the file directly typically produces more accurate results than referencing the file by URL. See Data Analysis: Overview for guidance.
Frequently asked questions
See also
- Chat Sharing - Share Assistant conversations with teammates
- File upload setup - Admin setup and configuration guide