File reading and chat history
The exact limits on text taken from attachments, earlier files in a long chat, and when to attach a file again.
June can work with an attachment even when only part of its text fits in the message. The limits below describe what the model receives first; on the web and in the app, ask June to open the full file in its sandbox when you need every row or page checked. For the steps to upload, preview, edit, or make a file, start with Files.
What June reads from each file
| File | What June gets |
|---|---|
| The PDF itself | |
| Image (PNG, JPEG, GIF, WebP) | The image itself |
| Text, Markdown, CSV, JSON | The text, up to 200,000 characters |
| Excel workbook | Each sheet as a table, the first 4,000 characters of each, up to 20,000 in all |
| PowerPoint deck | The text of each slide, up to 20,000 characters for the whole deck |
| Word document (.docx) | No text taken out: the file goes as it is, and the model may not read it. Attach a PDF export instead |
| Anything else | The file as it is, with no text taken out of it |
All the files in one message share the 200,000 characters, in order; once they are used up, the rest is left out and June is told so. On the web and in the app, June can still open the file in its sandbox and read the rest there.
The workbook has more rows than you can see. Open it in your sandbox and total column D across every sheet
Files in a long thread
June reads the files of your latest message in full, within the limits above. Files from earlier messages work differently:
- Text comes back shortened to its first 20,000 characters. On the web and in the app, June is told where the whole file is in its sandbox.
- A PDF or image comes back whole, and counts toward how much of the thread June reads back.
- In June's sandbox, on the web and in the app, every message's files have a folder of their own, so June can work on an earlier file in full. In an organization chat, those are the files you attached yourself. Earlier messages' files stay there as long as June still reads the message back (below); after that, attach the file again. In Slack, June has no sandbox.
On the web and in the app, June reads back the newest 500 messages of a thread. If they do not fit about 64,000 tokens (roughly 250,000 characters) less your new message, it drops whole turns, oldest first, each starting from one of your messages, until the rest fits. An earlier file or Google Drive link takes about 4,000 tokens of that. In a long thread with many files, start a new chat and attach what matters, or put the files in a project, where every chat in the project can open them.
If June cannot find an earlier attachment, attach it again with your next request. For an exact calculation, name the file, sheet and columns and ask June to use the full file rather than the shortened text.
When to start a new chat
For work that has moved to a different task, start a new chat and attach what matters. For several chats that need the same material, use project files. Say which document or sheet June should use instead of relying on an old attachment being in view.
What June can do in a reply
Every tool June can use while it answers, when it reaches for each, and where one is not available: the web and the app, Slack, or a scheduled run.
June's sandbox
The small Linux computer in the cloud where June runs code for you: which chats share one, how your work gets out, what is installed, and what it refuses.