The timeline assistant

The timeline assistant is an AI helper in the viewer. Ask it about the timeline you are looking at, or, if you can edit the timeline, have it draft new items and edits that you review before anything changes. It can be wrong, so check what matters.

Asking

Open a timeline in the viewer and choose Ask (or press A) to open the assistant. Type a question, or pick one of the suggestions, which fill the box without sending it. It reads the timeline's items, dates, tags and sources, so it can summarise a stretch of time, point out which events overlap, spot dates that don't fit together or tell you what led up to something. Answers stream in as they are written, and Stop ends one early.

When an answer draws on something that is not on the timeline, such as background the model knows about the period, it says so in the text, so you can tell the timeline's own facts from general knowledge. Treat that part with extra care. Each finished answer shows what it cost in credits.

The assistant also answers questions about how Timescrolly works, drawing on the help centre and linking you to the right article; that lookup uses your credits like any other question. The help assistant on the help pages answers the same questions for free, without an account and with daily limits. What you write there is not stored on our servers.

Drafting changes

If you can edit the timeline, you can ask for changes: describe what you want, such as "add the main events I'm missing", or paste an article and ask for it to be turned into items. Viewers and people who can only comment can ask questions but not have it draft.

Drafts arrive as Proposed changes under the answer, one card for each new item, edit or deletion, with periods holding the events proposed inside them. Nothing has changed yet. As you look at the cards:

• Proposed items appear as ghosts on the timeline, so you can see where they would land before you apply anything.
• Every card starts ticked. Untick the ones you don't want; unticking a new period also unticks the events inside it.
• Expand a card to see every field, and for an edit, what it was and what it would become.
• Edit changes a card's Title, Start and End in place, which is the quick way to fix a date the assistant got wrong.

Choose Apply when you are happy. Applying is all or nothing: either every ticked change is made, or none is. Each change is recorded in the timeline's History under your name, like any other edit. An edited item can be restored to its earlier version from there, a deleted one comes back from the trash, and a new one is removed by deleting it. Applied changes are your edits; the history does not mark them as the assistant's.

Changed your mind? Choose Undo under the applied changes in the assistant to take the whole set back in one step. Undo is all or nothing as well, and it works only while none of those items has been changed again since, adding an image or a file included: if one has, nothing is undone and the assistant lists the items in the way. It also works only within your plan's history window, and only while the changes are still in the assistant: a new chat, or closing the tab, takes Undo away, and the timeline's history is where you restore items from then. An undo can't itself be undone. The history still holds each change, the undo included.

Importing from documents

If you can edit a timeline, you can also have a document turned into items. Open the Imports tab on the timeline's dashboard pages and choose New import, or choose Import from a file in the assistant. It uses the same credits as the assistant.

You can bring:

• PDFs, up to 25 MB and 300 pages;
• photos and scans (JPEG, PNG, WebP, HEIC and HEIF), up to 10 MB each;
• text files (TXT, Markdown and CSV), up to 5 MB each;
• Word and Excel files (DOCX and XLSX), up to 5 MB each;
• text you paste in, up to 5 MB.

Up to 30 sources go into one import. Several are read as one document, in the order shown: Sort by name puts photos in the order they were taken, and the arrows move a page up or down. You can add an instruction, such as "only events in 1939–1945, skip school reports". If the timeline has a start or end, saying so in the instruction saves credits, because drafts outside it cost credits too.

A timeline has one unfinished import at a time, so finish or discard it before you start another. Your imports are private: only you see them, and other editors see the items once you apply them.

A long document is read in two steps. First a sample, a first look at the opening pages with a rough estimate of the items, the time and the credits the rest needs. From there you can adjust the instruction and read again, start the import or discard it. Starting hands the rest to the background: you can leave the page, and you are notified here and by email when the drafts are ready. A short document goes straight to the review.

Reading sets aside some credits while it works and charges only what it used, usually much less. So a reading can be refused even though you still have credits left, because it needs more than that to start. The refusal costs nothing. Reading again may use credits again, and discarding an import does not give back the credits it has already used.

The review lists every draft. Most start ticked; a likely duplicate, or a draft outside the timeline's range, starts unticked. Untick the ones you don't want, or edit one in place. Drafts can carry a flag:

• Date uncertain: the date is a guess or the document is vague about it, and the draft says why.
• Possibly already there: the timeline already seems to hold this item, and the draft links to it, in a new tab, so you can compare.
• Outside the range: the date falls before or after the timeline's range.
• Quote not found: the quote the draft rests on isn't in your document, so check it against the original.

Needs a look shows every flagged draft, and Untick flagged unticks them all at once. Choose Add to add the ticked drafts. Before they go in, you can switch on Cite this document as a source, which gives every item one shared source, with the page number for a PDF.

When ticked drafts fall outside the timeline's range, the Add dialog asks what to do. Either the range is changed to include them, and they are added, or those drafts are left out. A draft dated after today on a timeline with no end date can't be added by widening; give the timeline an end date first, or leave it out.

Adding is all or nothing, and each item is recorded in the timeline's History under your name, like any other edit. Changed your mind? Choose Undo import on the finished import to take all the items out in one step. It works only while none of them has been changed again since and within your plan's history window, and it can't be redone. Two things stay as they are: the source, if you cited the document (it stays in the library, and can be deleted once the undone items are permanently deleted from the trash), and the timeline's range, even if the import widened it (change it in the timeline's details).

Your files are sent to Google to be read, processed in the EU, and deleted once the reading is done. The drafts are kept until you apply or discard them, for at most 30 days, and then they are gone. The privacy policy has the details.

Starting a new timeline from a document

You can also begin a timeline with a document. On the New timeline page, choose From a document, give the timeline a title and add your files. The title is yours, and it sets the timeline's address (its slug), which you can change later from the Details tab. The description and category are suggested from the document, and you can change them in the review before anything is created.

The timeline's start and end come from the items you add, and you can change them afterwards, for example to make it ongoing. The new timeline stays private until you choose otherwise when it's ready.

If you stop, discard the import, or it fails or expires, the empty timeline stays until it's deleted. Each time, you're offered the delete or, if you can't delete the timeline yourself, told that a team owner can.

Credits

The assistant runs on credits, a monthly allowance that depends on your plan: Free has 100 a month, Pro has 3,000 a month (a Team plan gives you the same for your own timelines), and a team's shared pool is 15,000 a month. Credits reset on the 1st of each month, UTC. The number left shows in the assistant's header, and clicking it shows how much you have used and when it resets. Pricing shows the current allowances.

On a team's timeline, people who hold a seat in the team (owners, admins and contributors) use the team's pool, so their messages never touch their own credits. Everyone else, including a collaborator from outside the team, uses their own.

What things cost varies with the size of the timeline, but roughly:

• a question costs about 3 credits;
• a follow-up in the same conversation about 1;
• drafting changes about 7;
• each of these more on a large timeline, since the assistant reads all of it.

Because it can't know the cost of an answer before it has written it, a message first sets aside its worst case, and afterwards you are charged only what it really used. So a message can be refused even though you still have some credits left: the credits left are fewer than that worst case. The refusal costs nothing, your message stays in the box, and you can send it again after the reset.

If you stop an answer, or the connection drops before it finishes, it is charged its full worst case, because what it really used is only known once an answer is complete. Stopping does not save credits.

Privacy

To write an answer, Timescrolly sends your message, the conversation so far and the content of the timeline to Google, which processes it in the EU through Vertex AI. Timescrolly keeps no conversations: not your messages, not the answers. For each message it keeps a usage record (who sent it, on which timeline, when, and the tokens and credits it used) for 13 months. The privacy policy has the details.

The conversation stays in this browser tab. It survives a reload, but it is cleared when you sign out, and New chat clears it at any time. Another tab or another device starts fresh.