Changelog

The dashboard had five cards and none of them knew what day it was. Tasks have due dates, content has publish dates, artifacts go live and competitor scans come due — and opening Quiver told you none of it.

This week now sits at the top, in three columns, split by time so nothing is ever listed twice.

Needs you is what was missed or is waiting on a person. Content past its publish date and still not live comes first, because that is the most time-critical thing on the page. Then overdue tasks, tasks due this week, work an agent finished and handed back for you to close, artifacts waiting on approval, and research nobody has read yet — which clears as you open it.

Going out is what the world has still to see this week: everything dated to publish between now and Sunday, live or not, with the ones that are not marked plainly.

Already shipped is what actually happened — named by the thing itself rather than by what kind of thing it was — and it says whether it was you, an agent or the system.

Overdue work shows up whenever it fell due. Something that missed its date last Thursday does not stop mattering because the week rolled over.

The week runs Monday to Sunday in your own timezone, and clicking a task opens that task instead of dropping you on the full list.

Your agent sees the same list. The new get_this_week tool answers exactly what the dashboard shows, so an agent deciding what to pick up is working from what you are looking at — and if any part of it cannot be read, it says so rather than reporting a quiet week.

Two things got fixed on the way. A task an agent submitted for review was being treated as closed, so it vanished from your task list and from close-the-loop reminders; it counts as open now, because it is waiting on you. And the Quick Start panel is gone — starting a session belongs on Sessions, one click away in the sidebar.

An agent working in your workspace could mark a task done. Nobody had approved anything — the work simply disappeared off the list as complete. Research had the opposite problem: an agent could save an entry and there was no state at all for "somebody should read this", so it sat in the list looking identical to everything already reviewed.

Both are one boundary now. Tasks have a review status, and over MCP an agent can reach it and no further: submit_task_for_review hands the work back with an optional note about what it did, and a person closes it in Quiver. The old complete_task tool is gone rather than quietly changed — a tool called "complete" that no longer completes anything is worse than one that isn't there.

Research entries can be marked reviewed, and the dashboard's Needs you section lists the ones nobody has read yet.

Every move is recorded: who handed the work back, who closed it, and through which interface. That is the point of the boundary — it only means something if you can see afterwards that it happened.

An artifact's title and content can be edited in place on its detail page, with a preview toggle. Until now the only way to change one was through the API or an agent over MCP.

Editing goes through the same path an agent uses, so the rules are unchanged: the artifact keeps its id and its status, everything pointing at it keeps pointing at it, and the previous version is saved to its history where it can be restored. Editing an artifact that is already live asks for confirmation first, because something is already measuring what shipped.

An agent writing an artifact was limited to 50,000 characters, and the refusal suggested splitting the document across several artifacts. One document in several pieces is worse than the thing the limit was preventing, so the limit is now 200,000 characters and the suggestion is gone.

The same limit applies when an agent edits an existing artifact, which it did not before.

The session page reclaims a few hundred pixels of vertical space for the conversation itself. The session context panel is collapsed behind a control in the header rather than always occupying a row, and the artifact save prompt floats over the foot of the transcript instead of pushing it up.

The save prompt stays until you dismiss it or save the artifact. It is never dismissed on a timer, because that would throw away the draft.

Creating a workspace while you were already inside another one opened the one you were already in. From there, switching to the new workspace by hand could show an error before eventually landing on billing.

The workspace picker now says which workspace it just created or selected, rather than leaving the next screen to work it out from whichever one happened to be active — which, for a moment after a creation, is still the previous one.

If you open that screen for a workspace you are not a member of, it now tells you straight away instead of waiting.

A content piece's slug could only be set when the piece was created. You can change it now, on the content detail page next to where the slug is shown, and an agent can change it through the update_content tool with new_slug.

If the piece is already published, the editor says so before you save: the public URL changes with the slug, and anything already linking to the old one stops working. The recorded live URL is updated to match, and both the old and new addresses are cleared from the cache so nothing serves the piece at an address it no longer has.

A slug another piece is already using is refused by name rather than failing as a database error.

A dialog taller than your window used to run off the bottom of the screen, taking its buttons with it. Long modals scroll now, whatever the window size or browser font.

Deleting something that cannot be deleted — a campaign with artifacts attached, say — looked like nothing happened. The refusal was real but it rendered behind the dialog. Now it appears inside the dialog, naming what is attached, and the dialog closes only when the delete actually succeeds.

The MCP key and client config on the setup screen also have copy buttons.

Proposed context changes were clipped to three lines, so a real positioning statement ended in an ellipsis — and you cannot approve a change you cannot read. Current and proposed values now render in full.

The same page had a horizontal scrollbar for anyone with a long value in their context; that is gone too. And the Context indicator in the header now takes you straight to the version history rather than dropping you on the editor.

There is now a public getting started guide at quivergtm.dev/getting-started. It walks the whole loop in about thirty minutes: confirm the context Quiver drafted, add one piece of customer evidence, create a real campaign, run a focused session, save the work, and log what happened.

It is linked from the homepage and from the docs, and it is the first thing the welcome email points you at.