Changelog

The session transcript was printing raw markdown: literal hash marks for headings, asterisks around bold text and stray horizontal rules.

The models write structured markdown by design, and the artifact and content pages already rendered it. Now the session view does too.

Headings, lists, tables, bold, links and code blocks all render properly, while you are streaming and after the reply lands. Your own messages stay exactly as you typed them.

When your model provider rejected a request, the session showed an empty reply and nothing else. The response looked like the model had simply said nothing.

The Vercel AI SDK's text stream drops error events by design, so a rejected key, a wrong model id or a rate limit ended the stream cleanly with no text and no exception. We now read those errors explicitly and surface them.

• A failed turn shows the provider's actual reason, not a blank bubble.
• A turn that fails after producing some text keeps that text.
• A failed turn no longer leaves an empty assistant message in the transcript, so your session history stays clean and is not replayed back to the model.

API keys are redacted from every message we show.

Quiver's marketing skills ship with reference material: scoring rubrics, templates, tactic libraries, worked examples. Sessions were loading each skill's instructions but none of the material those instructions point at, so when a skill said to score against a rubric or cross-reference a library, the model had to make it up.

• Sessions now load the reference files too, with a budget so a single skill cannot crowd out the rest of your context.
• If something is too large to include, the session is told which parts were left out rather than inventing them.
• You will notice this most in Strategy sessions, where the plans and audits have the real frameworks behind them.

Starting a session from a campaign attached it to that campaign but never told the session anything about it.

• A campaign-scoped session now opens knowing the campaign's goal, status, channels, dates and what is already attached to it.
• That means less re-explaining and fewer suggestions that ignore work you have already done.

Parts of the context page used to hand you a block of raw JSON to edit by hand, where one misplaced comma meant the save failed.

• Your ICP is now four labelled fields: definition, decision maker, primary use case, and jobs to be done.
• Competitors, messaging pillars, proof points, hypotheses and customer quotes are rows with add and remove buttons.
• Nothing you have not touched gets rewritten when you save, so older or richer entries survive editing a different field.

Onboarding drafts your product category, business model, core problem, why the alternatives fall short and your key differentiators. Until now it asked for all five and then threw them away, because there was nowhere to put them. They are stored now, they appear in your context, and every session starts with them loaded.

• Core problem and key differentiators are the two that change AI output the most. Sessions now open knowing what you actually solve and why someone picks you.
• Anything drafted without direct evidence from your site is still flagged as an assumption, so you can see what to confirm.
• If you onboarded before today, open your context and fill the new fields in. They will not backfill themselves.

Paste your product URL and Quiver drafts your whole marketing context — positioning, ICP, messaging, competitors, customer language and proof points — then drops you straight into the workspace. The six-step wizard is gone.

• Anything drafted without direct evidence from your site is flagged as an assumption, so you know what to confirm instead of trusting a wall of text you did not write.
• Your dashboard shows how many fields are still assumptions. They clear themselves as you edit a field or approve a change, so the number only moves when something real happens.
• No website yet? Describe the product in a few sentences and you get the same result.
• Setup now ends by handing you an API key and your workspace's endpoint, so you can connect your agent before you do anything else.

Agent clients that run in a browser can now connect to a workspace by signing in, with no key to paste.

• Each workspace is its own connection, so a team with several workspaces keeps them separate.
• Every connection is listed in settings with who made it, and you can revoke any of them immediately.
• Members and admins can connect. Viewers cannot.
• Local clients keep working exactly as before with a workspace key.

When a session turns up something that changes what your company believes — a competitor you actually compete with, sharper language for your ICP, a customer phrase worth keeping — it proposes the change instead of quietly rewriting anything.

• Each proposal shows the current value, the proposed value and why. Accept or dismiss it in one click.
• Accepting creates a new context version, so nothing is lost and you can always see what changed.
• Early sessions in a new workspace open with one short question about an assumption rather than guessing at it.
• Next steps from a session can become tasks with owners, due dates and links back to the work that created them.

Tell Quiver your public site once and publishing records the live URL for you.

• Set a site address and a path pattern in settings; every piece you publish stores its own address from then on.
• That address is what future performance data attaches to, so numbers land on the right piece without anyone mapping them by hand.
• Publish dates and campaigns can now be set from your agent, so an imported archive keeps its original dates and lands in the right campaign.
• Distributions to other platforms now require the URL, because a distribution without one is just a note.