Watching her think

While Abbie is working on a reply, a strip appears above it showing the run as it happens: a header bar you can expand, then one block per step of the work. Not every reply takes the same route through this — a quick factual question and a genuinely hard one look different in the strip, and you have some say in that yourself.

​

Fast, Auto and Thorough

Next to the message box is a small chip showing the current mode and model. Selecting it opens a panel with three modes: The same panel has a model picker: Auto lets Abbie pick the model, or you can pin any active provider, or one specific model. A choice made here applies to every message you send in this chat until you change it again — it isn’t a one-off for the next message alone. To make it stick beyond this browser and this sign-in, the chat’s owner can select Make this the default for this chat, which saves it as this chat’s own default in Hub; it still shows up as a per-chat override on top of your account’s usual routing, so a chat that’s been quietly running on a lighter setting doesn’t read as “Abbie getting worse.” That option only appears for the chat’s owner, and is disabled until the chat’s first message has been sent, or while only a provider (not a specific model) is pinned.

​

When a request splits into more than a few pieces

Up to 6 independent pieces of work run automatically, without asking. Past that, Abbie shows a card first — “This splits into N pieces of work” — listing what those pieces are, with three options:

• Run all N in parallel
• Run as (a smaller capped number)
• Don’t run this

Run as N doesn’t drop the extra pieces — it groups them: the pieces beyond the cap are folded into the last slot’s task, so that slot’s work covers more than one piece at once, rather than leaving any of them undone. Nothing runs until you answer. If the moment passes before you do, the card shows No longer needed instead.

​

The run strip

Selecting the strip’s header expands it into blocks, appended one at a time as the run progresses — a plan line, then one block per expert slot, then judge and verify where they ran. The line at the top summarises the plan itself, for example:

Split into 3 tasks · will be checked · running 3 in parallel

or, for a request that’s being answered several different ways at once rather than split into parts:

Comparing 3 models · will be checked · running 3 in parallel

A one-task, fast-mode reply is plainer still — just One task, not checked. See What each stage means for what’s actually happening behind each block. Selecting a completed block expands it to show which model handled it, how long it took, and, for a signed-in session, a link into the full cognition log for that step.

​

What it cost

Each run strip has a credit icon. Selecting it breaks down what that run was charged, grouped into what it actually paid for — answering the question, checking the answer, and working out what to do before anything started — rather than a single opaque per-run price. Checking your own work is charged like anything else, and the breakdown is there so that’s visible rather than hidden inside one number.

​

Collapsed and expanded

While a run is fresh, the strip shows every block in full. Once you’ve moved on to read older messages, it collapses to a one-line summary — for example, “Thought for 12s · 3 steps · 6 credits” — how long the reply took, how many steps it went through, and, once the run has ended and the cost is known, what it was charged. Expand it again any time to see the detail.