Threads:
Branch a Conversation From Any Message

A thread is a new conversation branched off an existing one at a specific message. It starts with everything said up to that point, then goes its own way — leaving the original chat untouched. It's the feature for the moment when an answer raises an interesting side question you don't want to spend your main conversation on. This is one of the features of Secret Chat AI, the AI aggregator.

How to create one

Hover any message and press Thread. You're asked to name the branch, and a new chat appears in the sidebar nested under the conversation it came from. If the parent chat belongs to a project, the thread joins the same project automatically.

Threads can themselves be branched, so a genuinely tangled investigation ends up as a tree you can read rather than a single chat you have to scroll.

What it is for

  • Chasing a tangent. An answer mentions something worth a detour. Branch, chase it, and come back to a main conversation that never lost the plot.
  • Trying a second model on the same question. Branch at your question, switch the model in the thread, and you have two answers to the identical prompt sitting side by side in the sidebar: a manual second opinion, kept in the sidebar for as long as you want it. When you would rather have the comparison done for you, that's the AI Council's job (see below).
  • Testing a different approach. Rewriting a brief, arguing the opposite case, or asking for a much shorter version — all without throwing away the version you already like.
  • Splitting a conversation that grew two topics. Branch the newer topic out and each has room to keep going.

Threads versus editing a message

Both change the direction of a conversation, but they do different things:

  • Editing adds your reworded question to the same conversation as a new message — the original stays above it unless you delete it. Use it when the first attempt was simply worded badly and you want the retry in context.
  • Threading starts a separate chat from that point, so the two directions don't sit in one transcript. Use it when you want to compare answers, or come back to a tangent later.

Threads versus the AI Council

Both give you a second opinion; they differ in who does the comparing and how many models are involved.

  • A thread is one extra answer, from a model you pick, kept as its own chat. You read both and draw your own conclusion. It's the lightweight option, and it's the right one when you mainly want to keep the tangent.
  • The AI Council asks two, three or four models from different companies at the same time and then has a referee model compare their answers claim by claim, marking which model asserts each point, which contradicts it and which ignores it. Nothing lands in your sidebar to reconcile by hand; the disagreements are listed for you. A two-model Duet runs free within the daily quota.

In practice the two combine well: branch when you want to keep exploring, and switch a chat to a council when you have reached something you need to be right about. A council reads the whole conversation, so you can turn an existing chat into one and the answer under review is the one already on your screen.

Cost and privacy

A thread is an ordinary conversation, so messages sent in it use credits exactly as they would anywhere else — branching itself costs nothing, since it sends nothing to a model. Like every other chat, threads live in your own browser, are included when you export your data, and disappear with the parent if you delete it.