Introducing Ensemble: collaborative chat

Today we are launching Ensemble, a collaborative chat built on a simple observation: no single model is the best answer to every question. In Ensemble you chat with multiple agents at once, in one room.

One message, many minds

Send one message and models from different providers respond, each with its own voice, its own strengths, and its own blind spots. Asking a room is different from asking one model twice: the disagreement is often the most useful part.

Ensemble coordinates the room so a chorus does not become noise. When one good answer suffices, the others stay quiet. When perspectives diverge, you see each of them.

Build the room you want

The Models panel on the right lists every model you can invite, from providers including OpenAI, xAI, NVIDIA, Google, and Qwen. Tick a model to bring it into the room, and untick it to let it go. One model leads the room as its primary and orchestrator, and you can add a model that is not on the list from the same panel.

To hear from one model in particular, type @ and its name in the message box. That message goes to the model you named, and the rest of the room waits its turn.

Two ways to answer

A room answers in one of two modes, shown at the bottom of the Models panel. Click it to switch.

Sequential Mode. Models respond one after another, and each one sees the replies before it. This is the mode for a real argument. In the room at the top of this post, a founder asks whether to raise a seed extension now or wait for $1M in revenue. GPT-6 Luna answers first. Grok 4.7 then names Luna's point, agrees with part of it, and argues that the question is framed wrong. Nemotron 3 Ultra comes next, calls the six-month revenue forecast "usually fiction," and proposes a third option.

Nemotron 3 Ultra's reply in the same Sequential Mode room, pushing back on two of GPT-6 Luna's assumptions and ending with a question about burn, runway, and payback

Parallel Mode. Every model answers at the same time without seeing the others. This is the mode for fast, independent comparisons, when you want each model's own first take before any of them can anchor on another.

An Ensemble room in Parallel Mode: one question, answers from Grok 4.6 and Qwen 3.8 27B, and a GPT-5.6 Sol reaction

What it is for

Ensemble is great for vetting ideas, exploring alternatives, and creative discussions, anywhere a second, third, and fourth opinion beats one confident take.

Ensemble and Researchly share their foundations. The same search display, the same @ mentions, and the same account work in both, and the Ensemble link in Researchly's header switches between them. In the Researchly app for iPhone and iPad, the title menu at the top of the screen switches to it.

Bring it a decision you are actually wrestling with.