Progressive Mode

Normally, building a workflow in chat and running it are two separate steps: you describe what you want, the chat builder adds agents to the canvas, and then you press Run and wait. Progressive Mode collapses that into one conversation.

With it turned on, every message you send does all four things at once:

1. It figures out what you're asking for — restating the goal back to you as it understands it.
2. It builds the next piece — usually one new step (a small group of agents) added to your workflow.
3. It runs that piece — and only that piece. Everything built earlier keeps its results; nothing re-runs unless you've asked it to change something already done.
4. It reports back — what it built, what came out of it, what it cost — right there in the chat.

Then it checks whether your goal is met. If not, it keeps going — building, running, and reporting the next piece — up to a few steps per message, so you're never left waiting on a black box and never have to open a separate run screen.

Turning it on

Open the chat panel (the same panel you use for "chat to build") and look for the Progressive mode switch near the top. It's per-workflow — turning it on in one workflow doesn't affect any other, and turning it off any time brings back the ordinary chat-builder behavior exactly as it was.

The first time you send a message with nothing built yet, it will ask you to pick a folder to work in — it needs somewhere to write files as it goes, and it asks instead of guessing.

What you'll see as it works

Each step shows up as a line in the chat: it says what it's building, then shows it appearing on the canvas, then reports when the run finishes — including what it produced and what that step cost. Markdown results show up right in the chat message; if a step produces an HTML page or another file, you'll see a card you can click to open it. The full run view is still there if you want it (click through from any step), but it never pops up and takes over your screen on its own.

Changing something already built

If your next message asks it to change or fix something it already built — not just add to it — it detects that automatically. It will tell you exactly what it's about to redo and why, before it redoes it, then re-runs only the affected piece and whatever depends on it. Everything unrelated stays exactly as it was.

If a step fails

It gets one automatic attempt at diagnosing and fixing its own mistake before it stops and asks you. If the retry also fails, it tells you honestly what went wrong instead of pretending it succeeded or trying again indefinitely.

The cost guard

Progressive Mode is a free feature — but it runs on your own provider key (your own Anthropic, OpenAI, Gemini, or other configured account), so real spend is a real consideration. Two limits protect you:

• A per-step ceiling ($0.50 by default) — if a single step is projected to cost more than this, it pauses and asks before running it.
• A per-conversation budget ($5.00 by default) — the running total for the whole conversation. It stops and asks once this is reached, rather than quietly spending past it.

Your running total is always visible at the top of the chat panel while Progressive Mode is active, so you never have to guess what a conversation has cost so far. When either limit is hit, you get an inline choice to raise it and continue, or stop there. These are estimates checked before a step runs, not a hard guarantee down to the cent — the visible running total reflects what was actually spent, which is the number to trust.

Keep going, pause, or stop

While a conversation is building, you'll see pause / resume / stop controls at the top of the chat panel:

• Pause finishes the step in progress, then waits before starting the next one.
• Resume picks up where it left off.
• Stop ends the conversation's current run. Anything already built and run stays on your canvas.

You can also just keep typing — a message sent while it's mid-step is queued and folded into the next step, so nothing you type is lost or ignored.

Reaching the step limit

Each message you send builds at most a handful of steps (4 by default) before it stops on its own and tells you where it got to. This isn't a failure — it's a deliberate checkpoint so you stay in the loop. Send another message (even just "keep going") to continue from there.

Progressive Mode vs. Endless mode

Progressive Mode and Endless mode (found in the Run-mode menu next to the Run button) are two ways of using the same underlying engine — the difference is whether a human is in the loop:

• Progressive Mode always comes back to you every few steps to check in. It's free, and it's the conversational way to build something with Cerevisor doing the driving.
• Endless mode keeps going on its own, without stopping to check in, until it decides your goal is met or it needs your budget raised. Running unattended like that is a paid feature.

Mid-conversation, Progressive Mode can offer to hand off to Endless mode — "keep going without me" — so it can keep working while you step away, picking up exactly where the conversation left off (same goal, same progress, same running cost). On a free plan, this shows an upgrade prompt at that moment rather than blocking you earlier. Handing control back the other way — from Endless mode into a conversation — is always free, since checking in with you is never the paid direction.