The Workflow bar
Every button in the Workflow bar at the top right of the canvas: the model chip, the terminal, the cost estimate, working folder, Loop and Advanced Loop, Loop versions, Run and its options, Endless mode, Software factory, Free canvas, Schedule, Save as Power, and the More workflow actions menu on a
The Workflow bar is the row of small buttons at the top right of the canvas. Every button in it acts on the workflow in focus, the one you are working on. Hover any button and a green outline appears around that workflow, so you can see which one the button will act on. Hover a button to read its name.
From left to right:
| Button | What it does |
|---|---|
| Warning triangle (only sometimes) | A hint that your default model may be too small for this workflow |
| Model chip | Sets one model for every agent in the workflow |
| Terminal | Opens a terminal in the workflow's working folder |
| est. cost | What the next run is likely to cost |
| Folder | Opens the workflow's working folder |
| Loop, and its small arrow | Loop layout; the arrow holds Advanced Loop |
| Clock with an arrow | Loop versions |
| Play, and its small arrow | Runs the workflow; the arrow holds Loop & Enhance… |
| Infinity sign | Endless mode |
| Factory | Software factory (Factory mode) |
| Three scattered cards | Free canvas |
| Clock | Schedule |
| Star | Save as Power |
Model hint, model chip and cost estimate
The warning triangle shows only when the default model looks too small for how this workflow is built, for example many agents, conditions, loops or an agent that takes in the work of many others. Hover it to read why (Model may underperform this workflow or Workflow complexity note). It is advice only and never blocks a run.
The model chip shows a colored dot for the provider and the model your agents use. It reads Mixed when the agents use different models. Click it to open Model for all agents: pick a connection at the top, then a model, or press Use default for that connection's default model. Your pick is written onto every agent in the workflow, as one step you can undo, and that connection becomes the workflow's default. An agent you add later uses that connection with its default model. To change one agent only, use the round badge on its card instead. See Choosing models.
The cost estimate reads est. and a dollar amount, or a range with ~ when there is no history yet. Click it for the breakdown by model and by agent. It becomes more precise after three or more finished runs. To hide it, turn off Show cost projection before run in Settings → Workflow → Budget.
Terminal
The terminal button sits right after the model chip. Click it and a terminal window opens inside Cerevisor, already in the workflow's working folder. It is a real terminal, the same one you would open yourself: Command Prompt on Windows, your usual shell (zsh or bash) on a Mac or on Linux. Commands, colours, Ctrl+C, git, npm and full-screen programs such as vim work as they do there. A workflow that was never saved opens the terminal in your home folder.
- Move it by its title bar and resize it from any edge or corner. A new terminal window opens where you last left one, at the size you last gave it.
- Several terminals: the + in the title bar opens another one, in its own tab. Click a tab to switch. The x on a tab, a middle click on it, or typing
exitcloses that terminal. The window closes with its last tab. - Hide (the line at the top right) puts the window away and keeps everything running. The terminal button brings it back, with what was on screen. While a hidden terminal runs, a small green dot sits on the button. The button is lit while the window shows.
- Close (the x at the top right) ends every terminal of this workflow. Closing Cerevisor ends them too.
- Each workflow has its own terminal window, named after the workflow in its title bar.
Copy and paste work as in your own terminal. On Windows, Ctrl+C copies when text is selected and stops the running command when nothing is, Ctrl+V pastes, and a right click copies the selection or pastes. On Linux, use Ctrl+Shift+C and Ctrl+Shift+V. On a Mac, use Cmd+C and Cmd+V. Every other key, Ctrl+S included, goes to the terminal, not to Cerevisor.
What you type in a terminal and what it prints stay on your computer. Cerevisor does not record it, and agents cannot see it or type in it.
Working folder
The folder icon names the folder this workflow's runs work in. Hover it to read the full path; click it to open the folder. A workflow that was never saved has no folder yet, and the icon says No folder yet. Save the workflow (Ctrl+S) to give it one. For a workflow in Factory mode it is the factory's project folder.
Loop, Advanced Loop and Loop versions
- Loop (the circling arrows) turns on loop layout: the agents are arranged in a circle, and Play then opens a small window where you choose Iterations (blank for no limit) and Delay between runs (seconds) before the loop starts. Click Loop again to turn it off. While a loop runs, a green badge on the button counts the rounds.
- The small arrow next to Loop opens Advanced Loop: Fix, Grow, or Fix then Grow, with stop conditions. It is a paid feature. See Loop & Enhance.
- Loop versions lists past loop rounds. You can look at a round's output, compare rounds, and write a round's files back to the working folder.
Run and its options
Play runs the workflow. When the workflow's input is on, Play first takes you to its input panel, and you start the run there with Send. Before anything starts, Cerevisor checks what the run needs (Checking what this run needs…), then shows what the run may do. While the workflow runs, Play becomes Stop. Play pulses green when the workflow is ready to run. It is dimmed until the workflow is saved, because a run needs a folder to write to. See Run and monitor.
What Play says depends on the workflow:
- With loop layout on: Configure & run as loop.
- In Factory mode: Start factory, Pause factory, or Stop item while an item runs. See Software factory.
- On a free canvas: off. See below.
The small arrow next to Play (Run options) holds Loop & Enhance…: run the workflow in rounds and fix failures between rounds.
Endless mode, Software factory, Free canvas
- Endless mode (the infinity sign) runs the workflow until your intent is met: the last agent judges the result, and if it is not good enough, Cerevisor adds a new wave of agents and runs again. It opens a setup window first. See Progressive Mode.
- Software factory opens Settings → Factory mode. Its tooltip is Deploy the factory, or Factory mode settings once the factory is on (the button is then lit). See Software factory.
- Free canvas switches this workflow between waves and a free canvas. It sits right after the Software factory button and is lit while it is on. See the next section.
Free canvas or waves
A workflow can be laid out in two ways:
- Waves (the usual way): agents sit in columns from left to right and Run works through them in order, wave by wave. People also call this a structured, ordered or deterministic workflow.
- Free canvas: agents sit anywhere, with no order between them. You put them into named groups and work with each one through its chat, or message a whole group at once. People also call this a freeform canvas.
To switch, press the Free canvas button in the Workflow bar. Press it again to go back to waves: each agent goes into its nearest wave. On a narrow canvas, use the Free canvas row, which reads Back to waves while the free canvas is on.
On a free canvas Run, its options, Loop and Advanced Loop are off, and Endless mode and Loop & Enhance refuse, because there is no order to run in. Hover Play to read Free canvas is on, so the workflow does not run as a whole. Message your agents directly, or switch back to waves to run it. A chat with one agent, and Run this agent, still work. Everything about groups is in Free canvas.
Schedule and Save as Power
- Schedule (the clock) opens the schedule for this workflow. A blue count on it shows how many active schedules it has. An amber ! means a schedule in this world has no workflow to run; open Schedule to choose one. See Scheduling and pipelines.
- Save as Power (the star) saves this workflow as a reusable ability that other agents can call like a single tool. It is off until the workflow has at least one agent.
On a narrow canvas: More workflow actions
When the canvas gets narrow, the bar makes room in two steps. First the cost estimate hides and the model chip shrinks to its dot. Then the less used buttons fold into More workflow actions (the three dots). A blue dot on it means the workflow has an active schedule.
The menu holds, in order: Set model for all agents…, Terminal (Hide terminal while it shows, marked On while a hidden one runs), Open folder (with the folder's name), Loop versions, Endless mode, Schedule (with its count), the Software factory row (Deploy the factory or Factory mode settings), Free canvas (or Back to waves, marked On), Save as Power, Branch workflow and Share this workflow….
- Branch workflow copies this workflow into a subfolder so you can take it in a different direction. If it has run, the branch continues from where the run ended. It waits until a run has finished.
- Share this workflow… saves this workflow as one file you can send to someone. See Sharing a workflow.
Loop, Play and their small arrows always stay in the bar.
In simple view
Simple view (the eye button in the title bar) keeps the Workflow bar but makes it thin: smaller buttons, and no model hint, model chip or cost estimate. The terminal button stays. Turn simple view off to see the others again.