Build and run your first workflow

You have Cerevisor installed and at least one provider connected. Time to ship a workflow.

This page walks two paths:

1. Single-agent hello world: a one-agent workflow that produces real output. Five minutes.
2. Two-agent handoff: a researcher passes findings to a writer. Fifteen minutes. The smallest workflow that demonstrates what makes Cerevisor different from a chat window.

You can stop after path 1 if you just want to confirm everything's wired up.

Path 1: Single-agent hello world

Open a blank canvas

From the Home Screen, click Describe a job. Cerevisor asks you to pick a folder for this workspace (that's where your files and results will live), then opens a fresh canvas with the chat builder ready on the left. You can tell the chat builder the job in plain English and let it draft the team — or close it and build by hand, which is what this page walks through.

If you closed the Home Screen, use Ctrl/Cmd+N to create a new workflow.

Add an agent

Three ways to add an agent, pick the one that feels easiest:

• Drag from the left palette. Open the Agent Palette by clicking its icon on the left edge of the canvas, then drag any role card onto the canvas (or just click a role card to drop it in).
• Right-click the canvas → Add agent.
• Open the Manual Controls menu at the top-left of the canvas and click Add Agent, then pick a role from the menu.

For this hello world, pick Researcher.

Configure it

Click the agent card to select it. Selecting a single agent opens the Agent Config popup automatically.

Set these:

• Name: Quick Researcher
• Instructions: Research the current state of multi-agent AI workflow tools in 2026. Summarize the three most notable platforms, what makes each distinct, and who they're for. Cite sources.
• Output definition: A 300-word summary in markdown with inline source citations.

Leave everything else at defaults. Click Save.

Run it

Click the Run button in the top-right of the canvas.

A confirmation modal shows you the estimated cost and the permissions the workflow will need. Click Run.

The agent card lights up. You'll see streaming output appear in real time. When it finishes, the card shows a green check; click it to see the full output.

That's it. You just ran a multi-agent workflow with one agent.

What just happened? Cerevisor generated a prompt from your configuration, sent it to your default provider, granted the agent the default tools for its role (a Researcher gets read-only tools like read_file, glob, grep, and find_skills), let it call those tools in a loop until it produced the output you defined, then saved everything to ~/.cerevisor/audit-logs/. You can replay this workflow any time by hitting Run again.

Path 2: Two-agent handoff

Now let's chain two agents. The researcher you just built will feed a writer agent that turns the findings into a publishable blog post.

Add a second column

Right-click an empty area to the right of your existing column and choose Add column. Or open the Manual Controls menu at the top-left of the canvas and click Add Wave.

Name it Phase 2: Writing.

Add a writer agent

Drag a Blog post writer role into the new column (or use any of the three add-agent methods from Path 1).

Configure it:

• Name: Blog Writer
• Instructions: Turn the upstream research into a 600-word blog post for cerevisor.com. Match the existing cerevisor.com voice: clear, second-person, no hype. Include a 1-sentence summary at the top.
• Output definition: A markdown blog post, 600 words, with a title and a 1-sentence summary line.

Connect them

Hover over the right edge of Quick Researcher. A handle appears. Click and drag it to the left edge of Blog Writer. A connection line draws between them.

That's a handoff: the writer will read everything the researcher produced before starting.

You can also let Cerevisor auto-connect. With Smart auto-connect enabled (the default), any agent you add to a new column is automatically wired from the previous column's outputs. Manual wires you draw always override auto-connect.

Run

Click the Run button again. This time Cerevisor will tell you it's running two waves:

Wave 1: Quick Researcher
Wave 2: Blog Writer (after Wave 1)

When Wave 1 finishes, Wave 2 starts automatically. The writer's prompt will include the researcher's output as context.

When the run completes, click the writer card to read the blog post.

What to try next

You've covered the core loop. From here, real workflows usually add a few more things:

• Skills. Drag a skill (e.g. recency-research) from the Skills panel onto an agent for a more capable researcher. See Working with Skills.
• Markdown input/output files. Attach spec.md to the writer as an input file so it has context the researcher didn't generate. See Attach markdown files.
• A reviewer agent. Add a third column with a Reviewer role that critiques the blog post and proposes edits. See Connect agents.
• Conditional connections. Branch based on a confidence score (e.g. "if the researcher's confidence is low, send to a senior researcher; otherwise to the writer"). See Control flow.
• Save your workspace. Ctrl/Cmd+S writes a .cerevisor-world file you can re-open, share, or schedule. See File formats.
• Turn on Progressive Mode. Flip the switch at the top of the chat builder and your next message is built and run, step by step, with the cost reported as it goes. See Progressive Mode.

Where to go next

→ Guides → Building Workflows for the full set of things you can do with agents and connections.