Your First Workflow

Sign in and complete the onboarding wizard

The first time you open RondoFlow, a short three-step wizard appears. You can’t dismiss step 1 — a working directory is required for RondoFlow to function.

Step 1 — Where does your code live?

RondoFlow operates on your local files through Claude Code, so it needs a project folder.

• The wizard auto-detects a likely directory and pre-fills it. You can edit the path or click Browse to pick a folder.
• It also runs a health check: you’ll see Claude Code detected in green, or a warning to install the CLI with npm install -g @anthropic-ai/claude-code.
• If no Claude credential is found yet, you’ll get a non-blocking note. Set ANTHROPIC_API_KEY, or run claude setup-token and set CLAUDE_CODE_OAUTH_TOKEN to use your Claude subscription. You can also add it later in Settings.

The working directory can be changed any time in Workspace settings.

Step 2 — How do you want to work?

Pick a work mode:

Step 3 — What would you like to do first?

You’ll land on a final screen with a few options:

• Describe a task (the primary input) — type what you want and click Go. This is the path we follow below.
• Start from Template — open the gallery of pre-built Workflows.
• Create an Assistant — build one from scratch.
• Skip — go straight to an empty Workspace.

Describe a task and let RondoFlow generate a Workflow

Type a clear, specific goal — for example, Build a REST API with tests and code review. Your description must be at least 10 characters (and at most 1000).

RondoFlow sends it to its Workflow Architect, which analyzes the task and designs a complete team:

• 2–5 specialized Assistants, each with a focused role — no generic “helper” agents.
• A detailed Personality (persona) per Assistant describing its expertise, approach, and responsibilities.
• A purpose (writing, coding, analysis, chat, review, research, creative, data, or general).
• A recommended model — opus for deep analysis and complex reasoning, sonnet as the strong default for coding and writing, haiku for fast, simple tasks.
• Recommended skills drawn from the built-in catalog.
• Edges that define execution order as a DAG (no cycles), so output flows from one Assistant to the next. An edge may carry an optional regex condition that gates the next step.
• The Assistants are auto-laid out left-to-right by dependency layer.

Generated teams are always Claude Code Assistants — the Workflow Architect designs and runs against Claude. RondoFlow also supports OpenAI and Perplexity Assistants, but those aren’t part of generation: you add them yourself by dragging them from the palette onto the Workspace. See Providers.

The Director (orchestrator) is auto-enabled when the generated team has 3 or more Assistants or needs intelligent coordination, and left off for simple two-Assistant linear flows.

Review the generated team

Generation opens a review dialog before anything touches your Workspace. Here you can fine-tune the proposal:

• Edit the Workflow name.
• Edit each Assistant’s name, Personality, and model.
• Remove any suggested skill from an Assistant.
• Remove an Assistant entirely (its edges are removed too).
• Toggle Director orchestration on or off.

When you confirm, RondoFlow materializes the team onto the Workspace: it persists the Assistants and lays out their Cards (nodes) and connections so you can see the whole flow at a glance. From here you can keep editing on the canvas — see The Canvas for moving Cards, wiring connections, and adding skills or Safety Rules.

Press Run and watch live execution

When a Workflow is on the Workspace, a toolbar appears at the top center of the canvas.

• Click the green Run button. This opens the Execution Log, where you describe the task for this run and start it. (Running is triggered from inside that panel, not by the Run button alone.)
• Each Assistant then starts in dependency order, passing its output to the next.
• A step indicator below the toolbar shows progress like 2/4 with the current Assistant’s name.
• Each Card reflects its live state during the run:

The toolbar carries three orchestration controls beside Run/Stop:

If the Director is on, after each step it reviews the output and decides whether to continue, redirect (re-run a step with new guidance), or conclude early — shown as an inline badge in the toolbar. To stop everything immediately, click Stop, which terminates every active Assistant. See Planner, Director & Advisor for how all three work together.

Inspect the output and costs

Open the Workflow Chat toggle (the message icon) in the toolbar to watch Assistants run live, read their full raw output, and send follow-up messages to the Workflow.

For spend, RondoFlow tracks token usage and cost per run. The cost dashboard (GET /api/analytics/cost?days=30) summarizes total cost in USD, the number of conversations (sessions), and a per-model breakdown of input/output tokens and cost — plus a gap-filled per-day time series and a per-workspace rollup that power the Analytics view. If you haven’t run anything yet, it shows “No usage data yet.” See Monitoring for the full view.