A real drag-and-drop canvas
Drop nodes, drag connections, auto-arrange the graph, and pan with a hand tool. Step numbers, named branches and a minimap keep a real workflow readable instead of a spaghetti diagram.
workflow builder · no-code
Drag, drop, approve — AI workflows that do the work
Build a workflow on a drag-and-drop canvas: a trigger, the tools that do the work, an agent step where judgment is needed, and a human approval gate in front of anything that ships. Then watch every run, step by step.
No code required — and a code step for when you want one. dolv it.
automation that acts
Most builders move data. This one does the work.
A classic automation tool shuttles records between apps. A dolv workflow calls the same 84 tools your Director calls — so a step can draft the post, prepare the outreach, create the deal or book the meeting, grounded in your company context and gated by a human wherever it matters.
Every one of the 79 tools
Draft an email, publish a post, open a task, read your funnel — the Director’s own tools, wired into a graph you control. Effectful steps queue for approval rather than firing on their own.
Approval gates as a node
Drop an approval gate anywhere and the run pauses until a human approves or denies it in Approvals. Human-in-the-loop is a step in the graph, not a setting you hope is on.
Agent steps that think
An agent node is an LLM step with your instructions that can call tools mid-workflow. Use it where a decision needs judgment, not a hard-coded rule.
Branching that follows reality
Edges route on success, fail, approve, deny, true, false or always — so the unhappy path is designed, not an afterthought. Conditions branch on any upstream step’s output.
Point-and-click data references
Type {{ }} or hit slash to pull any earlier step’s output into a later one, with the values from your last run shown inline so you can see what you are actually referencing.
- 6 node types: trigger · tool · agent · approval · condition · code
- 79 callable tools available as workflow nodes
- 20 nodes per workflow, bounded so runs stay predictable
- 30 read + write integrations behind the tool nodes
the palette
Six node types. That's the whole language.
A workflow is a graph of at most twenty nodes, bounded on purpose so runs stay predictable and cheap. Everything you can build is made of these six.
- Trigger Starts the run — manually, or on a schedule.
- Tool One of 79 callable tools, run as a step.
- Agent An LLM step with your instructions; can call tools.
- Approval Pauses until a human approves or denies.
- Condition Routes true/false on an upstream step’s output.
- Code Your own JavaScript, run in an isolated sandbox.
build → gate → run
From an idea to a workflow you can trust
The point of a canvas is that you can see what will happen before it happens — and prove it one step at a time instead of shipping a black box.
- 01
Start from a trigger
Every workflow opens with a trigger — run it by hand, or put it on a schedule and let it fire on its own. The trigger payload is available to every downstream step.
- 02
Wire the work
Drag in tool nodes, agent steps, conditions and code. Connect them along success, fail or true/false edges so both the happy and unhappy paths are explicit.
- 03
Gate what matters
Drop an approval node in front of anything public or irreversible. The run parks there until a person approves or denies it in the Approvals inbox.
- 04
Test one step at a time
Run a read-only step, a condition or a code step on its own, synchronously, against sample upstream data — no waiting for a schedule. Steps that write or send are proved by a real run, never fired off as a test.
- 05
Run it and read the trace
Every run records a step-by-step trace: what fired, what each node returned, where it branched, and where it stopped. Debug from evidence, not guesswork.
describe it, don't draw it
Or just tell the Director to build it
You do not have to start on the canvas. Describe the motion in chat and the Director lays out the graph for you — then lists, updates and runs it on request. Refine the result by dragging nodes, or keep iterating in words.
Workflows export to a portable YAML or JSON file and import back the same way, so a workflow can live in git next to the rest of your stack. Pair it with AI agents, keep it honest with grounded AI, and read the outcome through funnel intelligence.
questions
Workflow builder FAQ
What is the dolv workflow builder?
It is a no-code, drag-and-drop canvas for building AI workflows. You compose a graph out of six node types — a trigger, tool steps, agent steps, approval gates, conditions and code — and dolv runs it. Because the tool palette holds 79 of the Director’s own callable tools, a workflow can read your live data and do real work across your connected stack, not just move data between apps.
Do I need to write code to build a workflow?
No. The whole builder is drag-and-drop: you add nodes from a palette, connect them by dragging edges, and configure each step with a form. A code node exists for the cases where you want to reshape data between steps yourself, but it is entirely optional — most workflows never use one.
Can I build a workflow by just describing it?
Yes. Ask the Director in chat to create a workflow and it will build the graph for you, then list, update or run it on request. It is often fastest to describe the motion in a sentence, let the Director lay out the nodes, and then refine the result on the canvas.
How do human approvals work inside a workflow?
An approval gate is a node you place anywhere in the graph. When the run reaches it, it pauses and queues the item in your Approvals inbox. Approving continues down the approve edge; denying routes down the deny edge, so you can design what happens when a human says no. Nothing public ships while a run is parked at a gate.
What can go wrong, and how do I debug it?
Each run stores a trace of every step — what executed, what it returned, and which branch it took. You can also test read-only steps, conditions and code steps on their own, synchronously, against sample upstream data; steps that write or send are deliberately not test-fireable, so a test can never send something real. Failures route down the fail edge rather than silently ending the run, so you can build a recovery path instead of just an error.
Can I move a workflow between workspaces?
Yes. Workflows export to a portable YAML or JSON file and import back the same way, so you can version a workflow in git, share it with a teammate, or rebuild it in another workspace without clicking it together again.
Build your first workflow today
Start free for 14 days — the full product, no card. Every plan includes the workflow builder; plans differ by tokens, seats and runs, never by features.