Skip to Content
almyty docs — v1
AgentsVisual Pipeline Builder

Visual Pipeline Builder

The pipeline builder is the primary interface for designing workflow agents. It provides a ReactFlow canvas where you drag, connect, and configure nodes to define your agent’s execution graph.

Visual pipeline builder

Interface Overview

The builder has three main areas:

AreaLocationPurpose
Node PaletteLeft panelLists all 10 node types available to drag onto the canvas
CanvasCenterThe DAG workspace where you arrange and connect nodes
Node Config PanelRight panel (opens on node click)Configure the selected node’s properties

At the top of the builder, two tabs switch between Workflow (visual DAG) and Autonomous (form-based) modes.

In the UI

Adding Nodes

  1. Open an agent in the builder
  2. Find the node type you need in the Node Palette on the left
  3. Drag it onto the canvas
  4. The node appears with default configuration and unconnected handles

Connecting Nodes

  1. Hover over a node’s output handle (right side)
  2. Click and drag to another node’s input handle (left side)
  3. Release to create an edge
  4. Data flows left-to-right along edges at execution time

To remove a connection, click the edge and press Delete or Backspace.

Configuring a Node

  1. Click any node on the canvas
  2. The Node Config Panel opens on the right
  3. Fill in the required fields (varies by node type)
  4. Changes are saved to the draft automatically

Saving and Testing

  1. Click Save in the top bar to persist the pipeline
  2. Validation runs automatically — nodes with errors show a red border
  3. Fix any issues listed in the error panel at the bottom
  4. Click Test to open the test drawer, provide sample input, and run the pipeline
  5. Review the output and per-node execution details in the test results

Node Types

The palette contains all 10 node types. Each has a distinct color on the canvas for quick identification.

Node TypePurpose
InputEntry point — defines the JSON Schema for incoming data
OutputExit point — maps the final result from upstream nodes
LLM CallSends a prompt to an LLM provider and returns the response
Tool CallInvokes a tool with explicit parameter mapping
ConditionBranches the pipeline based on a boolean expression (true/false edges)
TransformRuns sandboxed JavaScript to reshape data
LoopIterates over an array, running child nodes for each item
ParallelExecutes multiple branches simultaneously and aggregates results
MergeCombines outputs from multiple upstream nodes into a single value
Sub-AgentInvokes another agent as a step in this pipeline

See Node Types for full configuration details on each.

Workflow vs Autonomous Mode

Workflow Mode

The visual DAG shown above. You explicitly define every step, connection, and data transformation. Best for deterministic, repeatable pipelines where you need full control over execution order.

Autonomous Mode

Switching to the Autonomous tab replaces the canvas with a form:

  • System Instructions — free-text prompt describing the agent’s behavior
  • Tools — select which tools the agent can call
  • Memory — toggle persistent memory across conversations
  • Collaboration — optionally connect other agents for delegation

The LLM decides at runtime which tools to call and in what order. Best for open-ended tasks where rigid sequencing is unnecessary.

Canvas Controls

ActionShortcut
PanClick and drag on empty canvas
ZoomScroll wheel or pinch
Select multipleShift + click, or drag a selection box
Delete selectedDelete or Backspace
UndoCtrl/Cmd + Z
RedoCtrl/Cmd + Shift + Z

Via the API

The pipeline structure set in the builder maps directly to the API’s pipeline object:

curl -X PATCH /agents/{id} \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "pipeline": {
      "nodes": [
        { "id": "input_1", "type": "input", "data": { "schema": { "type": "object", "properties": { "query": { "type": "string" } }, "required": ["query"] } } },
        { "id": "llm_1", "type": "llm_call", "data": { "providerId": "provider-uuid", "userPromptTemplate": "Answer: {{input.query}}" } },
        { "id": "output_1", "type": "output", "data": { "mapping": "{{nodes.llm_1.output}}" } }
      ],
      "edges": [
        { "id": "e1", "source": "input_1", "target": "llm_1" },
        { "id": "e2", "source": "llm_1", "target": "output_1" }
      ]
    }
  }'

Node positions on the canvas are stored in each node’s position field ({ x, y }) and are included in export/import payloads.

AgentsNode Types