Skip to main content

Assistant Workflow Editor

Configure the LLM pipeline that powers your assistant's responses.

What is the embedded workflow?

Every assistant has its own private embedded workflow — an LLM pipeline that processes incoming chat messages and generates responses. This workflow:

  • Is created from within the assistant editor (not the Workflows section)
  • Does not appear in the Workflows dashboard (which only lists standalone task workflows)
  • Is the engine behind every chat conversation the assistant has

Getting to the editor

  1. Open an assistant at /assistants-config/{id}
  2. In the workflow section, click + Add workflow (if no workflow exists) — this creates one automatically
  3. Or click an existing workflow card to open it

The editor opens at /assistants-config/{id}/workflow/{workflowId}.


Canvas overview

Embedded workflow editor showing Input, Agent, and Output fixed nodes on the canvas

The canvas has three fixed nodes that cannot be deleted:

NodeRole
InputReceives the user's chat message
AgentProcesses the message using the assigned LLM and any attached sources or tools
OutputReturns the agent's response to the chat

The left panel contains draggable items you can attach to the Agent node.


Configuring the Agent node

Assign an LLM (required to save)

Drag a Chat Setting from the LLMs section in the left panel and drop it onto the LLM drop area on the Agent node. The assigned setting's name appears in the drop area.

To remove it, click the trash icon next to the name. Note: you must always have an LLM assigned — saving without one is blocked with an error toast: "Please drag a chat setting inside Agent before saving the workflow."

Attach sources

Drag a Source from the left panel onto the Agent node to enable document retrieval (RAG) for this assistant. Attached sources are listed on the node.

Attach tools

Drag a Tool (action) from the left panel onto the Agent node. Available tools include:

ToolWhat it does
Get Answers in a FileSaves the AI response as a downloadable file (docx, pdf, xlsx, csv, txt)
Execute Python CodeRuns Python snippets for calculations or data tasks
Get Interactive Visuals in AnswersGenerates interactive charts alongside responses

Set the mode

Use the Mode selector on the Agent node to switch between:

ModeDescription
AgentSingle LLM processes each message — default and recommended for most assistants
OrchestratorRoutes queries to specialized sub-agents, each with their own tools and sources
note

Orchestrator mode is available in the UI but requires sub-agents to be configured before saving. Saving in Orchestrator mode without sub-agents will return an error. Use Agent mode for standard assistants.


Toolbar actions

ActionBehaviour
Save workflowPersists all canvas changes. Blocked if no LLM is assigned.
CancelReturns to the assistant editor without saving. An "Unsaved changes" dialog appears if there are unsaved changes.
DeletePermanently deletes the embedded workflow. A confirmation modal appears. After deletion, you return to the assistant editor with the "+ Add workflow" placeholder.
ExportDownloads the workflow as a .json file (named workflow-{name}.json)
Import JSON fileUploads a .json to replace the workflow structure
Full ScreenExpands the canvas to fill the browser window
New ParameterOpens the "Add New Variable" popup to define custom input parameters

Input parameters

Input parameters let you define variables that users fill in before a workflow run. In chat-based assistants, parameters are referenced in the workflow's prompt template using {{parameter name}} syntax.

Add a parameter

Click New Parameter in the toolbar. The "Add New Variable" popup opens.

Add New Variable popup showing Name, Is Required, Input Type, Placeholder, Value, and Description fields
FieldRequiredNotes
NameYesUsed as the variable name in prompt templates — no spaces
Is RequiredNoTrue / False radio (default: False). Stored in the workflow but not visually indicated on the parameter chip.
Input TypeNoControls the input control shown to users (see types below)
PlaceholderNoHint text shown inside the input
ValueNoDefault value pre-filled for users
DescriptionNoTooltip or helper text

Click Create to add the parameter. A chip with the parameter name appears in the Input node's parameters section.

Input types

TypeOptions field shownInput control
StringNoSingle-line text
EmailNoEmail input
PasswordNoPassword (masked) input
NumberNoNumeric input
BooleanNoTrue / False toggle
TextareaNoMulti-line text
SelectYesDropdown with your defined options
RadioYesRadio button group
CheckboxYesCheckbox group

For Select, Radio, and Checkbox types, an Add Option field appears in the popup — type an option label and click Add to create a chip. Options can be removed individually.

Edit or delete a parameter

  • Edit — click the pencil icon on the parameter chip. The "Edit Variable" popup opens, pre-filled with current values.
  • Delete — click the trash icon on the parameter chip to remove it.

Save and navigation

After configuring the canvas:

  1. Click Save workflow — a success toast confirms the save
  2. Click Cancel to return to the assistant editor

Changes do not take effect in chat until the workflow is saved.