Output Steps

For common fields, string substitutions, metadata filters, and execution order, see the Agent Steps Overview.

Display Result Step

Marks a step's output as the final visible result of the agent run. The output of this step is what users see in the UI and what is returned in the API response.

Fields

Field	Type	Required	Default	Description
`template`	string	No	`null`	A template for the output. If null, the step's input is used directly. Supports substitutions.
`html_format`	boolean	No	`true`	When `true`, the output is treated as HTML for rendering in the UI
`attachments`	list	No	`null`	Selects which files to attach to the result — a list of `{{…}}` attachment reference expressions (e.g. `{{attachments}}`, `{{step.gen.attachments[*.png]}}`). When set, the resolved files become the step's files alongside its formatted text — the agent's final result shows generated images / files inline.

Display result steps cannot have child steps.

When attachments resolves to one or more attachments, they are the step's files: its output stays the text, and the run-details UI renders the files below it as it does for prompt-call outputs. With no references, a result step surfaces the files its input carries. On Display Result, references replace that default, so list {{attachments}} alongside them to keep the input's files. Streaming Result's saved output uses its references only while streaming, and not when its parent prompt_call outputs files. See the Attachments guide for a generate-and-display example.

Use Case Examples

Simple pass-through:

No template, html_format: true — Displays the previous step's output as HTML.

Figure 1.Simple pass-through — the prompt call's output flows directly into a display result step with no template transformation.

Formatted output with template:

Template:

<div class="result">
  <h2>Analysis Results</h2>
  {{step.analysis.output}}
  <p><em>Generated on {{date UTC}} by {{agent.name}}</em></p>
</div>

Plain text output:

Template: {{step.final.output}}

html_format: false — Returns plain text without HTML rendering.

Figure 2.Formatted display — an extract data step analyzes content, then the display result step wraps it in an HTML template.

Figure 3.Display Result with governance — the display step shows the final output while a non-blocking governance evaluation audits the content in the background.

Streaming Result Step

Streams LLM output tokens in real-time from a preceding prompt_call to the caller via Server-Sent Events (SSE). This enables chat-like interfaces where users see the response being generated token by token.

Must be a direct child of a prompt_call step. The prompt_call's model must support streaming (supports_streaming capability). Both simple format (plain text) and advanced/JSON template prompt call configurations support streaming.

Additional requirements:

The agent must use a dynamic_input or template_input trigger (the two trigger types with an SSE consumer)
Runs must use priority: true (priority mode)
Only one streaming_result step is allowed per agent definition

Fields

Field	Type	Required	Default	Description
`attachments`	list	No	`null`	Selects which files to attach to the result — a list of `{{…}}` attachment reference expressions (e.g. `{{attachments}}`, `{{step.gen.attachments[*.png]}}`). When set, the resolved files become the step's files alongside the streamed text. SSE token events stream live as text; the files appear on the trace once the stream completes.
`show_reasoning`	bool	No	`false`	Publishes the model's reasoning on the live stream as `stream_reasoning` events while it thinks; only models served through Together AI or Novita can stream their reasoning. Never part of the output; withheld when a governance policy redacts this step. Reasoning can restate the system prompt, retrieved content and tool results. API clients receive it only when the run request sets `include_reasoning`.

Otherwise the step inherits all behavior from its parent prompt_call step.

Streaming result steps cannot have child steps.

Guidance

Place as a direct child of a prompt_call step
The agent trigger must be dynamic_input or template_input
Runs must use priority mode (priority: true)
The model selected in the parent prompt_call must support streaming
For side-effects (save to memory, send email, etc.), add those as sibling branches under the same prompt_call — they receive the full buffered output after streaming completes
To stream and return a schema-validated payload, put an extract_content (with json_schema) → display_result chain in a sibling branch. A display_result takes precedence over the stream whichever finishes first, so its checked output replaces the streamed text as the run's final result — see Streaming plus a schema-validated result
If governance policies with output blocking are active, streaming falls back to non-streaming delivery
For the full streaming setup guide, see Agent Streaming

Use Case Example

Chat-style Q&A agent:

Trigger: dynamic_input (priority: true)
Step 1: Retrieval — Search knowledge base for relevant context
  Step 2: Prompt Call — Answer the user's question using the retrieved context
    Step 3: Streaming Result — Stream the response token-by-token to the user
    Step 4: Add Memory — Save the assistant's reply to conversation history

The streaming result delivers tokens in real-time while the memory step saves the full buffered output after streaming completes.

Figure 4.Streaming RAG chat — retrieval feeds a prompt call, the streaming step delivers tokens live, and memory saves the full reply.

Figure 5.Streaming Result — a retrieval-augmented chat agent streams the LLM response in real-time while saving to memory as a side-effect.

← Back to Agent Steps Overview