The Chat block sends a prompt to the workspace assistant and returns its answer to the next block. Use it for tasks that need workspace context, memory, or several tool calls, such as finding information across connected services and summarizing the results.
Chat chooses its model using the workspace's available models and credentials, and decides which tools to call. Use the Agent block when you want to configure a model, messages, and tools directly on the block.
Prompt
Enter the task in Prompt. Include the information the assistant needs and the result you want. You can reference upstream outputs, for example:
Summarize the following results and identify any items that need attention:
<api.data>Prompt is required. Each block execution starts a new assistant turn; previous workflow runs are not automatically replayed as conversation history. The assistant can use the workspace tools and memory available to the user running the workflow.
Outputs
For a block named Chat, downstream blocks can reference:
| Reference | Value |
|---|---|
<chat.content> | The assistant's response text |
<chat.model> | The model used for the turn, or workspace-assistant when no model ID is reported |
<chat.tokens> | Token usage with prompt, completion, and total fields, when reported |
<chat.cost> | Model cost with input, output, and total fields, when reported |
<chat.toolCalls.count> | Number of tool calls made during the turn |
<chat.toolCalls.list> | Tool calls with their names, arguments, results, errors, and durations |
Model cost uses the resolved model's rates, including cached input discounts and long-context pricing for each model call. Hosted model usage includes the configured model-cost multiplier. With your own provider key (BYOK), SteelEngine's model cost is zero, while token usage is still reported; the provider bills that usage directly. The workflow's execution charge is separate. See Cost Calculation.
Example Workflow
Connect Start → Chat → Response. Give Chat a prompt, then set the Response block's JSON data to <chat.content> to return the answer.
The workflow waits for Chat to finish before continuing. If the assistant turn fails, Chat fails the step and the workflow follows its configured error handling. Cancelling the workflow also cancels the active assistant turn.
Chat checks the executing user's workspace access and uses the same model access rules as workspace Chat. Tools remain subject to workspace permissions and their confirmation requirements.
For self-hosted installations, deploy the Chat runtime fix to the workflow workers as well as the web app. A worker without the assistant runtime reports Workspace assistant port is not configured when it reaches this block. See Temporal execution for worker setup.