Reasoning Steps

Reka Research gives you full visibility into how responses are generated.

Each call includes a reasoning trace in message.reasoning_steps — showing every step the agent takes: what it thinks, which tools it uses, and what it returns. The the agent may sometimes encounter errors during tool use which will be reflected in the reasoning trace.

This makes it easy to debug, explain, and audit how your AI works — essential for building apps that users and stakeholders can trust.

Reasoning Step

Each reasoning step shows how the agent decides what to do — and why.

role
string

The author of the reasoning step — either the assistant or a tool (for tool execution results).

reasoning_content
string

The agent’s internal reasoning — a natural language explanation of its intent before acting.

tool_calls
array

A list of tool calls made by the agent during this reasoning step.

content
object

The result from the execution of the tool call, if applicable. See Tool Output for details.

tool_call_id
string

The unique identifier of the tool call that produced this reasoning step. This links the step to a prior tool call.

Tool Output

Tool steps return structured output with metadata about the tool run and its result.

tool_name
string

The name of the tool executed. One of search_web or analyze.

tool_output
array

The result returned by the tool. The format depends on the tool used and may include web search results or analysis result.

tool_error
string

The error message from the tool execution, if applicable.

Web search result

Represents a search result from the search_web tool.

url
string

The URL of the webpage.

title
string

The title of the webpage.

snippet
string

A brief snippet or summary of the webpage content.

age
string

A string representing the age of the web search result (e.g. ‘2h ago’). Optional.

Analysis result

Represents the result from the analyze tool.

url
string

The URL of the webpage.

analysis
string

The analysis of the webpage content.