Types

Provider-agnostic wire and history types shared across the framework.

Run items

• augments.adk.types.RunItem
• augments.adk.types.RunItemBase
• augments.adk.types.UserItem
• augments.adk.types.SystemItem
• augments.adk.types.MessageOutputItem
• augments.adk.types.ReasoningItem
• augments.adk.types.ToolCallItem
• augments.adk.types.ToolCallOutputItem
• augments.adk.types.ToolApprovalItem
• augments.adk.types.ToolSearchCallItem
• augments.adk.types.ToolSearchOutputItem
• augments.adk.types.HandoffCallItem
• augments.adk.types.HandoffOutputItem
• augments.adk.types.MCPListToolsItem
• augments.adk.types.MCPApprovalRequestItem
• augments.adk.types.MCPApprovalResponseItem
• augments.adk.types.CompactionItem
• augments.adk.types.ItemHelpers

Result

augments.adk.types.RunResult

Result of a completed (or interrupted) agent run. Contains the final output (if the run completed), all items generated during execution, and supports HITL interruptions via deferred_requests.

Fields:

• final_output — the final output from the agent, or None if interrupted for approval.
• user_prompt — the original user prompt passed to the run.
• new_items — Layer 3 RunItem values generated during this run (messages, tool calls, results).
• context — the run context with usage tracking.
• last_agent — the last agent that was active.
• recovered — True when an error handler produced final_output after the run raised; recovered runs skip session and memory persistence.
• deferred_requests — tools captured for approval or external execution; None if the run completed.
• state — serializable state for resuming interrupted runs.
• guardrail_results — per-phase agent-level guardrail audit trail (input and output slots).
• guardrail_audit — per-action guardrail audit records across every level (agent, tool, flow), captured as hashes, never raw payloads.
• swarm_yield — set only by the swarm driver when an agent turn yielded control; None on every plain Runner.arun() path.
• sandbox_usage — aggregate sandbox resource and cost usage, or None when no sandbox session ran.

Members:

• requires_action — property; True when human approval or external action is pending.
• interruptions — property; tool calls awaiting human approval, as a flat list.
• last_response_id — property; the response_id of the most recent LLM response in this run, or None.
• release_agents(*, release_new_items=True) — drop strong references to agents and, optionally, run items.
• to_input_list() — convert to a Layer 1 input list for a continued conversation.
• final_output_as(output_type) — cast the final output to the expected type.

Example:

result = await Runner.arun(agent, "Delete user 123")
if result.requires_action:
    for req in list(result.deferred_requests.approvals):
        if await confirm(f"Approve {req.tool_name}?"):
            result.state.approve(req)
        else:
            result.state.reject(req, "Denied")
    result = await Runner.arun(agent, result.state)

Built-in tool call and result types

• augments.adk.types.WebSearchToolCall
• augments.adk.types.WebSearchToolCallResult
• augments.adk.types.WebSearchResult
• augments.adk.types.FileSearchToolCall
• augments.adk.types.FileSearchToolCallResult
• augments.adk.types.FileSearchResult
• augments.adk.types.CodeInterpreterToolCall
• augments.adk.types.CodeInterpreterToolCallResult
• augments.adk.types.CodeInterpreterOutput
• augments.adk.types.ComputerToolCall
• augments.adk.types.ComputerToolCallResult
• augments.adk.types.ComputerAction
• augments.adk.types.ImageGenerationToolCall
• augments.adk.types.ImageGenerationToolCallResult
• augments.adk.types.ShellToolCall
• augments.adk.types.ShellToolCallResult
• augments.adk.types.ApplyPatchToolCall
• augments.adk.types.ApplyPatchToolCallResult
• augments.adk.types.ToolSearchToolCall
• augments.adk.types.ToolSearchToolCallResult
• augments.adk.types.ToolSearchResultEntry
• augments.adk.types.MCPListTools
• augments.adk.types.MCPListToolsTool
• augments.adk.types.MCPCall
• augments.adk.types.MCPCallResult
• augments.adk.types.MCPApprovalRequest
• augments.adk.types.MCPApprovalResponse

Tracing span data

• augments.adk.types.SpanData
• augments.adk.types.AgentSpanData
• augments.adk.types.FunctionSpanData
• augments.adk.types.GenerationSpanData
• augments.adk.types.GuardrailSpanData
• augments.adk.types.HandoffSpanData
• augments.adk.types.ResponseSpanData
• augments.adk.types.CustomSpanData
• augments.adk.types.AnySpanData

How the type layers fit together is explained in the Types guide.