---
title: "PUT /rerun-call-analysis/{call_id}"
method: PUT
path: "/rerun-call-analysis/{call_id}"
---

# PUT /rerun-call-analysis/{call_id}

`PUT /rerun-call-analysis/{call_id}`

## Request

- Base URL: `https://api.retellai.com`
- URL: `https://api.retellai.com/rerun-call-analysis/{call_id}`
- Auth: HTTP bearer

Rerun post-call analysis for a specific call. This operation incurs charges.

## Path parameters

- `call_id` string, required

## Response `201`

Call analysis rerun successfully.

- union
  - V2WebCallResponse
    - `call_type` 'web_call', required: Type of the call. Used to distinguish between web call and phone call.
    - `access_token` string, required: Access token to enter the web call room. This needs to be passed to your frontend to join the call.
    - `call_id` string, required: Unique id of the call. Used to identify the call in the LLM websocket and used to authenticate in the audio websocket.
    - `agent_id` string, required: Corresponding agent id of this call.
    - `agent_name` string: Name of the agent.
    - `agent_version` integer, required: The version of the agent.
    - `agent_tag` string, nullable: Tag pointing at the agent version used for this call, captured at call creation time and frozen thereafter (unaffected by later tag reassignments). Populated whether the caller dispatched by tag, numeric version, "latest", or "latest_published" — when the caller specified a tag, that tag wins; otherwise the most-recently- assigned tag on the resolved version is used. Absent when no tag points at the resolved version (or for calls created before this field was introduced).
    - `call_status` 'registered' | 'not_connected' | 'ongoing' | 'ended' | 'error', required: Status of call. - `registered`: Call id issued, starting to make a call using this id. - `ongoing`: Call connected and ongoing. - `ended`: The underlying websocket has ended for the call. Either user or agent hung up, or call transferred. - `error`: Call encountered error.
    - `metadata` object: An arbitrary object for storage purpose only. You can put anything here like your internal customer id associated with the call. Not used for processing. You can later get this field from the call object.
    - `retell_llm_dynamic_variables` object: Add optional dynamic variables in key value pairs of string that injects into your Response Engine prompt and tool description. Only applicable for Response Engine.
    - `collected_dynamic_variables` object: Dynamic variables collected from the call. Only available after the call ends.
    - `custom_sip_headers` object: Custom SIP headers to be added to the call.
    - `data_storage_setting` 'everything' | 'everything_except_pii' | 'basic_attributes_only', nullable: Data storage setting for this call's agent. "everything" stores all data, "everything_except_pii" excludes PII when possible, "basic_attributes_only" stores only metadata.
    - `opt_in_signed_url` boolean: Whether this agent opts in for signed URLs for public logs and recordings. When enabled, the generated URLs will include security signatures that restrict access and automatically expire after 24 hours.
    - `start_timestamp` integer: Begin timestamp (milliseconds since epoch) of the call. Available after call starts.
    - `end_timestamp` integer: End timestamp (milliseconds since epoch) of the call. Available after call ends.
    - `transfer_end_timestamp` integer: Transfer end timestamp (milliseconds since epoch) of the call. Available after transfer call ends.
    - `duration_ms` integer: Duration of the call in milliseconds. Available after call ends.
    - `transcript` string: Transcription of the call. Available after call ends.
    - `transcript_object` Utterance[]: Transcript of the call in the format of a list of utterance, with timestamp. Available after call ends.
      - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
      - `content` string, required: Transcript of the utterances.
      - `metadata` object
        - `response_id` integer: Call-scoped agent response identifier, when available.
        - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
        - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
        - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
          - `model` string, required: Actual selected model after routing and retries.
          - `primary_model` string, required: Primary model configured for this request.
          - `backup_model` string, required: Backup model configured for this request.
          - `reason` 'primary_failed' | 'preference_expired' | 'first_usable', required: Why the backup won. primary_failed means primary attempts failed; preference_expired means backup was ready within the primary preference window; first_usable means backup became ready after that window and won the race.
          - `selection_latency_ms` number, required: Elapsed milliseconds from race start to winner selection.
          - `primary_ready_ms` number: Elapsed milliseconds until primary produced usable output. Omitted if primary was not ready when the winner was selected.
          - `backup_ready_ms` number: Elapsed milliseconds until backup produced usable output. Omitted when unavailable.
      - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
        - `word` string: Word transcript (with punctuation if applicable).
        - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
        - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
    - `transcript_with_tool_calls` UtteranceOrToolCall[]: Transcript of the call weaved with tool call invocation and results. It precisely captures when (at what utterance, which word) the tool was invoked and what was the result. Available after call ends.
      - union
        - Utterance
          - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
          - `content` string, required: Transcript of the utterances.
          - `metadata` object
            - `response_id` integer: Call-scoped agent response identifier, when available.
            - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
            - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
            - `word` string: Word transcript (with punctuation if applicable).
            - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
            - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
        - ToolCallInvocationUtterance
          - `role` 'tool_call_invocation', required: This is a tool call invocation.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `name` string, required: Name of the tool. This is the configured tool name, or the node name for a built-in conversation flow node, so it can contain spaces and non-ASCII characters. To identify the kind of tool, use `type`.
          - `type` string: Kind of tool. Known values are `custom`, `code`, `mcp`, `integration_app`, `end_call`, `transfer_call`, `bridge_transfer`, `cancel_transfer`, `agent_swap`, `press_digit`, `send_sms`, `extract_dynamic_variable`, `adjust_voice_speed`, `book_appointment_cal`, and `check_availability_cal`. New values can be added. Absent on tool calls sent by a custom LLM, and on some older calls and chats.
          - `arguments` string, required: Arguments for this tool call, it's a stringified JSON object.
          - `metadata` object
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `thought_signature` string: Optional thought signature from Google Gemini thinking models. This is used internally to maintain reasoning chain in multi-turn function calling.
        - ToolCallResultUtterance
          - `role` 'tool_call_result', required: This is the result of a tool call.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `content` string, required: Result of the tool call, can be a string, a stringified json, etc.
          - `successful` boolean: Whether the tool call was successful.
        - NodeTransitionUtterance
          - `role` 'node_transition', required: This is result of a node transition
          - `former_node_id` string, required: Former node id
          - `former_node_name` string, required: Former node name
          - `new_node_id` string, required: New node id
          - `new_node_name` string, required: New node name
          - `transition_type` 'global' | 'global_go_back' | 'interrupt_go_back' | 'normal': How this node was reached. "global" means a global node transition, "global_go_back" means returning from a global node, "interrupt_go_back" means going back due to user interruption, and "normal" means a regular edge transition.
        - DTMFUtterance
          - `role` 'dtmf', required: Digit pressed by the user from their phone keypad.
          - `digit` string, required: The digit pressed by the user. Will be a single digit string like "1", "2", "3", "*", "#" etc.
        - SmsUtterance
          - `role` 'sms', required: SMS message received from the user during the call (for example while the agent is leaving a voicemail). Not part of the spoken conversation.
          - `content` string, required: Text content of the SMS message.
          - `multimedia` SmsMultimediaItem[]: Multimedia attachments (MMS). Display only; not relayed into the spoken conversation.
            - `url` string, required: URL of the multimedia attachment.
            - `summary` string: Optional textual summary of the attachment.
          - `time_sec` number, required: Time the SMS was received, in seconds relative to the start of the call.
        - InjectedUtterance
          - `role` 'injected', required: External context injected into the conversation via the update-live-call API. Not spoken by either party.
          - `content` string, required: The injected context text.
          - `time_sec` number, required: Time the context was injected, in seconds relative to the start of the call.
    - `scrubbed_transcript_with_tool_calls` UtteranceOrToolCall[]: Transcript of the call weaved with tool call invocation and results, without PII. It precisely captures when (at what utterance, which word) the tool was invoked and what was the result. Available after call ends.
      - union
        - Utterance
          - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
          - `content` string, required: Transcript of the utterances.
          - `metadata` object
            - `response_id` integer: Call-scoped agent response identifier, when available.
            - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
            - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
            - `word` string: Word transcript (with punctuation if applicable).
            - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
            - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
        - ToolCallInvocationUtterance
          - `role` 'tool_call_invocation', required: This is a tool call invocation.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `name` string, required: Name of the tool. This is the configured tool name, or the node name for a built-in conversation flow node, so it can contain spaces and non-ASCII characters. To identify the kind of tool, use `type`.
          - `type` string: Kind of tool. Known values are `custom`, `code`, `mcp`, `integration_app`, `end_call`, `transfer_call`, `bridge_transfer`, `cancel_transfer`, `agent_swap`, `press_digit`, `send_sms`, `extract_dynamic_variable`, `adjust_voice_speed`, `book_appointment_cal`, and `check_availability_cal`. New values can be added. Absent on tool calls sent by a custom LLM, and on some older calls and chats.
          - `arguments` string, required: Arguments for this tool call, it's a stringified JSON object.
          - `metadata` object
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `thought_signature` string: Optional thought signature from Google Gemini thinking models. This is used internally to maintain reasoning chain in multi-turn function calling.
        - ToolCallResultUtterance
          - `role` 'tool_call_result', required: This is the result of a tool call.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `content` string, required: Result of the tool call, can be a string, a stringified json, etc.
          - `successful` boolean: Whether the tool call was successful.
        - NodeTransitionUtterance
          - `role` 'node_transition', required: This is result of a node transition
          - `former_node_id` string, required: Former node id
          - `former_node_name` string, required: Former node name
          - `new_node_id` string, required: New node id
          - `new_node_name` string, required: New node name
          - `transition_type` 'global' | 'global_go_back' | 'interrupt_go_back' | 'normal': How this node was reached. "global" means a global node transition, "global_go_back" means returning from a global node, "interrupt_go_back" means going back due to user interruption, and "normal" means a regular edge transition.
        - DTMFUtterance
          - `role` 'dtmf', required: Digit pressed by the user from their phone keypad.
          - `digit` string, required: The digit pressed by the user. Will be a single digit string like "1", "2", "3", "*", "#" etc.
        - SmsUtterance
          - `role` 'sms', required: SMS message received from the user during the call (for example while the agent is leaving a voicemail). Not part of the spoken conversation.
          - `content` string, required: Text content of the SMS message.
          - `multimedia` SmsMultimediaItem[]: Multimedia attachments (MMS). Display only; not relayed into the spoken conversation.
            - `url` string, required: URL of the multimedia attachment.
            - `summary` string: Optional textual summary of the attachment.
          - `time_sec` number, required: Time the SMS was received, in seconds relative to the start of the call.
        - InjectedUtterance
          - `role` 'injected', required: External context injected into the conversation via the update-live-call API. Not spoken by either party.
          - `content` string, required: The injected context text.
          - `time_sec` number, required: Time the context was injected, in seconds relative to the start of the call.
    - `pre_session_transcript_with_tool_calls` UtteranceOrToolCall[]: Tool call invocations and results of integration tools run before the session started (pre-session). Stored separately from the main transcript. Available after call ends if pre-session integration tools ran.
      - union
        - Utterance
          - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
          - `content` string, required: Transcript of the utterances.
          - `metadata` object
            - `response_id` integer: Call-scoped agent response identifier, when available.
            - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
            - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
            - `word` string: Word transcript (with punctuation if applicable).
            - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
            - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
        - ToolCallInvocationUtterance
          - `role` 'tool_call_invocation', required: This is a tool call invocation.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `name` string, required: Name of the tool. This is the configured tool name, or the node name for a built-in conversation flow node, so it can contain spaces and non-ASCII characters. To identify the kind of tool, use `type`.
          - `type` string: Kind of tool. Known values are `custom`, `code`, `mcp`, `integration_app`, `end_call`, `transfer_call`, `bridge_transfer`, `cancel_transfer`, `agent_swap`, `press_digit`, `send_sms`, `extract_dynamic_variable`, `adjust_voice_speed`, `book_appointment_cal`, and `check_availability_cal`. New values can be added. Absent on tool calls sent by a custom LLM, and on some older calls and chats.
          - `arguments` string, required: Arguments for this tool call, it's a stringified JSON object.
          - `metadata` object
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `thought_signature` string: Optional thought signature from Google Gemini thinking models. This is used internally to maintain reasoning chain in multi-turn function calling.
        - ToolCallResultUtterance
          - `role` 'tool_call_result', required: This is the result of a tool call.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `content` string, required: Result of the tool call, can be a string, a stringified json, etc.
          - `successful` boolean: Whether the tool call was successful.
        - NodeTransitionUtterance
          - `role` 'node_transition', required: This is result of a node transition
          - `former_node_id` string, required: Former node id
          - `former_node_name` string, required: Former node name
          - `new_node_id` string, required: New node id
          - `new_node_name` string, required: New node name
          - `transition_type` 'global' | 'global_go_back' | 'interrupt_go_back' | 'normal': How this node was reached. "global" means a global node transition, "global_go_back" means returning from a global node, "interrupt_go_back" means going back due to user interruption, and "normal" means a regular edge transition.
        - DTMFUtterance
          - `role` 'dtmf', required: Digit pressed by the user from their phone keypad.
          - `digit` string, required: The digit pressed by the user. Will be a single digit string like "1", "2", "3", "*", "#" etc.
        - SmsUtterance
          - `role` 'sms', required: SMS message received from the user during the call (for example while the agent is leaving a voicemail). Not part of the spoken conversation.
          - `content` string, required: Text content of the SMS message.
          - `multimedia` SmsMultimediaItem[]: Multimedia attachments (MMS). Display only; not relayed into the spoken conversation.
            - `url` string, required: URL of the multimedia attachment.
            - `summary` string: Optional textual summary of the attachment.
          - `time_sec` number, required: Time the SMS was received, in seconds relative to the start of the call.
        - InjectedUtterance
          - `role` 'injected', required: External context injected into the conversation via the update-live-call API. Not spoken by either party.
          - `content` string, required: The injected context text.
          - `time_sec` number, required: Time the context was injected, in seconds relative to the start of the call.
    - `scrubbed_pre_session_transcript_with_tool_calls` UtteranceOrToolCall[]: Pre-session integration tool call invocations and results, without PII. Available after call ends if pre-session integration tools ran and PII scrubbing is enabled.
      - union
        - Utterance
          - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
          - `content` string, required: Transcript of the utterances.
          - `metadata` object
            - `response_id` integer: Call-scoped agent response identifier, when available.
            - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
            - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
            - `word` string: Word transcript (with punctuation if applicable).
            - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
            - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
        - ToolCallInvocationUtterance
          - `role` 'tool_call_invocation', required: This is a tool call invocation.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `name` string, required: Name of the tool. This is the configured tool name, or the node name for a built-in conversation flow node, so it can contain spaces and non-ASCII characters. To identify the kind of tool, use `type`.
          - `type` string: Kind of tool. Known values are `custom`, `code`, `mcp`, `integration_app`, `end_call`, `transfer_call`, `bridge_transfer`, `cancel_transfer`, `agent_swap`, `press_digit`, `send_sms`, `extract_dynamic_variable`, `adjust_voice_speed`, `book_appointment_cal`, and `check_availability_cal`. New values can be added. Absent on tool calls sent by a custom LLM, and on some older calls and chats.
          - `arguments` string, required: Arguments for this tool call, it's a stringified JSON object.
          - `metadata` object
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `thought_signature` string: Optional thought signature from Google Gemini thinking models. This is used internally to maintain reasoning chain in multi-turn function calling.
        - ToolCallResultUtterance
          - `role` 'tool_call_result', required: This is the result of a tool call.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `content` string, required: Result of the tool call, can be a string, a stringified json, etc.
          - `successful` boolean: Whether the tool call was successful.
        - NodeTransitionUtterance
          - `role` 'node_transition', required: This is result of a node transition
          - `former_node_id` string, required: Former node id
          - `former_node_name` string, required: Former node name
          - `new_node_id` string, required: New node id
          - `new_node_name` string, required: New node name
          - `transition_type` 'global' | 'global_go_back' | 'interrupt_go_back' | 'normal': How this node was reached. "global" means a global node transition, "global_go_back" means returning from a global node, "interrupt_go_back" means going back due to user interruption, and "normal" means a regular edge transition.
        - DTMFUtterance
          - `role` 'dtmf', required: Digit pressed by the user from their phone keypad.
          - `digit` string, required: The digit pressed by the user. Will be a single digit string like "1", "2", "3", "*", "#" etc.
        - SmsUtterance
          - `role` 'sms', required: SMS message received from the user during the call (for example while the agent is leaving a voicemail). Not part of the spoken conversation.
          - `content` string, required: Text content of the SMS message.
          - `multimedia` SmsMultimediaItem[]: Multimedia attachments (MMS). Display only; not relayed into the spoken conversation.
            - `url` string, required: URL of the multimedia attachment.
            - `summary` string: Optional textual summary of the attachment.
          - `time_sec` number, required: Time the SMS was received, in seconds relative to the start of the call.
        - InjectedUtterance
          - `role` 'injected', required: External context injected into the conversation via the update-live-call API. Not spoken by either party.
          - `content` string, required: The injected context text.
          - `time_sec` number, required: Time the context was injected, in seconds relative to the start of the call.
    - `post_session_transcript_with_tool_calls` UtteranceOrToolCall[]: Tool call invocations and results of integration tools run after post-call analysis (post-session). Stored separately from the main transcript. Available after call ends if post-session integration tools ran.
      - union
        - Utterance
          - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
          - `content` string, required: Transcript of the utterances.
          - `metadata` object
            - `response_id` integer: Call-scoped agent response identifier, when available.
            - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
            - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
            - `word` string: Word transcript (with punctuation if applicable).
            - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
            - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
        - ToolCallInvocationUtterance
          - `role` 'tool_call_invocation', required: This is a tool call invocation.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `name` string, required: Name of the tool. This is the configured tool name, or the node name for a built-in conversation flow node, so it can contain spaces and non-ASCII characters. To identify the kind of tool, use `type`.
          - `type` string: Kind of tool. Known values are `custom`, `code`, `mcp`, `integration_app`, `end_call`, `transfer_call`, `bridge_transfer`, `cancel_transfer`, `agent_swap`, `press_digit`, `send_sms`, `extract_dynamic_variable`, `adjust_voice_speed`, `book_appointment_cal`, and `check_availability_cal`. New values can be added. Absent on tool calls sent by a custom LLM, and on some older calls and chats.
          - `arguments` string, required: Arguments for this tool call, it's a stringified JSON object.
          - `metadata` object
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `thought_signature` string: Optional thought signature from Google Gemini thinking models. This is used internally to maintain reasoning chain in multi-turn function calling.
        - ToolCallResultUtterance
          - `role` 'tool_call_result', required: This is the result of a tool call.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `content` string, required: Result of the tool call, can be a string, a stringified json, etc.
          - `successful` boolean: Whether the tool call was successful.
        - NodeTransitionUtterance
          - `role` 'node_transition', required: This is result of a node transition
          - `former_node_id` string, required: Former node id
          - `former_node_name` string, required: Former node name
          - `new_node_id` string, required: New node id
          - `new_node_name` string, required: New node name
          - `transition_type` 'global' | 'global_go_back' | 'interrupt_go_back' | 'normal': How this node was reached. "global" means a global node transition, "global_go_back" means returning from a global node, "interrupt_go_back" means going back due to user interruption, and "normal" means a regular edge transition.
        - DTMFUtterance
          - `role` 'dtmf', required: Digit pressed by the user from their phone keypad.
          - `digit` string, required: The digit pressed by the user. Will be a single digit string like "1", "2", "3", "*", "#" etc.
        - SmsUtterance
          - `role` 'sms', required: SMS message received from the user during the call (for example while the agent is leaving a voicemail). Not part of the spoken conversation.
          - `content` string, required: Text content of the SMS message.
          - `multimedia` SmsMultimediaItem[]: Multimedia attachments (MMS). Display only; not relayed into the spoken conversation.
            - `url` string, required: URL of the multimedia attachment.
            - `summary` string: Optional textual summary of the attachment.
          - `time_sec` number, required: Time the SMS was received, in seconds relative to the start of the call.
        - InjectedUtterance
          - `role` 'injected', required: External context injected into the conversation via the update-live-call API. Not spoken by either party.
          - `content` string, required: The injected context text.
          - `time_sec` number, required: Time the context was injected, in seconds relative to the start of the call.
    - `scrubbed_post_session_transcript_with_tool_calls` UtteranceOrToolCall[]: Post-session integration tool call invocations and results, without PII. Available after call ends if post-session integration tools ran and PII scrubbing is enabled.
      - union
        - Utterance
          - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
          - `content` string, required: Transcript of the utterances.
          - `metadata` object
            - `response_id` integer: Call-scoped agent response identifier, when available.
            - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
            - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
            - `word` string: Word transcript (with punctuation if applicable).
            - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
            - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
        - ToolCallInvocationUtterance
          - `role` 'tool_call_invocation', required: This is a tool call invocation.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `name` string, required: Name of the tool. This is the configured tool name, or the node name for a built-in conversation flow node, so it can contain spaces and non-ASCII characters. To identify the kind of tool, use `type`.
          - `type` string: Kind of tool. Known values are `custom`, `code`, `mcp`, `integration_app`, `end_call`, `transfer_call`, `bridge_transfer`, `cancel_transfer`, `agent_swap`, `press_digit`, `send_sms`, `extract_dynamic_variable`, `adjust_voice_speed`, `book_appointment_cal`, and `check_availability_cal`. New values can be added. Absent on tool calls sent by a custom LLM, and on some older calls and chats.
          - `arguments` string, required: Arguments for this tool call, it's a stringified JSON object.
          - `metadata` object
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `thought_signature` string: Optional thought signature from Google Gemini thinking models. This is used internally to maintain reasoning chain in multi-turn function calling.
        - ToolCallResultUtterance
          - `role` 'tool_call_result', required: This is the result of a tool call.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `content` string, required: Result of the tool call, can be a string, a stringified json, etc.
          - `successful` boolean: Whether the tool call was successful.
        - NodeTransitionUtterance
          - `role` 'node_transition', required: This is result of a node transition
          - `former_node_id` string, required: Former node id
          - `former_node_name` string, required: Former node name
          - `new_node_id` string, required: New node id
          - `new_node_name` string, required: New node name
          - `transition_type` 'global' | 'global_go_back' | 'interrupt_go_back' | 'normal': How this node was reached. "global" means a global node transition, "global_go_back" means returning from a global node, "interrupt_go_back" means going back due to user interruption, and "normal" means a regular edge transition.
        - DTMFUtterance
          - `role` 'dtmf', required: Digit pressed by the user from their phone keypad.
          - `digit` string, required: The digit pressed by the user. Will be a single digit string like "1", "2", "3", "*", "#" etc.
        - SmsUtterance
          - `role` 'sms', required: SMS message received from the user during the call (for example while the agent is leaving a voicemail). Not part of the spoken conversation.
          - `content` string, required: Text content of the SMS message.
          - `multimedia` SmsMultimediaItem[]: Multimedia attachments (MMS). Display only; not relayed into the spoken conversation.
            - `url` string, required: URL of the multimedia attachment.
            - `summary` string: Optional textual summary of the attachment.
          - `time_sec` number, required: Time the SMS was received, in seconds relative to the start of the call.
        - InjectedUtterance
          - `role` 'injected', required: External context injected into the conversation via the update-live-call API. Not spoken by either party.
          - `content` string, required: The injected context text.
          - `time_sec` number, required: Time the context was injected, in seconds relative to the start of the call.
    - `recording_url` string: Recording of the call. Available after call ends.
    - `recording_multi_channel_url` string: Recording of the call, with each party's audio stored in a separate channel. Available after the call ends.
    - `scrubbed_recording_url` string: Recording of the call without PII. Available after call ends.
    - `scrubbed_recording_multi_channel_url` string: Recording of the call without PII, with each party's audio stored in a separate channel. Available after the call ends.
    - `public_log_url` string: Public log of the call, containing details about all the requests and responses received in LLM WebSocket, latency tracking for each turntaking, helpful for debugging and tracing. Available after call ends.
    - `knowledge_base_retrieved_contents_url` string: URL to the knowledge base retrieved contents of the call. Available after call ends if the call utilizes knowledge base feature. It consists of the respond id and the retrieved contents related to that response. It's already rendered in call history tab of dashboard, and you can also manually download and check against the transcript to view the knowledge base retrieval results.
    - `latency` object: Latency tracking of the call, available after call ends. Not all fields here will be available, as it depends on the type of call and feature used.
      - `e2e` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
      - `asr` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
      - `llm` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
      - `llm_websocket_network_rtt` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
      - `tts` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
      - `knowledge_base` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
      - `s2s` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
    - `disconnection_reason` 'user_hangup' | 'agent_hangup' | 'user_requested_dnc' | 'user_requested_callback' | 'call_transfer' | 'voicemail_reached' | 'ivr_reached' | 'inactivity' | 'max_duration_reached' | 'concurrency_limit_reached' | 'no_concurrency_fallback' | 'no_valid_payment' | 'credit_exhausted' | 'budget_reached' | 'scam_detected' | 'dial_busy' | 'dial_failed' | 'dial_no_answer' | 'invalid_destination' | 'telephony_provider_permission_denied' | 'telephony_provider_unavailable' | 'sip_routing_error' | 'marked_as_spam' | 'network_blocked' | 'user_declined' | 'error_llm_websocket_open' | 'error_llm_websocket_lost_connection' | 'error_llm_websocket_runtime' | 'error_llm_websocket_corrupt_payload' | 'error_no_audio_received' | 'error_asr' | 'error_retell' | 'error_unknown' | 'error_user_not_joined' | 'registered_call_timeout' | 'transfer_bridged' | 'transfer_cancelled' | 'manual_stopped' | 'call_take_over'
    - `transfer_destination` string, nullable: The destination number or identifier where the call was transferred to. Only populated when the disconnection reason was `call_transfer`. Can be a phone number or a SIP URI. SIP URIs are prefixed with "sip:" and may include a ";transport=..." portion (if transport is known) where the transport type can be "tls", "tcp" or "udp".
    - `call_analysis` CallAnalysis
      - `call_summary` string: A high level summary of the call.
      - `in_voicemail` boolean: Whether the call is entered voicemail.
      - `user_sentiment` 'Negative' | 'Positive' | 'Neutral' | 'Unknown': Sentiment of the user in the call.
      - `call_successful` boolean: Whether the agent seems to have a successful call with the user, where the agent finishes the task, and the call was complete without being cutoff.
      - `custom_analysis_data` object: Custom analysis data that was extracted based on the schema defined in agent post call analysis data. Can be empty if nothing is specified.
    - `call_cost` object: Cost of the call, including all the products and their costs and discount.
      - `product_costs` ProductCost[], required: List of products with their unit prices and costs in cents
        - `product` string, required: Product name that has a cost associated with it.
        - `unit_price` number: Unit price of the product in cents per second.
        - `cost` number, required: Cost for the product in cents for the duration of the call.
        - `is_transfer_leg_cost` boolean: True if this cost item is for a transfer segment.
      - `total_duration_seconds` number, required: Total duration of the call in seconds
      - `total_duration_unit_price` number, required: Total unit duration price of all products in cents per second
      - `combined_cost` number, required: Combined cost of all individual costs in cents
    - `unbilled_response_periods` object[]: Spans of the call with no user or agent speech and no call event (start of a DTMF entry, SMS, node transition, tool call start or result, or injected message), minus the first 10 seconds of each such gap. A tool call's run time counts as silence, since no LLM or TTS is working then. Payment time is not included. LLM, TTS, realtime (speech-to-speech) model time and the LLM token surcharge are not billed over these spans. Other products, such as the voice engine, and full-duplex (GPT Live) voice models are still billed over them. Each product's reduction is rounded down to whole seconds. Times are in milliseconds from the start of the call. Only present when such spans occurred.
      - `start_ms` number, required: Start of the span in milliseconds from call start.
      - `end_ms` number, required: End of the span in milliseconds from call start.
      - `type` 'cascading' | 's2s' | 'duplex': The engine that ran during the span. A span is split where an agent swap changes the engine. `cascading` credits LLM and TTS time, `s2s` credits the realtime model's time, and `duplex` credits only the backend model's time, because the duplex voice model is still billed in full. Absent when the engine is not known.
    - `unbilled_response_seconds` number: Total seconds of `unbilled_response_periods`, rounded down. See `unbilled_response_periods` for what is and is not billed over them. Only present when `unbilled_response_periods` is present.
    - `llm_token_usage` object: LLM token usage of the call, available after call ends. Not populated if using custom LLM, realtime API, or no LLM call is made.
      - `values` number[], required: All the token count values in the call.
      - `average` number, required: Average token count of the call.
      - `num_requests` number, required: Number of requests made to the LLM.
  - V2PhoneCallResponse
    - `call_type` 'phone_call', required: Type of the call. Used to distinguish between web call and phone call.
    - `from_number` string, required: The caller number.
    - `to_number` string, required: The callee number.
    - `direction` 'inbound' | 'outbound', required: Direction of the phone call.
    - `telephony_identifier` object: Telephony identifier of the call, populated when available. Tracking purposes only.
      - `twilio_call_sid` string: Twilio call sid.
    - `call_id` string, required: Unique id of the call. Used to identify the call in the LLM websocket and used to authenticate in the audio websocket.
    - `agent_id` string, required: Corresponding agent id of this call.
    - `agent_name` string: Name of the agent.
    - `agent_version` integer, required: The version of the agent.
    - `agent_tag` string, nullable: Tag pointing at the agent version used for this call, captured at call creation time and frozen thereafter (unaffected by later tag reassignments). Populated whether the caller dispatched by tag, numeric version, "latest", or "latest_published" — when the caller specified a tag, that tag wins; otherwise the most-recently- assigned tag on the resolved version is used. Absent when no tag points at the resolved version (or for calls created before this field was introduced).
    - `call_status` 'registered' | 'not_connected' | 'ongoing' | 'ended' | 'error', required: Status of call. - `registered`: Call id issued, starting to make a call using this id. - `ongoing`: Call connected and ongoing. - `ended`: The underlying websocket has ended for the call. Either user or agent hung up, or call transferred. - `error`: Call encountered error.
    - `metadata` object: An arbitrary object for storage purpose only. You can put anything here like your internal customer id associated with the call. Not used for processing. You can later get this field from the call object.
    - `retell_llm_dynamic_variables` object: Add optional dynamic variables in key value pairs of string that injects into your Response Engine prompt and tool description. Only applicable for Response Engine.
    - `collected_dynamic_variables` object: Dynamic variables collected from the call. Only available after the call ends.
    - `custom_sip_headers` object: Custom SIP headers to be added to the call.
    - `data_storage_setting` 'everything' | 'everything_except_pii' | 'basic_attributes_only', nullable: Data storage setting for this call's agent. "everything" stores all data, "everything_except_pii" excludes PII when possible, "basic_attributes_only" stores only metadata.
    - `opt_in_signed_url` boolean: Whether this agent opts in for signed URLs for public logs and recordings. When enabled, the generated URLs will include security signatures that restrict access and automatically expire after 24 hours.
    - `start_timestamp` integer: Begin timestamp (milliseconds since epoch) of the call. Available after call starts.
    - `end_timestamp` integer: End timestamp (milliseconds since epoch) of the call. Available after call ends.
    - `transfer_end_timestamp` integer: Transfer end timestamp (milliseconds since epoch) of the call. Available after transfer call ends.
    - `duration_ms` integer: Duration of the call in milliseconds. Available after call ends.
    - `transcript` string: Transcription of the call. Available after call ends.
    - `transcript_object` Utterance[]: Transcript of the call in the format of a list of utterance, with timestamp. Available after call ends.
      - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
      - `content` string, required: Transcript of the utterances.
      - `metadata` object
        - `response_id` integer: Call-scoped agent response identifier, when available.
        - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
        - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
        - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
          - `model` string, required: Actual selected model after routing and retries.
          - `primary_model` string, required: Primary model configured for this request.
          - `backup_model` string, required: Backup model configured for this request.
          - `reason` 'primary_failed' | 'preference_expired' | 'first_usable', required: Why the backup won. primary_failed means primary attempts failed; preference_expired means backup was ready within the primary preference window; first_usable means backup became ready after that window and won the race.
          - `selection_latency_ms` number, required: Elapsed milliseconds from race start to winner selection.
          - `primary_ready_ms` number: Elapsed milliseconds until primary produced usable output. Omitted if primary was not ready when the winner was selected.
          - `backup_ready_ms` number: Elapsed milliseconds until backup produced usable output. Omitted when unavailable.
      - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
        - `word` string: Word transcript (with punctuation if applicable).
        - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
        - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
    - `transcript_with_tool_calls` UtteranceOrToolCall[]: Transcript of the call weaved with tool call invocation and results. It precisely captures when (at what utterance, which word) the tool was invoked and what was the result. Available after call ends.
      - union
        - Utterance
          - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
          - `content` string, required: Transcript of the utterances.
          - `metadata` object
            - `response_id` integer: Call-scoped agent response identifier, when available.
            - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
            - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
            - `word` string: Word transcript (with punctuation if applicable).
            - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
            - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
        - ToolCallInvocationUtterance
          - `role` 'tool_call_invocation', required: This is a tool call invocation.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `name` string, required: Name of the tool. This is the configured tool name, or the node name for a built-in conversation flow node, so it can contain spaces and non-ASCII characters. To identify the kind of tool, use `type`.
          - `type` string: Kind of tool. Known values are `custom`, `code`, `mcp`, `integration_app`, `end_call`, `transfer_call`, `bridge_transfer`, `cancel_transfer`, `agent_swap`, `press_digit`, `send_sms`, `extract_dynamic_variable`, `adjust_voice_speed`, `book_appointment_cal`, and `check_availability_cal`. New values can be added. Absent on tool calls sent by a custom LLM, and on some older calls and chats.
          - `arguments` string, required: Arguments for this tool call, it's a stringified JSON object.
          - `metadata` object
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `thought_signature` string: Optional thought signature from Google Gemini thinking models. This is used internally to maintain reasoning chain in multi-turn function calling.
        - ToolCallResultUtterance
          - `role` 'tool_call_result', required: This is the result of a tool call.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `content` string, required: Result of the tool call, can be a string, a stringified json, etc.
          - `successful` boolean: Whether the tool call was successful.
        - NodeTransitionUtterance
          - `role` 'node_transition', required: This is result of a node transition
          - `former_node_id` string, required: Former node id
          - `former_node_name` string, required: Former node name
          - `new_node_id` string, required: New node id
          - `new_node_name` string, required: New node name
          - `transition_type` 'global' | 'global_go_back' | 'interrupt_go_back' | 'normal': How this node was reached. "global" means a global node transition, "global_go_back" means returning from a global node, "interrupt_go_back" means going back due to user interruption, and "normal" means a regular edge transition.
        - DTMFUtterance
          - `role` 'dtmf', required: Digit pressed by the user from their phone keypad.
          - `digit` string, required: The digit pressed by the user. Will be a single digit string like "1", "2", "3", "*", "#" etc.
        - SmsUtterance
          - `role` 'sms', required: SMS message received from the user during the call (for example while the agent is leaving a voicemail). Not part of the spoken conversation.
          - `content` string, required: Text content of the SMS message.
          - `multimedia` SmsMultimediaItem[]: Multimedia attachments (MMS). Display only; not relayed into the spoken conversation.
            - `url` string, required: URL of the multimedia attachment.
            - `summary` string: Optional textual summary of the attachment.
          - `time_sec` number, required: Time the SMS was received, in seconds relative to the start of the call.
        - InjectedUtterance
          - `role` 'injected', required: External context injected into the conversation via the update-live-call API. Not spoken by either party.
          - `content` string, required: The injected context text.
          - `time_sec` number, required: Time the context was injected, in seconds relative to the start of the call.
    - `scrubbed_transcript_with_tool_calls` UtteranceOrToolCall[]: Transcript of the call weaved with tool call invocation and results, without PII. It precisely captures when (at what utterance, which word) the tool was invoked and what was the result. Available after call ends.
      - union
        - Utterance
          - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
          - `content` string, required: Transcript of the utterances.
          - `metadata` object
            - `response_id` integer: Call-scoped agent response identifier, when available.
            - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
            - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
            - `word` string: Word transcript (with punctuation if applicable).
            - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
            - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
        - ToolCallInvocationUtterance
          - `role` 'tool_call_invocation', required: This is a tool call invocation.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `name` string, required: Name of the tool. This is the configured tool name, or the node name for a built-in conversation flow node, so it can contain spaces and non-ASCII characters. To identify the kind of tool, use `type`.
          - `type` string: Kind of tool. Known values are `custom`, `code`, `mcp`, `integration_app`, `end_call`, `transfer_call`, `bridge_transfer`, `cancel_transfer`, `agent_swap`, `press_digit`, `send_sms`, `extract_dynamic_variable`, `adjust_voice_speed`, `book_appointment_cal`, and `check_availability_cal`. New values can be added. Absent on tool calls sent by a custom LLM, and on some older calls and chats.
          - `arguments` string, required: Arguments for this tool call, it's a stringified JSON object.
          - `metadata` object
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `thought_signature` string: Optional thought signature from Google Gemini thinking models. This is used internally to maintain reasoning chain in multi-turn function calling.
        - ToolCallResultUtterance
          - `role` 'tool_call_result', required: This is the result of a tool call.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `content` string, required: Result of the tool call, can be a string, a stringified json, etc.
          - `successful` boolean: Whether the tool call was successful.
        - NodeTransitionUtterance
          - `role` 'node_transition', required: This is result of a node transition
          - `former_node_id` string, required: Former node id
          - `former_node_name` string, required: Former node name
          - `new_node_id` string, required: New node id
          - `new_node_name` string, required: New node name
          - `transition_type` 'global' | 'global_go_back' | 'interrupt_go_back' | 'normal': How this node was reached. "global" means a global node transition, "global_go_back" means returning from a global node, "interrupt_go_back" means going back due to user interruption, and "normal" means a regular edge transition.
        - DTMFUtterance
          - `role` 'dtmf', required: Digit pressed by the user from their phone keypad.
          - `digit` string, required: The digit pressed by the user. Will be a single digit string like "1", "2", "3", "*", "#" etc.
        - SmsUtterance
          - `role` 'sms', required: SMS message received from the user during the call (for example while the agent is leaving a voicemail). Not part of the spoken conversation.
          - `content` string, required: Text content of the SMS message.
          - `multimedia` SmsMultimediaItem[]: Multimedia attachments (MMS). Display only; not relayed into the spoken conversation.
            - `url` string, required: URL of the multimedia attachment.
            - `summary` string: Optional textual summary of the attachment.
          - `time_sec` number, required: Time the SMS was received, in seconds relative to the start of the call.
        - InjectedUtterance
          - `role` 'injected', required: External context injected into the conversation via the update-live-call API. Not spoken by either party.
          - `content` string, required: The injected context text.
          - `time_sec` number, required: Time the context was injected, in seconds relative to the start of the call.
    - `pre_session_transcript_with_tool_calls` UtteranceOrToolCall[]: Tool call invocations and results of integration tools run before the session started (pre-session). Stored separately from the main transcript. Available after call ends if pre-session integration tools ran.
      - union
        - Utterance
          - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
          - `content` string, required: Transcript of the utterances.
          - `metadata` object
            - `response_id` integer: Call-scoped agent response identifier, when available.
            - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
            - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
            - `word` string: Word transcript (with punctuation if applicable).
            - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
            - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
        - ToolCallInvocationUtterance
          - `role` 'tool_call_invocation', required: This is a tool call invocation.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `name` string, required: Name of the tool. This is the configured tool name, or the node name for a built-in conversation flow node, so it can contain spaces and non-ASCII characters. To identify the kind of tool, use `type`.
          - `type` string: Kind of tool. Known values are `custom`, `code`, `mcp`, `integration_app`, `end_call`, `transfer_call`, `bridge_transfer`, `cancel_transfer`, `agent_swap`, `press_digit`, `send_sms`, `extract_dynamic_variable`, `adjust_voice_speed`, `book_appointment_cal`, and `check_availability_cal`. New values can be added. Absent on tool calls sent by a custom LLM, and on some older calls and chats.
          - `arguments` string, required: Arguments for this tool call, it's a stringified JSON object.
          - `metadata` object
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `thought_signature` string: Optional thought signature from Google Gemini thinking models. This is used internally to maintain reasoning chain in multi-turn function calling.
        - ToolCallResultUtterance
          - `role` 'tool_call_result', required: This is the result of a tool call.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `content` string, required: Result of the tool call, can be a string, a stringified json, etc.
          - `successful` boolean: Whether the tool call was successful.
        - NodeTransitionUtterance
          - `role` 'node_transition', required: This is result of a node transition
          - `former_node_id` string, required: Former node id
          - `former_node_name` string, required: Former node name
          - `new_node_id` string, required: New node id
          - `new_node_name` string, required: New node name
          - `transition_type` 'global' | 'global_go_back' | 'interrupt_go_back' | 'normal': How this node was reached. "global" means a global node transition, "global_go_back" means returning from a global node, "interrupt_go_back" means going back due to user interruption, and "normal" means a regular edge transition.
        - DTMFUtterance
          - `role` 'dtmf', required: Digit pressed by the user from their phone keypad.
          - `digit` string, required: The digit pressed by the user. Will be a single digit string like "1", "2", "3", "*", "#" etc.
        - SmsUtterance
          - `role` 'sms', required: SMS message received from the user during the call (for example while the agent is leaving a voicemail). Not part of the spoken conversation.
          - `content` string, required: Text content of the SMS message.
          - `multimedia` SmsMultimediaItem[]: Multimedia attachments (MMS). Display only; not relayed into the spoken conversation.
            - `url` string, required: URL of the multimedia attachment.
            - `summary` string: Optional textual summary of the attachment.
          - `time_sec` number, required: Time the SMS was received, in seconds relative to the start of the call.
        - InjectedUtterance
          - `role` 'injected', required: External context injected into the conversation via the update-live-call API. Not spoken by either party.
          - `content` string, required: The injected context text.
          - `time_sec` number, required: Time the context was injected, in seconds relative to the start of the call.
    - `scrubbed_pre_session_transcript_with_tool_calls` UtteranceOrToolCall[]: Pre-session integration tool call invocations and results, without PII. Available after call ends if pre-session integration tools ran and PII scrubbing is enabled.
      - union
        - Utterance
          - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
          - `content` string, required: Transcript of the utterances.
          - `metadata` object
            - `response_id` integer: Call-scoped agent response identifier, when available.
            - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
            - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
            - `word` string: Word transcript (with punctuation if applicable).
            - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
            - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
        - ToolCallInvocationUtterance
          - `role` 'tool_call_invocation', required: This is a tool call invocation.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `name` string, required: Name of the tool. This is the configured tool name, or the node name for a built-in conversation flow node, so it can contain spaces and non-ASCII characters. To identify the kind of tool, use `type`.
          - `type` string: Kind of tool. Known values are `custom`, `code`, `mcp`, `integration_app`, `end_call`, `transfer_call`, `bridge_transfer`, `cancel_transfer`, `agent_swap`, `press_digit`, `send_sms`, `extract_dynamic_variable`, `adjust_voice_speed`, `book_appointment_cal`, and `check_availability_cal`. New values can be added. Absent on tool calls sent by a custom LLM, and on some older calls and chats.
          - `arguments` string, required: Arguments for this tool call, it's a stringified JSON object.
          - `metadata` object
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `thought_signature` string: Optional thought signature from Google Gemini thinking models. This is used internally to maintain reasoning chain in multi-turn function calling.
        - ToolCallResultUtterance
          - `role` 'tool_call_result', required: This is the result of a tool call.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `content` string, required: Result of the tool call, can be a string, a stringified json, etc.
          - `successful` boolean: Whether the tool call was successful.
        - NodeTransitionUtterance
          - `role` 'node_transition', required: This is result of a node transition
          - `former_node_id` string, required: Former node id
          - `former_node_name` string, required: Former node name
          - `new_node_id` string, required: New node id
          - `new_node_name` string, required: New node name
          - `transition_type` 'global' | 'global_go_back' | 'interrupt_go_back' | 'normal': How this node was reached. "global" means a global node transition, "global_go_back" means returning from a global node, "interrupt_go_back" means going back due to user interruption, and "normal" means a regular edge transition.
        - DTMFUtterance
          - `role` 'dtmf', required: Digit pressed by the user from their phone keypad.
          - `digit` string, required: The digit pressed by the user. Will be a single digit string like "1", "2", "3", "*", "#" etc.
        - SmsUtterance
          - `role` 'sms', required: SMS message received from the user during the call (for example while the agent is leaving a voicemail). Not part of the spoken conversation.
          - `content` string, required: Text content of the SMS message.
          - `multimedia` SmsMultimediaItem[]: Multimedia attachments (MMS). Display only; not relayed into the spoken conversation.
            - `url` string, required: URL of the multimedia attachment.
            - `summary` string: Optional textual summary of the attachment.
          - `time_sec` number, required: Time the SMS was received, in seconds relative to the start of the call.
        - InjectedUtterance
          - `role` 'injected', required: External context injected into the conversation via the update-live-call API. Not spoken by either party.
          - `content` string, required: The injected context text.
          - `time_sec` number, required: Time the context was injected, in seconds relative to the start of the call.
    - `post_session_transcript_with_tool_calls` UtteranceOrToolCall[]: Tool call invocations and results of integration tools run after post-call analysis (post-session). Stored separately from the main transcript. Available after call ends if post-session integration tools ran.
      - union
        - Utterance
          - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
          - `content` string, required: Transcript of the utterances.
          - `metadata` object
            - `response_id` integer: Call-scoped agent response identifier, when available.
            - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
            - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
            - `word` string: Word transcript (with punctuation if applicable).
            - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
            - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
        - ToolCallInvocationUtterance
          - `role` 'tool_call_invocation', required: This is a tool call invocation.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `name` string, required: Name of the tool. This is the configured tool name, or the node name for a built-in conversation flow node, so it can contain spaces and non-ASCII characters. To identify the kind of tool, use `type`.
          - `type` string: Kind of tool. Known values are `custom`, `code`, `mcp`, `integration_app`, `end_call`, `transfer_call`, `bridge_transfer`, `cancel_transfer`, `agent_swap`, `press_digit`, `send_sms`, `extract_dynamic_variable`, `adjust_voice_speed`, `book_appointment_cal`, and `check_availability_cal`. New values can be added. Absent on tool calls sent by a custom LLM, and on some older calls and chats.
          - `arguments` string, required: Arguments for this tool call, it's a stringified JSON object.
          - `metadata` object
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `thought_signature` string: Optional thought signature from Google Gemini thinking models. This is used internally to maintain reasoning chain in multi-turn function calling.
        - ToolCallResultUtterance
          - `role` 'tool_call_result', required: This is the result of a tool call.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `content` string, required: Result of the tool call, can be a string, a stringified json, etc.
          - `successful` boolean: Whether the tool call was successful.
        - NodeTransitionUtterance
          - `role` 'node_transition', required: This is result of a node transition
          - `former_node_id` string, required: Former node id
          - `former_node_name` string, required: Former node name
          - `new_node_id` string, required: New node id
          - `new_node_name` string, required: New node name
          - `transition_type` 'global' | 'global_go_back' | 'interrupt_go_back' | 'normal': How this node was reached. "global" means a global node transition, "global_go_back" means returning from a global node, "interrupt_go_back" means going back due to user interruption, and "normal" means a regular edge transition.
        - DTMFUtterance
          - `role` 'dtmf', required: Digit pressed by the user from their phone keypad.
          - `digit` string, required: The digit pressed by the user. Will be a single digit string like "1", "2", "3", "*", "#" etc.
        - SmsUtterance
          - `role` 'sms', required: SMS message received from the user during the call (for example while the agent is leaving a voicemail). Not part of the spoken conversation.
          - `content` string, required: Text content of the SMS message.
          - `multimedia` SmsMultimediaItem[]: Multimedia attachments (MMS). Display only; not relayed into the spoken conversation.
            - `url` string, required: URL of the multimedia attachment.
            - `summary` string: Optional textual summary of the attachment.
          - `time_sec` number, required: Time the SMS was received, in seconds relative to the start of the call.
        - InjectedUtterance
          - `role` 'injected', required: External context injected into the conversation via the update-live-call API. Not spoken by either party.
          - `content` string, required: The injected context text.
          - `time_sec` number, required: Time the context was injected, in seconds relative to the start of the call.
    - `scrubbed_post_session_transcript_with_tool_calls` UtteranceOrToolCall[]: Post-session integration tool call invocations and results, without PII. Available after call ends if post-session integration tools ran and PII scrubbing is enabled.
      - union
        - Utterance
          - `role` 'agent' | 'user' | 'transfer_target', required: Documents whether this utterance is spoken by agent or user.
          - `content` string, required: Transcript of the utterances.
          - `metadata` object
            - `response_id` integer: Call-scoped agent response identifier, when available.
            - `interrupt_id` integer: Custom LLM agent interruption identifier, when available.
            - `is_backchannel` boolean: Whether the utterance is a backchannel acknowledgment.
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `words` object[], required: Array of words in the utterance with the word timestamp. Useful for understanding what word was spoken at what time. Note that the word timestamp is not guaranteed to be accurate, it's more like an approximation.
            - `word` string: Word transcript (with punctuation if applicable).
            - `start` number: Start time of the word in the call in second. This is relative audio time, not wall time.
            - `end` number: End time of the word in the call in second. This is relative audio time, not wall time.
        - ToolCallInvocationUtterance
          - `role` 'tool_call_invocation', required: This is a tool call invocation.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `name` string, required: Name of the tool. This is the configured tool name, or the node name for a built-in conversation flow node, so it can contain spaces and non-ASCII characters. To identify the kind of tool, use `type`.
          - `type` string: Kind of tool. Known values are `custom`, `code`, `mcp`, `integration_app`, `end_call`, `transfer_call`, `bridge_transfer`, `cancel_transfer`, `agent_swap`, `press_digit`, `send_sms`, `extract_dynamic_variable`, `adjust_voice_speed`, `book_appointment_cal`, and `check_availability_cal`. New values can be added. Absent on tool calls sent by a custom LLM, and on some older calls and chats.
          - `arguments` string, required: Arguments for this tool call, it's a stringified JSON object.
          - `metadata` object
            - `backup_model_used` BackupModelUsed: Selection details for a winning backup model used in a response or tool invocation. Timings are measured from the start of the candidate race to selection or first usable output, not completion of generation.
              - …
          - `thought_signature` string: Optional thought signature from Google Gemini thinking models. This is used internally to maintain reasoning chain in multi-turn function calling.
        - ToolCallResultUtterance
          - `role` 'tool_call_result', required: This is the result of a tool call.
          - `tool_call_id` string, required: Tool call id, globally unique.
          - `content` string, required: Result of the tool call, can be a string, a stringified json, etc.
          - `successful` boolean: Whether the tool call was successful.
        - NodeTransitionUtterance
          - `role` 'node_transition', required: This is result of a node transition
          - `former_node_id` string, required: Former node id
          - `former_node_name` string, required: Former node name
          - `new_node_id` string, required: New node id
          - `new_node_name` string, required: New node name
          - `transition_type` 'global' | 'global_go_back' | 'interrupt_go_back' | 'normal': How this node was reached. "global" means a global node transition, "global_go_back" means returning from a global node, "interrupt_go_back" means going back due to user interruption, and "normal" means a regular edge transition.
        - DTMFUtterance
          - `role` 'dtmf', required: Digit pressed by the user from their phone keypad.
          - `digit` string, required: The digit pressed by the user. Will be a single digit string like "1", "2", "3", "*", "#" etc.
        - SmsUtterance
          - `role` 'sms', required: SMS message received from the user during the call (for example while the agent is leaving a voicemail). Not part of the spoken conversation.
          - `content` string, required: Text content of the SMS message.
          - `multimedia` SmsMultimediaItem[]: Multimedia attachments (MMS). Display only; not relayed into the spoken conversation.
            - `url` string, required: URL of the multimedia attachment.
            - `summary` string: Optional textual summary of the attachment.
          - `time_sec` number, required: Time the SMS was received, in seconds relative to the start of the call.
        - InjectedUtterance
          - `role` 'injected', required: External context injected into the conversation via the update-live-call API. Not spoken by either party.
          - `content` string, required: The injected context text.
          - `time_sec` number, required: Time the context was injected, in seconds relative to the start of the call.
    - `recording_url` string: Recording of the call. Available after call ends.
    - `recording_multi_channel_url` string: Recording of the call, with each party's audio stored in a separate channel. Available after the call ends.
    - `scrubbed_recording_url` string: Recording of the call without PII. Available after call ends.
    - `scrubbed_recording_multi_channel_url` string: Recording of the call without PII, with each party's audio stored in a separate channel. Available after the call ends.
    - `public_log_url` string: Public log of the call, containing details about all the requests and responses received in LLM WebSocket, latency tracking for each turntaking, helpful for debugging and tracing. Available after call ends.
    - `knowledge_base_retrieved_contents_url` string: URL to the knowledge base retrieved contents of the call. Available after call ends if the call utilizes knowledge base feature. It consists of the respond id and the retrieved contents related to that response. It's already rendered in call history tab of dashboard, and you can also manually download and check against the transcript to view the knowledge base retrieval results.
    - `latency` object: Latency tracking of the call, available after call ends. Not all fields here will be available, as it depends on the type of call and feature used.
      - `e2e` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
      - `asr` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
      - `llm` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
      - `llm_websocket_network_rtt` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
      - `tts` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
      - `knowledge_base` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
      - `s2s` CallLatency
        - `p50` number: 50 percentile of latency, measured in milliseconds.
        - `p90` number: 90 percentile of latency, measured in milliseconds.
        - `p95` number: 95 percentile of latency, measured in milliseconds.
        - `p99` number: 99 percentile of latency, measured in milliseconds.
        - `max` number: Maximum latency in the call, measured in milliseconds.
        - `min` number: Minimum latency in the call, measured in milliseconds.
        - `num` number: Number of data points (number of times latency is tracked).
        - `values` number[]: All the latency data points in the call, measured in milliseconds.
    - `disconnection_reason` 'user_hangup' | 'agent_hangup' | 'user_requested_dnc' | 'user_requested_callback' | 'call_transfer' | 'voicemail_reached' | 'ivr_reached' | 'inactivity' | 'max_duration_reached' | 'concurrency_limit_reached' | 'no_concurrency_fallback' | 'no_valid_payment' | 'credit_exhausted' | 'budget_reached' | 'scam_detected' | 'dial_busy' | 'dial_failed' | 'dial_no_answer' | 'invalid_destination' | 'telephony_provider_permission_denied' | 'telephony_provider_unavailable' | 'sip_routing_error' | 'marked_as_spam' | 'network_blocked' | 'user_declined' | 'error_llm_websocket_open' | 'error_llm_websocket_lost_connection' | 'error_llm_websocket_runtime' | 'error_llm_websocket_corrupt_payload' | 'error_no_audio_received' | 'error_asr' | 'error_retell' | 'error_unknown' | 'error_user_not_joined' | 'registered_call_timeout' | 'transfer_bridged' | 'transfer_cancelled' | 'manual_stopped' | 'call_take_over'
    - `transfer_destination` string, nullable: The destination number or identifier where the call was transferred to. Only populated when the disconnection reason was `call_transfer`. Can be a phone number or a SIP URI. SIP URIs are prefixed with "sip:" and may include a ";transport=..." portion (if transport is known) where the transport type can be "tls", "tcp" or "udp".
    - `call_analysis` CallAnalysis
      - `call_summary` string: A high level summary of the call.
      - `in_voicemail` boolean: Whether the call is entered voicemail.
      - `user_sentiment` 'Negative' | 'Positive' | 'Neutral' | 'Unknown': Sentiment of the user in the call.
      - `call_successful` boolean: Whether the agent seems to have a successful call with the user, where the agent finishes the task, and the call was complete without being cutoff.
      - `custom_analysis_data` object: Custom analysis data that was extracted based on the schema defined in agent post call analysis data. Can be empty if nothing is specified.
    - `call_cost` object: Cost of the call, including all the products and their costs and discount.
      - `product_costs` ProductCost[], required: List of products with their unit prices and costs in cents
        - `product` string, required: Product name that has a cost associated with it.
        - `unit_price` number: Unit price of the product in cents per second.
        - `cost` number, required: Cost for the product in cents for the duration of the call.
        - `is_transfer_leg_cost` boolean: True if this cost item is for a transfer segment.
      - `total_duration_seconds` number, required: Total duration of the call in seconds
      - `total_duration_unit_price` number, required: Total unit duration price of all products in cents per second
      - `combined_cost` number, required: Combined cost of all individual costs in cents
    - `unbilled_response_periods` object[]: Spans of the call with no user or agent speech and no call event (start of a DTMF entry, SMS, node transition, tool call start or result, or injected message), minus the first 10 seconds of each such gap. A tool call's run time counts as silence, since no LLM or TTS is working then. Payment time is not included. LLM, TTS, realtime (speech-to-speech) model time and the LLM token surcharge are not billed over these spans. Other products, such as the voice engine, and full-duplex (GPT Live) voice models are still billed over them. Each product's reduction is rounded down to whole seconds. Times are in milliseconds from the start of the call. Only present when such spans occurred.
      - `start_ms` number, required: Start of the span in milliseconds from call start.
      - `end_ms` number, required: End of the span in milliseconds from call start.
      - `type` 'cascading' | 's2s' | 'duplex': The engine that ran during the span. A span is split where an agent swap changes the engine. `cascading` credits LLM and TTS time, `s2s` credits the realtime model's time, and `duplex` credits only the backend model's time, because the duplex voice model is still billed in full. Absent when the engine is not known.
    - `unbilled_response_seconds` number: Total seconds of `unbilled_response_periods`, rounded down. See `unbilled_response_periods` for what is and is not billed over them. Only present when `unbilled_response_periods` is present.
    - `llm_token_usage` object: LLM token usage of the call, available after call ends. Not populated if using custom LLM, realtime API, or no LLM call is made.
      - `values` number[], required: All the token count values in the call.
      - `average` number, required: Average token count of the call.
      - `num_requests` number, required: Number of requests made to the LLM.

## Other responses

- `400`: Bad Request
- `401`: Unauthorized
- `402`: Payment Required
- `404`: Not Found
- `422`: Unprocessable Content
- `429`: Too Many Requests
- `500`: Internal Server Error

## Changes

- **2026-11-05** `e482acaa292c`: 4 info
  - added the optional property `oneOf[#/components/schemas/V2PhoneCallResponse]/allOf[#/components/schemas/V2CallBase]/unbilled_response_periods` to the response with the `201` status
  - added the optional property `oneOf[#/components/schemas/V2PhoneCallResponse]/allOf[#/components/schemas/V2CallBase]/unbilled_response_seconds` to the response with the `201` status
  - added the optional property `oneOf[#/components/schemas/V2WebCallResponse]/allOf[#/components/schemas/V2CallBase]/unbilled_response_periods` to the response with the `201` status
  - added the optional property `oneOf[#/components/schemas/V2WebCallResponse]/allOf[#/components/schemas/V2CallBase]/unbilled_response_seconds` to the response with the `201` status
- …earlier changes not shown

[Operation history](https://skmtc.dev/retellai/apis/retell-sdk/changes/rerun-call-analysis/:call_id/put.md)

---

[API](https://skmtc.dev/retellai/apis/retell-sdk.md) - [All operations](https://skmtc.dev/retellai/apis/retell-sdk/llms.txt) - [OpenAPI document](https://skmtc.dev/retellai/apis/retell-sdk/revisions/e482acaa292c?raw)