chat_analyzed | Tried Docs

Sent after post-chat analysis is stored, with the chat summary, success flag, sentiment and custom fields.

chat_analyzed is the last chat event. It repeats the ended chat and adds chat_analysis.

When it fires#

After chat_ended has been delivered (or has used up its retries), Tried runs post-chat analysis with the chat agent's post_chat_analysis_model (default gpt-5.6-terra), stores it and sends chat_analyzed.

It fires again when you re-run analysis on an ended chat with POST /rerun-chat-analysis/:chat_id.

Which settings enable it#

chat_analyzed in the chat agent's webhook_events (on by default).
URL: the chat agent's webhook_url, else the workspace URL.
Timeout: the workspace timeout, else 10 s.
Custom fields come from the chat agent's post_chat_analysis_data. analysis_summary_prompt and analysis_successful_prompt customise the summary and success judgement.

Payload#

{
  "event": "chat_analyzed",
  "chat": {
    "chat_id": "chat_2d7f1a4c8e0b3f6a9c5e1d42",
    "chat_agent_id": "chat_agent_0c4f8e2a6b1d3f5a7c9e2b40",
    "agent_name": "Receptionist (chat)",
    "chat_type": "api_chat",
    "channel": "widget",
    "chat_status": "ended",
    "retell_llm_dynamic_variables": {
      "business_name": "Bondi Dental"
    },
    "metadata": {
      "origin": "https://www.bondidental.com.au",
      "public_key_id": "pk_4b8e1d5a9c2f6e0a3d7b1c58"
    },
    "start_timestamp": 1760679000210,
    "end_timestamp": 1760679391020,
    "transcript": "Agent: Hi, I'm Ava from Bondi Dental. How can I help?\nUser: Are you open on Saturday?\nAgent: Yes, we're open Saturday from 9am to 1pm.\nUser: Great, thanks.\nAgent: You're welcome. Anything else I can help with?",
    "message_with_tool_calls": [
      { "role": "agent", "content": "Hi, I'm Ava from Bondi Dental. How can I help?", "created_timestamp": 1760679000845 },
      { "role": "user", "content": "Are you open on Saturday?", "created_timestamp": 1760679012100 },
      { "role": "agent", "content": "Yes, we're open Saturday from 9am to 1pm.", "created_timestamp": 1760679013320 },
      { "role": "user", "content": "Great, thanks.", "created_timestamp": 1760679030880 },
      { "role": "agent", "content": "You're welcome. Anything else I can help with?", "created_timestamp": 1760679031750 }
    ],
    "chat_analysis": {
      "chat_summary": "The visitor asked about Saturday opening hours and was told the clinic is open 9am to 1pm.",
      "chat_successful": true,
      "user_sentiment": "Positive",
      "custom_analysis_data": {
        "topic": "opening hours"
      }
    },
    "chat_cost": { "combined_cost": 0.01 }
  }
}

Fields#

The chat fields are described on chat_started. This event adds:

Field	Type	Description
`chat_analysis.chat_summary`	string	Short summary of the chat
`chat_analysis.chat_successful`	boolean	Whether the chat met its goal
`chat_analysis.user_sentiment`	string	`Positive`, `Negative`, `Neutral` or `Unknown`
`chat_analysis.custom_analysis_data`	object	One key per post-chat field. Fields the model could not fill are left out

Chats have no voicemail, so there is no in_voicemail field.

If the analysis model fails twice, chat_summary says the analysis could not be generated, chat_successful is false and user_sentiment is Unknown. In keyless local mode the values come from a mock model. Needs keys

How to respond#

Reply 2xx. The body is ignored. Retries: up to 3 (1 s, 2 s, 4 s).