Call Notes & Scoring

Overview

Two post-conversation tools, both keyed by a conversation id (the id from Conversations; every voice call and chat has one):

• Call notes — POST /api/v1/calls/{conversation_id}/notes reads the transcript and writes a structured note: a summary, action items, a disposition, a follow-up date and tags. Edit it, then push it to your CRM.
• Call scoring — write a scorecard (a rubric of weighted criteria), then POST /api/v1/calls/{conversation_id}/score grades the conversation against it. Every criterion comes back with a 0–100 score and the transcript line that justifies it.

Both run one language-model completion and are billed in credits at that model's per-token LLM rate. Reading, editing and pushing notes, and managing scorecards, are free.

Authentication

Authorization: Bearer cm_your_api_key

Base URL: https://api.callmissed.com. Errors are {"detail": "..."}. A conversation or scorecard outside your workspace returns 404, never 403.

Call notes

The notes object

{
  "conversation_id": "c0ffee00-1111-2222-3333-444455556666",
  "notes": {
    "summary": "Caller asked to move her Thursday appointment. Agent offered Friday 11:00, which she accepted.",
    "action_items": ["Send the Friday confirmation by WhatsApp"],
    "disposition": "rescheduled",
    "follow_up_date": "2026-10-09",
    "tags": ["appointment", "reschedule"],
    "generated_at": "2026-10-02T09:30:00+00:00",
    "edited_at": null,
    "edited_by": null,
    "pushed": null
  }
}

notes is null until notes have been generated.

GET /api/v1/calls/{conversation_id}/notes

Returns the notes object — notes: null if none have been generated yet.

POST /api/v1/calls/{conversation_id}/notes

Generates notes from the transcript and saves them.

curl -X POST https://api.callmissed.com/api/v1/calls/c0ffee00-1111-2222-3333-444455556666/notes \
  -H "Authorization: Bearer cm_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{}'

The note is written by the conversation agent's configured model, or gpt-oss-120b when the agent has none, and billed at that model's token rate.

PATCH /api/v1/calls/{conversation_id}/notes

Edit the stored notes. Every field is optional; only what you send changes.

404 when no notes have been generated yet.

POST /api/v1/calls/{conversation_id}/notes/push

Pushes the stored notes to your CRM as a note, attached to the matching CRM contact when the conversation's contact has an email that exists there.

Connect HubSpot under Integrations in the console first.

{
  "provider": "hubspot",
  "note_id": "81234567890",
  "contact_id": "5123",
  "associated": true
}

associated is false when no matching contact was found — the note is still created, just not attached.

Scorecards

A scorecard is your QA rubric.

{
  "id": "5c0a1b2c-3d4e-4f50-8a6b-7c8d9e0f1a2b",
  "name": "Support QA",
  "criteria": [
    { "label": "Verified caller identity", "weight": 2, "guidance": "Asked for name and booking reference before discussing the booking." },
    { "label": "Resolved or escalated", "weight": 3, "guidance": null },
    { "label": "Polite close", "weight": 1, "guidance": null }
  ],
  "created_at": "2026-10-01T08:00:00Z",
  "updated_at": "2026-10-01T08:00:00Z"
}

A scorecard has 1–20 criteria. A rule violation returns 400 naming the criterion.

GET /api/v1/scorecards

Newest first. limit 1..500 (default 100), offset 0..100000.

POST /api/v1/scorecards

Body: name (1–200 characters) and criteria. Returns 201 with the scorecard.

curl -X POST https://api.callmissed.com/api/v1/scorecards \
  -H "Authorization: Bearer cm_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Support QA",
    "criteria": [
      {"label": "Verified caller identity", "weight": 2},
      {"label": "Resolved or escalated", "weight": 3},
      {"label": "Polite close", "weight": 1}
    ]
  }'

GET /api/v1/scorecards/{scorecard_id}

Returns one scorecard.

PUT /api/v1/scorecards/{scorecard_id}

Replaces name and criteria (both required, same rules as create). Scores already taken keep their own breakdown, so past grades are unchanged.

DELETE /api/v1/scorecards/{scorecard_id}

Returns 204. Scores taken with it survive with scorecard_id: null.

Scoring a call

POST /api/v1/calls/{conversation_id}/score

curl -X POST https://api.callmissed.com/api/v1/calls/c0ffee00-1111-2222-3333-444455556666/score \
  -H "Authorization: Bearer cm_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{"scorecard_id": "5c0a1b2c-3d4e-4f50-8a6b-7c8d9e0f1a2b"}'

{
  "conversation_id": "c0ffee00-1111-2222-3333-444455556666",
  "scorecard_id": "5c0a1b2c-3d4e-4f50-8a6b-7c8d9e0f1a2b",
  "total": 76.67,
  "breakdown": [
    { "criterion": "Verified caller identity", "score": 90, "evidence_quote": "Agent: Could I have your booking reference, please?" },
    { "criterion": "Resolved or escalated", "score": 70, "evidence_quote": "Agent: I've moved you to Friday at eleven." },
    { "criterion": "Polite close", "score": 60, "evidence_quote": null }
  ],
  "model": "gpt-5-mini",
  "created_at": "2026-10-02T09:31:00Z"
}

• total is the weight-weighted mean of the criterion scores, on a 0–100 scale.
• breakdown always follows your rubric, in order. A criterion the model skipped scores 0 rather than disappearing, so a missing line can never inflate the total.
• evidence_quote is a verbatim transcript line, or null when nothing in the transcript supports the criterion.
• Re-scoring the same conversation with the same scorecard replaces the previous score.
• Billed at the model's LLM token rate. Long transcripts are trimmed from the middle before grading, keeping the opening and the close.

GET /api/v1/calls/{conversation_id}/score

The newest stored score for the conversation, across all scorecards. 404 This call has not been scored yet if there is none.