Post-call actions over the API

When a call ends, your agent runs its post-call actions. Each action sends the call's results somewhere: an email report, a webhook to your server, a new row in a Google Sheet, a Slack message or a CRM record.

With the API you manage these actions one at a time. Add a webhook, and the agent's other actions stay exactly as they were, including the ones set up in the dashboard.

What your agent can send after a call

How it works

Before you start

• Your API key goes in the Authorization: Bearer header.
• For Google Sheets, Slack, Salesforce, HubSpot or Zoho Bigin, connect the account once in the dashboard under Integrations. Signing in to Google or Slack has to happen in a browser, so the API can't do that part.
• Resellers can manage a client's agent by adding user_id. See Acting as clients.

Step by step

curl "https://backend.omnidim.io/api/v1/agents/158910/post-call-actions" \
  -H "Authorization: Bearer YOUR_API_KEY"

{
  "actions": [
    {
      "id": 63535,
      "type": "email",
      "recipients": ["demo@example.com"],
      "include": ["summary"],
      "extracted_variables": [],
      "trigger_call_statuses": ["completed", "voicemail_detected"]
    }
  ],
  "applies_after_publish": false
}

curl -X POST "https://backend.omnidim.io/api/v1/agents/158910/post-call-actions" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: 6f1c2d9e-0b7a-4c1e-9a55-3e2f8b1d7c40" \
  -d '{
    "type": "webhook",
    "url": "https://crm.acme-corp.io/omnidim-callback",
    "include": ["summary", "extracted_variables"],
    "extracted_variables": [
      { "key": "outcome", "prompt": "Did the caller book an appointment?" }
    ],
    "trigger_call_statuses": ["completed", "failed"]
  }'

{
  "action": {
    "id": 63536,
    "type": "webhook",
    "url": "https://crm.acme-corp.io/omnidim-callback",
    "include": ["summary", "extracted_variables"],
    "extracted_variables": [
      { "key": "outcome", "prompt": "Did the caller book an appointment?" }
    ],
    "trigger_call_statuses": ["completed", "failed"]
  },
  "applies_after_publish": false
}

curl "https://backend.omnidim.io/api/v1/integrations" \
  -H "Authorization: Bearer YOUR_API_KEY"

curl -X POST "https://backend.omnidim.io/api/v1/agents/158910/post-call-actions" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "type": "google_sheets",
    "integration_id": 812,
    "spreadsheet_id": "1BxiMVs0XRA5nFMdKvBdBZjgmUUqptlbs74OgvE2upms",
    "worksheet_name": "Calls",
    "include": ["summary", "extracted_variables"],
    "extracted_variables": [
      { "key": "customer_name", "prompt": "The caller'\''s full name." }
    ]
  }'

curl -X PATCH "https://backend.omnidim.io/api/v1/agents/158910/post-call-actions/63536" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "url": "https://crm.acme-corp.io/omnidim-callback-v2" }'

curl -X DELETE "https://backend.omnidim.io/api/v1/agents/158910/post-call-actions/63536" \
  -H "Authorization: Bearer YOUR_API_KEY"

What a webhook receives

After each call, a webhook action sends one POST to your url, with a JSON body like this:

{
  "call_id": 7651752,
  "call_sid": "CA4b1f9e2d7c...",
  "call_request_id": 91422,
  "bot_id": 158910,
  "bot_name": "Front desk",
  "call_direction": "outbound",
  "from_number": "+15551230000",
  "to_number": "+15551234567",
  "phone_number": "+15551230000",
  "call_date": "2026-10-16 14:02:11",
  "start_time": "2026-10-16 14:02:11",
  "end_time": "2026-10-16 14:04:37",
  "call_duration": 146,
  "call_status": "completed",
  "hangup_source": "user",
  "hangup_reason": "The caller hung up.",
  "user_email": "demo@example.com",
  "recording_url": "https://backend.omnidim.io/...",
  "recording_available_at": "2026-10-16 14:06:07",
  "low_interaction": false,
  "metadata": { "lead_id": "L-1042" },
  "call_report": {
    "summary": "The caller booked a cleaning for Friday at 10 am.",
    "sentiment": "Positive",
    "extracted_variables": { "outcome": "booked" },
    "full_conversation": "...",
    "recording_url": "https://backend.omnidim.io/...",
    "interactions": [
      {
        "sequence": 1,
        "user_query": "Hi, I want to book a cleaning.",
        "bot_response": "Sure. What day works for you?",
        "time": "2026-10-16 14:02:15"
      }
    ]
  }
}

Call fields

Every call field above is always in the body, but some can be empty. On a call with no recording or no hang-up details, such as most no-answer and busy calls:

• recording_url, hangup_source and hangup_reason are false.
• recording_available_at is null.

Treat both false and null as "no value".

Fields added only in certain cases

These keys are left out of the body entirely unless their condition is true.

The call report

call_report is always in the body. Its keys depend on include and on what the call produced:

A call nobody answered has no conversation, so it has no summary, sentiment, extracted_variables or full_conversation, even if you asked for them. Check that a key exists before you read it.

Delivery and retries

• Your server should answer with any 2xx status. Do the heavy work after you answer, because a request is given up after 30 seconds.
• If your server is down, times out, or answers 408, 425, 429, 500, 502, 503 or 504, the webhook is sent again, up to 4 more times, with growing waits between tries.
• Any other answer, such as 400 or 404, is final. It is not retried, so fix the endpoint and the next call will deliver.
• If your server fails several times in a row, deliveries to it pause for a short while, then resume on their own. They are kept and sent, not dropped, for up to about 20 minutes.
• Retries can send the same call twice. Use call_id to ignore a call you already processed.

Action types

A workflow action starts the workflow's published version with the call's details and transcript, plus the summary, sentiment and extracted variables when the call produced them. To create and publish a workflow from your code, see Build with the API. Its List workflows call gives you the workflow_id.

Every type also accepts three options:

• include: what to send. Pick from summary, extracted_variables, fullConversation and sentiment. Zoho Bigin, workflow and WhatsApp always receive extracted variables. See How extracted variables work.
• extracted_variables: details for the agent to pull out of the conversation, each a key and a prompt. Salesforce and HubSpot send a fixed set of fields, so they take none.
• trigger_call_statuses: which call outcomes run the action. If you leave it out, the action runs on completed and voicemail_detected.

Every field is listed in the Add a post-call action reference.

When a change goes live

Most agents: right away. The next call uses it, and every response says "applies_after_publish": false.

Agents you publish in the dashboard: calls use the published version. The response says "applies_after_publish": true, and your change goes live the next time you publish. See Agent versioning and publishing.

Finding your action again

An action's id can change. Saving post-call changes in the dashboard, or restoring an older version, gives every action a new id.

So don't keep the id. When you need your action again, list the actions and match it by what it points at:

• a webhook by its url
• an email by its recipients
• Sheets, Slack or a CRM by its integration_id

Your actions are safe from the dashboard, too. If someone has the agent open in the dashboard while you add an action, their next save keeps yours. Their own post-call change in that save is not applied: the dashboard shows the current list and asks them to make it again.

How extracted variables work

An extracted variable is a detail the agent pulls out of the conversation, like customer_name or appointment_date. Each one is a key (the name you get back) and a prompt (what the agent should look for). An action can have as many as you need.

One extraction per call. After the call, the agent reads the conversation once and fills in the variables of all the agent's actions together. Adding more actions doesn't add more work per call.

Each action gets its own keys. The results are then split by action:

Same key, same prompt. One key has one prompt per agent, because it is extracted only once. If your webhook asks for outcome and a Slack action asks for outcome too, both prompts must match. A new action that reuses a key with a different prompt is refused with a 400, and nothing changes. Give it a new key instead, such as slack_outcome.

Different keys, different prompts. Keys only need to differ from each other. You can give the webhook budget and the email follow_up_date, each with its own prompt, and each action receives its own.

Errors

Every error comes back as { "error": "...", "error_description": "..." }, and the description tells you what to fix.

Related