AI Calling API: Trigger AI Phone Calls From Your CRM

The lead who filled your form two minutes ago is still looking at their phone. Two hours later they have spoken to three other companies. Most teams know this, which is why fast follow-up and AI callbacks have become standard.

But if the moment that should trigger a call lives in your system (your CRM, your booking app, your loan origination tool) you don’t want a second dashboard. You want your system to say “call this person now, with this agent”, and to hear back when the call is over. That is what an AI calling API is for.

When an API Beats a No-Code Dashboard

For most teams, starting AI calls from a dashboard is enough: upload a list, pick an agent, press start. Reach for the API when one of these is true.

• The trigger lives in your own software. A form on your site, a stage change in your CRM, an abandoned checkout, an EMI that is due tomorrow. The API lets that event place the call within seconds, with no person in between.
• The result has to land back in your system. If your team works in your CRM, the call outcome should appear there, not in a tool they never open.
• You run calling at volume on a schedule. Creating campaigns, adding contacts and starting them from your own jobs beats a weekly spreadsheet upload.
• You are building a product. Agencies and software companies use an AI calling API to offer calling inside their own product, without building speech, telephony and an LLM stack themselves.

Eight Things to Check in Any AI Calling API in India

Global voice AI APIs are built for US numbers and English. An Indian deployment fails on different things. Ask every vendor these eight questions before you write code.

What Sahay’s AI Calling API Covers

The API is a REST API over HTTPS with JSON bodies. It covers the four things you need to run AI calling from your own system.

• Agents. List your AI agents and the voices available, edit an agent’s draft script, publish it as a new version, and roll back to an earlier version.
• Calls. Start one call to one number, list and filter calls by status, agent or date, fetch one call’s details, and fetch its recording.
• Campaigns. Create a calling campaign, add contacts with a name and your own variables, then start, pause, resume or stop it, and read each contact’s status.
• Webhooks. Register one HTTPS URL and pick the events you want: call.completed for every call, campaign.contact.completed as each campaign contact finishes, and campaign.completed when a whole campaign ends. Your system never has to poll.

Every endpoint, field, error code and webhook payload is documented in the Sahay API reference at docs.sahay.io, with copy-paste examples in cURL, Python and Node.

Start an AI Call With One Request

Authenticate with an API key in the Authorization: Bearer header (or X-API-Key). Send the agent to use and the lead’s number in E.164 format. Anything you put in metadata, such as your own lead ID, comes back to you on the call and its webhook, so you can match the result to your record.

curl https://sales.sahay.io/api/public/v1/calls \
  -H "Authorization: Bearer $SAHAY_API_KEY" \
  -H "Idempotency-Key: lead-48213-first-call" \
  -H "Content-Type: application/json" \
  -d '{
    "agent_id": "YOUR_AGENT_ID",
    "to_number": "+919876543210",
    "metadata": { "crm_lead_id": "48213", "source": "website-form" }
  }'

The API answers 202 Accepted with the new call and its status straight away. The call itself happens in the background. If the agent has no published version, its number is paused, or the plan has no AI voice minutes left, you get a clear error code (agent_not_published, agent_number_paused, insufficient_voice_minutes) instead of a silent failure.

Get the Result Back With a Signed Webhook

When the call ends, Sahay POSTs a call.completed event to your webhook URL. It carries the call ID, the agent, the number, the final status, the duration, the end time and your metadata. Every delivery has a Sahay-Signature header of the form t=<unix time>,v1=<HMAC-SHA256>, computed over the timestamp and the raw body with your webhook secret. Check it before you trust the event:

import hmac, hashlib, time

def verify(secret: str, header: str, raw_body: str, tolerance=300) -> bool:
    # header looks like: t=1762588800,v1=5f2b...
    parts = dict(p.split("=", 1) for p in header.split(","))
    t, sig = parts.get("t"), parts.get("v1")
    if not t or not sig or abs(time.time() - int(t)) > tolerance:
        return False
    expected = hmac.new(secret.encode(), f"{t}.{raw_body}".encode(),
                        hashlib.sha256).hexdigest()
    return hmac.compare_digest(expected, sig)

Limits, So Nothing Surprises You

• Requests: 30 a minute per account by default, raised on request for higher volumes. Every response tells you where you stand in X-RateLimit-Limit, X-RateLimit-Remaining and X-RateLimit-Reset.
• Calls in progress: 2 live API calls at a time by default, also raised for larger teams. Campaigns are the better tool for calling a long list.
• Keys: up to 10 active keys per account, so each system gets its own key and you can revoke one without touching the others.

The API docs list every error code and the rate-limit headers in full.

Three Patterns Teams Build First

1. Speed-to-lead. Your website form handler calls POST /calls the moment a lead submits. The AI qualifies them while they are still interested, and the webhook moves the lead to the right stage in your CRM.
2. Reminders and renewals. A nightly job creates a campaign with everyone whose payment, appointment or renewal is due, adds their details as contact variables, and starts it at 10 am.
3. Re-engaging cold leads. Pull leads that went quiet 30 days ago from your CRM, run a short check-in campaign, and route the ones who say yes to a person.

If your team doesn’t have its own CRM yet, you don’t need the API at all: Sahay’s own CRM starts AI callbacks for new leads on its own. Our AI CRM comparison explains how, and how Sahay compares with other AI calling agents covers the product side.

Frequently Asked Questions