Copy page

Guides

Production Checklist

Use this checklist before going live. Each section covers a critical area — skip nothing. Production issues are almost always caused by a missed step here.

Agent Configuration

• System prompt tested — conversation flow verified with test calls
• Brain settings finalized — LLM provider, model, and temperature locked in
• Voice previewed — TTS provider, voice model, and speaking rate sound natural
• Transcriber configured — STT provider and language set for your callers
• Conversation settings tuned — silence timeout, max duration, interruption handling
• First message set — the greeting callers hear when the agent picks up
• Tools tested — all MCP tools respond correctly and handle errors gracefully
• Transfer targets configured — warm/blind transfer destinations and fallback numbers
• Knowledge base loaded — all reference documents uploaded and indexed
• Post-call analytics enabled — success evaluation rubric configured if needed

Verify via API

Code
curl https://api.neuratel.ai/v1/agents/agent_xyz789 \
  -H "Authorization: Bearer nk_live_your_api_key"

Check that the response includes:

• brain.provider and brain.model set to production values
• voice.provider and voice.voice_id configured
• transcriber.provider and transcriber.language set
• conversation.first_message is not empty
• tools array contains your expected tool definitions
• analytics.enabled is true if you want post-call analysis

Phone Numbers

• Numbers in your account — at least one number provisioned by the Neuratel team
• Numbers assigned to agents — each live number has an agent_id
• Numbers active — all numbers show is_active: true
• Outbound caller ID set — if making outbound calls, a number is configured as caller ID

Verify via API

Code
curl https://api.neuratel.ai/v1/numbers \
  -H "Authorization: Bearer nk_live_your_api_key"

Every number you plan to use should have agent_id set and is_active: true.

Webhooks

• Endpoint live — your webhook URL is publicly accessible
• HTTPS only — webhook URL uses https://
• Signature verification implemented — your server validates X-Webhook-Signature on every request
• Response under timeout — endpoint returns 200 within the configured timeout (default 30s)
• Idempotency implemented — duplicate deliveries are handled gracefully using event_id
• Test webhook sent — POST /v1/webhooks/{id}/test returned a successful delivery

Test Signature Verification

Code
import hmac
import hashlib

def verify_signature(body: bytes, signature: str, secret: str) -> bool:
    expected = hmac.new(secret.encode(), body, hashlib.sha256).hexdigest()
    return hmac.compare_digest(expected, signature)

Code
const crypto = require("crypto");

function verifySignature(body, signature, secret) {
  const expected = crypto
    .createHmac("sha256", secret)
    .update(body)
    .digest("hex");
  return crypto.timingSafeEqual(
    Buffer.from(signature),
    Buffer.from(expected)
  );
}

Verify via API

Code
# List webhooks and check status
curl https://api.neuratel.ai/v1/webhooks \
  -H "Authorization: Bearer nk_live_your_api_key"

# Send a test event
curl -X POST https://api.neuratel.ai/v1/webhooks/wh_abc123/test \
  -H "Authorization: Bearer nk_live_your_api_key"

# Check delivery logs
curl https://api.neuratel.ai/v1/webhooks/wh_abc123/logs \
  -H "Authorization: Bearer nk_live_your_api_key"

API Keys

• Keys created — at least one nk_live_ key exists for each service
• Scoped appropriately — each key has only the permissions its service needs
• Stored securely — keys are in environment variables or a secrets manager, never in code

Review your keys in Settings → API Keys and check scopes on each key.

Account Readiness

• Confirm with your account team that your account is provisioned for production traffic and that any low-balance signal destinations are correctly wired

You can sanity-check your live balance and usage at any time — see Usage Visibility. Anything beyond reading those values goes through your account team.

Testing

• Inbound test calls made — called each production number and verified agent answers correctly
• Outbound test calls made — if using campaigns or API-initiated calls, verified they connect
• Tool calls tested — triggered each MCP tool during a test call and verified correct behavior
• Transfer tested — verified warm and blind transfers connect to the right destination
• Edge cases covered — tested silence, interruptions, long pauses, and unusual inputs
• Webhook delivery confirmed — checked webhook logs for successful deliveries after test calls

Monitoring

• Webhook logs reviewed — no consistent failures in delivery history
• Call logs reviewed — test calls show expected duration, outcome, and transcript
• Analytics dashboard checked — data appearing correctly for test calls
• Alert thresholds set — balance alerts and error rate monitoring configured

Check Recent Calls

Code
curl "https://api.neuratel.ai/v1/voice-sessions?limit=10&sort=-created_at" \
  -H "Authorization: Bearer nk_live_your_api_key"

Review the last 10 calls for expected outcomes and durations.

Compliance

• Recording consent — if recording calls, agent states consent notice at start of call (required in many jurisdictions)
• DNC list loaded — Do Not Call numbers excluded from outbound campaigns
• Data retention configured — retention periods match your data processing agreements
• GDPR contact — deletion request process documented if serving EU callers

Quick Verification Script

Run all checks in one pass:

Code
API_KEY="nk_live_your_api_key"
BASE="https://api.neuratel.ai/v1"

echo "=== Agent ==="
curl -s "$BASE/agents" -H "Authorization: Bearer $API_KEY" | jq '.items[] | {name, status: .is_active}'

echo "=== Numbers ==="
curl -s "$BASE/numbers" -H "Authorization: Bearer $API_KEY" | jq '.items[] | {did, agent_id, is_active}'

echo "=== Webhooks ==="
curl -s "$BASE/webhooks" -H "Authorization: Bearer $API_KEY" | jq '.items[] | {url, is_active}'

echo "=== Balance ==="
curl -s "$BASE/billing/balance" -H "Authorization: Bearer $API_KEY" | jq '{has_credits}'

Post-Launch

After going live, monitor these daily for the first week:

• Call success rate and average duration
• Webhook delivery success rate
• Credit balance consumption rate
• Agent error logs

Next Steps

• Troubleshooting — solutions when things go wrong
• Billing and Usage — monitor ongoing costs
• Security — ongoing security best practices