עבריתLog in

Use cases

What AI agents do with OneLi, with request and result examples.

Each scenario shows what the agent sends, where a consultation may happen and what comes back. Examples are trimmed: a real request also needs idempotency_key and to_number.

Personal assistant books a doctor, garage or hairdresser

Situation
The user writes “book my car in for its 10,000 km service this week”. The agent knows from the calendar when the user is free.
Mid-call consultation
The garage only offers a time outside the given slots, or a price above the cap. The voice agent asks; your agent checks the calendar or asks the user.
What comes back
outcome goal_achieved with the slot in the summary. The agent adds it to the calendar.
POST /v1/call-tasks
{
  "to_number": "03-XXXXXXX",
  "goal": "לקבוע טיפול 10,000 לרכב השבוע",
  "instructions": "חלונות פנויים: ג׳ 08:00-10:00, ה׳ אחרי 16:00. לשאול כמה זמן הטיפול לוקח.",
  "constraints": { "max_price_ils": 900, "fallback": "לאסוף את המועדים הפנויים ולסיים בנימוס" }
}

Restaurant booking with fallback times

Situation
“Book us a table for four on Friday evening.”
Mid-call consultation
20:00 is full and they offer 18:30 or 21:30. The voice agent asks which one works.
What comes back
A confirmed booking, or needs_user_input with the alternatives offered.
POST /v1/call-tasks
{
  "goal": "להזמין שולחן ל-4 סועדים ביום שישי ב-20:00",
  "instructions": "אם 20:00 תפוס, לשאול על 19:30 עד 21:00. שם להזמנה: דנה."
}

Comparing quotes from three suppliers

Situation
“Get a price for moving a 3-room flat from Tel Aviv to Haifa.” The agent creates three tasks, one per supplier, with the same goal.
Mid-call consultation
Usually none: this is an inquiry. A consultation only happens if a supplier asks for a commitment.
What comes back
Three summaries. The agent compares them and shows the user a table and a recommendation.
Node.js
// one task per supplier, each with its own idempotency_key
for (const s of suppliers) await oneli("POST", "/v1/call-tasks", {
  idempotency_key: `move-quote-${s.id}`,
  to_number: s.phone,
  goal: "לקבל הצעת מחיר להובלת דירת 3 חדרים מתל אביב לחיפה ב-15 בחודש",
  instructions: "לא להתחייב. לשאול מה כלול במחיר והאם יש מע״מ.",
});

Following up on an order or delivery

Situation
The order is late. The agent calls the shop with the order number.
Mid-call consultation
The shop offers a refund instead of delivery. The voice agent asks before agreeing.
What comes back
An updated delivery date in the summary, or a decision handed back to the user.
Chat prompt (MCP)
״תתקשר לחנות ותברר מה קורה עם הזמנה 58213 שהייתה אמורה להגיע אתמול״

Rescheduling or cancelling an appointment

Situation
“Cancel Monday’s dentist appointment and ask for a slot next week.”
Mid-call consultation
The only new slot is in the morning. The voice agent asks whether to accept.
What comes back
Old slot cancelled and a new one booked, or only the cancellation with needs_user_input.
POST /v1/call-tasks
{
  "goal": "לבטל את התור של יעל כהן ביום שני ב-10:00 ולקבוע תור חדש בשבוע הבא",
  "instructions": "מועד חדש רק אחרי 15:00. אם אין, לבטל בכל זאת ולשאול מתי כדאי לחזור."
}

Checking availability or stock in a store

Situation
Before driving over, the agent checks whether an item is in stock at the nearest branch.
Mid-call consultation
Only a similar model is available. The voice agent asks whether to have it put aside.
What comes back
Yes or no, price and opening hours in the summary.
Chat prompt (MCP)
״תברר אם יש בסניף בהרצליה מקדחה בוש GSB 18V במלאי, ועד מתי פתוח היום״

B2B: confirming meetings with leads the day before

Situation
Your sales system creates one task per meeting tomorrow, with the client’s name and time.
Mid-call consultation
The client asks to move it. The voice agent asks; your webhook checks the rep’s calendar and answers approve or ask_user.
What comes back
Confirmed, moved (new time in the summary) or cancelled. The result goes back to the CRM.
POST /v1/call-tasks
{
  "idempotency_key": "meeting-8812-confirm",
  "goal": "לאשר את הפגישה עם משה מחר ב-11:00 במשרדי החברה",
  "instructions": "אם צריך להזיז, להציע מחר 14:00-17:00 או ביום ה׳ בבוקר.",
  "dynamic_variables": { "opening_line": "שלום משה, אני עוזרת דיגיטלית מטעם צוות המכירות, מתקשרת לגבי הפגישה של מחר." }
}

Operations: chasing a contractor for a status update

Situation
Work on site is running late. The ops agent calls the contractor to ask when it will be finished and what is missing.
Mid-call consultation
The contractor asks for extra payment. The voice agent asks; answering ask_user hands the decision to the person in charge.
What comes back
An estimated finish date and a list of what is missing, for the project system.
POST /v1/call-tasks
{
  "goal": "לברר מתי תסתיים עבודת האינסטלציה בדירה ברחוב הרצל 12 ומה מעכב",
  "instructions": "לא לאשר תוספות תשלום. לבקש תאריך סיום מחייב."
}

Support agent escalates to a phone call with a business

Situation
Your product’s support agent cannot resolve something in writing with a vendor (a lab, a courier). It opens a call task with the full context.
Mid-call consultation
The vendor offers a fix that needs approval. The decision reaches the support agent, which answers by your policy.
What comes back
A resolution or a ticket number in the summary, which the agent passes on to the customer.
POST /v1/call-tasks
{
  "goal": "לברר למה משלוח 7741 עדיין לא נאסף ולקבל מועד איסוף חדש",
  "instructions": "מספר לקוח 220431. לבקש מספר פנייה.",
  "max_duration_s": 300
}

Automation: new form lead → confirmation call → write back to CRM

Situation
In n8n or Make: new form → a call task that confirms the details and agrees a time for a follow-up → wait → read the result → update the CRM.
Mid-call consultation
Usually none. A webhook can answer by fixed rules.
What comes back
outcome and summary are written to the lead record.
POST /v1/call-tasks
{
  "idempotency_key": "lead-{{ lead.id }}",
  "to_number": "{{ lead.phone }}",
  "goal": "לאשר שהפנייה על {{ lead.topic }} רלוונטית ולתאם שעה לשיחה עם נציג",
  "instructions": "להציג את עצמך כעוזרת דיגיטלית של העסק. אם לא מעוניינים, להודות ולסיים."
}

What OneLi does not do

  • No inbound calls. Every call is outbound, started by your agent.
  • Israeli numbers only: mobile 05X, landline 02/03/04/08/09 and 07X.
  • Not for marketing, telemarketing or cold calls. The person called can ask not to be called again, and that applies to every account. There is also a daily limit per number.
  • The voice agent always introduces itself as a digital assistant, even with a custom opening_line.
  • No live audio API. You get the transcript, events and a summary.