This guide walks you from a fresh signup to your first live phone call answered by an Aitelier agent. Plan to spend about 15 minutes the first time. Once you know the flow, a new project takes 5.
1. Create your account
Go to app.beta.aitelier.org and sign in with your email. We send a 6-digit code — paste it and you are in. No password to remember; the code expires after 10 minutes.
The first time you sign in, the platform creates a Tenant for you — your billing and isolation boundary. Everything you create from this account belongs to that tenant. New tenants start on a free trial: 14 days and up to 200 conversations, no card required.
2. Create your first Project
A Project represents one business with one voice agent. You can have many projects under one account — for example, an agency running agents for several clients.
Click New project on the dashboard. The platform offers two starts:
- I have a website — paste the URL of your business. The platform crawls the public pages, extracts your business category, hours, services and contact info, and pre-fills the agent. This is the fastest path and what most customers pick.
- I'll describe it — tell the platform about your business in a few sentences instead. Use this if you do not have a website yet.
If you want full manual control, there is also a link to create an empty project and configure everything yourself.
3. Tell the platform about your goals
After the project is created, an AI Copilot walks you through a short goals dialog. You describe what you want the agent to do — "book tables for my restaurant", "qualify inbound sales leads", "route citizens to the right office" — and the Copilot proposes a complete starting setup:
- One or more Goals with success criteria
- A matching Industry Pack (Restaurants / Sales / Real estate / Government)
- A draft Scenario with the right slots and prompt
Review the proposal. Apply it if it fits, or hit Refine and tell the Copilot what to change — it revises and proposes again. When you apply, you land on the project's main page.
4. Configure connections
Open the Connections tab. It holds your industry pack, your knowledge base and your plugin connections. Connect the integrations your scenario uses — a Google Calendar for meetings, a webhook into your booking or CRM system, a Telegram or WhatsApp channel for text conversations.
Plugins are optional in the strict sense — without them the agent still captures leads and structured slot values in the platform's own session log — but having them turns calls into calendar entries, bookings and CRM rows automatically.
5. Add your business knowledge
In the Knowledge base panel, you add anything the agent should ground its answers in: a menu, a price list, service descriptions, FAQ, special policies. The platform indexes the content; the agent retrieves from it at runtime instead of guessing.
The website crawl from step 2 already populated the knowledge base. You can add more — upload PDFs (menus, datasheets, brochures), or re-crawl when your site changes. A daily auto-refresh keeps crawled content current.
Read the knowledge base guide.
6. Run a test call
On the project Overview, find the Test call card. Allow microphone access in your browser. The platform places you in a WebRTC session with the agent, running the current scenario. You can interrupt, ask awkward questions, simulate a difficult caller. Every test call is recorded and playable afterwards, and test calls are free during your trial.
When the agent does something you do not like:
- Open the session record — the full transcript and the tool-call trace are there
- Find the point where it took a wrong turn
- Use Discuss with AI on the session, or ask the Copilot to adjust the goal or the prompt
- Run another test call to confirm the fix
This loop is the main day-to-day work after the initial setup. Most scenarios reach production-quality behaviour in a few iterations.
7. Connect a phone number
Open the project Settings tab. Two options:
- Bring your own SIP trunk — point your existing trunk (Twilio, your carrier, your IP-PBX) at the platform. You can place a test call against the trunk to validate the call path before going live.
- Use a number we provide — we provision a phone number for you. Reach out via your workspace or the contact page and we set it up for your country.
Once a number is attached, every inbound call to that number reaches your agent.
8. Trigger an outbound call (optional)
If your project needs outbound calls — appointment reminders, lead follow-up — you trigger them through the API:
- Single call —
POST /v1/callswith a scenario, a phone number and optional per-call parameters. - Batch — upload a CSV of contacts via
POST /v1/calls/bulk. The platform paces the dial-outs and records every result; you poll the batch status through the API.
9. Watch your first real calls
Every call lands in the Calls tab as a full session record: audio, transcript, structured slot values, the tool-call trace, the outcome (success / real conversation / spam) and a Commercial Intent Score. If you have not set up webhooks, the platform emails a per-call summary to your project's notification address, so nothing slips past you.
Most owners read every session in the first week to learn what the agent handles well and what it does not. For programmatic notification — a booking landed, a call ended — subscribe to webhooks.
What you have now
You have:
- A live voice agent answering inbound calls
- A test-call loop for ongoing iteration
- A knowledge base it grounds answers in
- Connections to your calendar / booking / CRM / messengers
- A free trial, then a pay-as-you-go wallet (top up when ready)
What you do next depends on what you want to grow:
- More volume — add scenarios for other call types, raise the duration cap, add outbound batches.
- More polish — iterate on test calls, refine the goal definitions with the Copilot, tighten the prompts.
- More automation — wire up webhooks so a new booking triggers your downstream systems.
The rest of the docs cover each of these in detail. The Glossary is a useful tab to keep open while reading.