Quickstart
By the end of this page you will have talked to an agent you designed, twice: once from your browser, once over an actual phone line. Along the way you’ll meet every object in Talkif — flow, number, call, credits — in the order you’ll use them every day.
The dashboard shows the same five steps as a checklist on your home screen until you’ve completed them. This page is the long version, with the why for each one.
Before you start
- An account. Sign up at app.talkif.ai with email + password or Continue with Google. Email sign-ups get a verification link (valid 24 hours) and an SMS code for the phone number you register. Both must be confirmed before the account can place calls.
- Credits. New accounts receive a $5.00 welcome credit. It’s enough to build, test in the browser, and take a few short calls.
Welcome credit is a bonus balance. Bonus and promotional balances pay for calls but are never used for phone-number purchases or monthly rental. To buy a number in step 4 you’ll add a payment method and top up first — the number itself costs a dollar or two a month, and the top-up minimum is shown on the Billing page. Everything before step 4 works on the welcome credit alone.
What you’re going to build
A single-agent support line: it greets the caller, asks what they need, answers from the instructions you give it, and ends the call politely. One agent, one flow — the same structure scales to multi-agent flows later without redoing anything here.
Create a flow and give the agent its instructions
Open Flow Builder in the sidebar and choose Create Flow. You can start from a template or from an empty canvas; for this walkthrough start empty so you see every piece.
The canvas opens with a Main Agent already placed. It’s the entry point of every call and the only node that carries the call’s persona. Select it and fill in the two prompts:
-
System Prompt — who the agent is, for the whole call. Keep it to identity, tone, and hard limits:
System Prompt -
Agent Prompt — what this agent must accomplish while the conversation is on it:
Agent Prompt
Prompts are instructions, not scripts. Tell the agent what to achieve and what to collect; let it choose the words. Scripted line-by-line prompts sound robotic and break the moment a caller says something unexpected. Prompts can also use {{contact.first_name}}-style variables that are filled in when the call starts — see Write prompts.
Attach the three providers
An agent can’t hear, think, or speak until a provider node is connected for each job. Add an STT, an LLM, and a TTS node to the canvas and connect each one to the Main Agent.
The live list of vendors and models is in Models and voices. Publishing is refused if any of the three is missing, so the builder flags it early.
Talk to it in your browser
Open the Flow Tester from the builder toolbar and click the phone button. Allow microphone access when the browser asks. You’re now on a WebRTC call with the draft you have open — unsaved edits included — so you can iterate on prompts without publishing.
Say “Hi, what time do you close on Friday?” and then “I’d like to book a cleaning.” Watch the transcript, the per-turn latency (LLM TTFB) and token counts on the right; those numbers tell you whether a model is fast enough for phone use before a real caller finds out.
Test calls use the same STT/LLM/TTS as a real call and are billed the same way, at the same rates. They need an available balance of at least $1.00 to start — the welcome credit covers this.
Adjust the prompts until the conversation feels right, then hang up with the red button.
Publish the flow
Click Publish. Talkif validates the flow and refuses to publish until it passes. The checks that matter most:
- exactly one Main Agent, and it’s the entry node
- the Main Agent has a System Prompt; every agent has an Agent Prompt
- an LLM, an STT and a TTS provider are configured, with models the platform currently offers
- every transition points at an agent that exists; agent and function names are unique
Publishing creates an immutable version. The draft stays editable; phone numbers and API calls use the published version until you publish again. Versioning, rollback and pinning are explained in Flows and versions.
Get a phone number
Open Phone Numbers in the sidebar and choose Purchase Number. Pick a country, then search by area code or digit pattern. Each result shows its capabilities (voice, SMS) and its monthly price. Confirm to buy; the number is provisioned instantly.
Purchases and rental are paid from your purchased balance — add a payment method and top up under Billing if you haven’t. Numbers renew monthly against the same balance; see Buy a number for pricing and what happens if renewal fails.
Dashboard
API
Phone Numbers → Purchase Number, choose country → search → buy.
Already have numbers with a SIP carrier or a WhatsApp Business account? Skip the purchase and connect them instead — Bring your own SIP trunk, WhatsApp Business Calling.
Connect the flow to the number
Open the number you just bought and connect your published flow to it. From this moment, anyone who dials the number is answered by the agent — this is what makes a number inbound-capable.
Dashboard
API
Phone Numbers → select the number → Select a published flow. Leave the version on latest published unless you have a reason to pin.
A number is connected to at most one flow. If someone dials a number with no flow connected, Talkif answers busy and records a failed call in your history with the reason “no connected flow” — so a misconfigured number is visible, not silent. The same busy response is returned when your balance is below the $1.00 minimum.
Make a real call
Two ways, and it’s worth doing both:
Inbound — dial the number from your own phone. This is the path your customers will take.
Outbound — have the agent call you. From the dashboard use Start a call on the home screen (or Call Center → Calls), pick the flow, the number to call from, and your mobile as the destination. From the API, the same thing is one request:
Dashboard
API
Home screen → Start a call → flow, from-number, destination → Call.
Read the result
Open Call Center → History and select the call. You get three things for every call, test or real:
- the transcript, turn by turn, with who spoke and when
- the recording, playable in place
- the cost breakdown — telephony seconds, STT seconds, LLM tokens, TTS characters — each priced, summed to the amount taken from your balance
Costs are where model choice becomes visible: swap the LLM or TTS in the draft, run the same conversation in the Flow Tester, and compare. Cost breakdown explains every line.
You now have
A published flow, a live number connected to it, and at least one completed call with a transcript and a cost. Everything else in Talkif is a way to do this at scale or with more precision:
What happened between “Call” and the transcript — statuses, queueing, end reasons, billing.
Split the job across agents, add transitions, and hand off cleanly.
Let the agent look things up or take action in your systems mid-call.
Call a list of contacts with one flow, with pacing, retries and analytics.