Skip to navigation

Build agents

Design the conversation, give it a voice and a brain, connect it to your systems, and prove it works before it goes live.

A flow is where a Talkif agent is designed. You work on a canvas: agent nodes hold the instructions, provider nodes give the agents hearing, thinking and speech, function nodes let them act — and edges say what is connected to what. There is no code, but there is real engineering: prompt quality, model choice and hand-off design decide whether callers get a good experience.

These pages are in the order you’ll need them.

Design a conversation

Main Agent and sub-agents, transitions between them, and how to split a job so each prompt stays short.

Write prompts

System Prompt vs Agent Prompt, what works on a phone line, dynamic {{variables}}, and the Prompts workspace.

Choose models and voices

STT, LLM and TTS providers — how to trade latency, quality and cost, and every setting each one exposes.

Call your backend

Flow functions: let the agent look things up or take action in your systems mid-call, with signed requests.

Transfer and end calls

Ending well, scripted lines, silence handling, maximum length, and voicemail.

Transfer calls to people

Hand a live call to a person — warm or cold, to numbers, teammates or teams.

Test and publish

The Flow Tester, what validation checks, and what publishing changes.

Templates

Five ready-made flows to start from instead of an empty canvas.

Not sure how flows, versions and phone numbers relate? Read Flows and versions first — it’s five minutes and it answers most “why isn’t it live?” questions.