Let Flipa test it
It throws three requests at it, two ordinary ones and a hard one it makes up, and tells you for each whether the result holds up and why. It takes half a minute.
AI agents
AI agents for small business: just explain the job
AI agents for small business tend to sound like forms and settings, or like a bot that phones your customers. Flipa's are neither. They work inside your team and start the way a new hire starts: you say what the job is, the agent asks what it doesn't know, and before you trust it, you watch it work. It drafts; someone on your team sends. It doesn't make calls, send email, or post anything.
The example on this page is a made-up language school. The app's interface is in Spanish today.
I need an agent that drafts replies to people who write in asking about our classes
OK. How you talk and which classes you offer I already know from your documents, so I won't ask you that. I'm missing two things: where people write to you, and whether it can quote prices.
Classes 2026-27.pdfTone guide
Watch it
Build an AI agent without coding, from request to proposal
The animation recreates the real screen, with what Flipa asked and proposed in a test account. There's no canvas of boxes and arrows: flowchart tools are a good fit if you think in flowcharts, and this no-code AI agent builder is a conversation instead. If you've already spent a while explaining something, after three messages a button offers to turn the conversation into an agent.
- Before it asks, it checks what it already knows about your company. If it knows your tone, it doesn't ask about it.
- It asks with options to click. You can always type something else or skip the question.
- The proposal says what the agent is for and when it'll step in, brings sample questions, and shows underneath the credits it took to prepare.
The interview
What it asks you, and why it asks
One at a time, three at most, and you can skip any of them. What you don't answer isn't left hanging: Flipa decides it and shows you, so you can change it. Walk through it with the language school's agent.
I need an agent that drafts replies to people who ask about our classes
Before asking
First it checks what it already knows about the school
- Already knowHow you talk: friendly, no jargon Tone guide
- Already knowWhich classes you offer, with levels and schedules Classes 2026-27.pdf
- Already knowThat you never promise a score on an official exam What we can't say
Question 1 of 3
How do people who ask about classes write to you?
Question 2 of 3
When someone asks what a class costs, what should it do?
Question 3 of 3
Who signs the replies?
The proposal
New agent: Class info
- What it's for
- Drafts the reply to anyone who writes in asking about classes, levels, and schedules.
- When it'll step in
- When someone pastes an email asking about classes, levels, or schedules.
- What you told me
- Nothing yet: you got here without answering.
After it's created
It steps in when it should, and tells you it was the one
You don't have to remember which agents exist. The one that fits the request steps in, one at most, and underneath you see who answered. If you'd rather choose, you pick from the who-answers menu. It's the same in the Chrome side panel, which is in testing.
- Four come built in: SEO, Social media, Legal and claims, and Customer support. That last one writes drafts; your team sends them.
- Each one has its own face. Yours do too.
- Your brand's rules outrank any agent's instructions.
The test
Before you trust it, you test it
Nobody gets the keys on day one. An agent doesn't get saved before you've seen it work, either.
Test it yourself
You chat with the draft without saving it. Paste in that email your team always struggles to answer.
Fix it by talking
Tell it what to change and it shows you only the difference: red for what comes out, green for what goes in. You can undo it.
Go back
Every change keeps the previous version. If yesterday's worked better, you return to it.
Try it here
A custom AI agent in three answers
It's a mock-up, not the product, but the idea is the same: you answer and the profile fills in.
What job are you giving it?
How should it write?
What should it never do without asking you?
Agent profile
Class info
- How it talks
- Friendly and clear, like your tone guide
- Never without asking you
- Quote prices: it offers a placement test instead
- What it already knows
- Everything you've told Flipa about your company
What for
Three jobs the school used to leave to “whoever has time”
Some jobs have a ready-made starting point. Only the one for review replies is tested end to end, so that's the one we vouch for today.
Frequently asked questions
Before you hand a job to an agent
Do I need to know how to code?
No. You build an AI agent without coding and without forms: you ask in the chat, answer by clicking options, and approve the proposal with one click. If you want to adjust something by hand later, the profile is editable.
What if the agent says something it shouldn't?
Your brand's rules outrank its instructions. We tested that on purpose: an agent was ordered to promise refunds and to say the products cure illness, and it didn't. It warned that the order clashed with what the company allows itself to say. Even so, what it writes is a draft.
Does it reply to my customers itself?
No. An agent prepares the text, and someone on your team copies it and sends it. Flipa doesn't send email, doesn't post on social media, doesn't make calls, and doesn't speak with your customers.
Is the app in English?
Not yet: the interface is in Spanish today. Agents answer in the language you write in, so the drafts can be in English.
How much does it cost to create one?
It runs on credits. Creating an agent is a long job. In our tests an ordinary job used 45 to 138 credits, and creating an agent uses more, depending on how much it has to read. The proposal itself tells you what it took to prepare; in the animation on this page it was 486. It comes out of your plan's credits, and plans start at $99/mo, plus applicable taxes: see pricing.
Can anyone on the team create agents?
Today the account admin creates them. Opening it to everyone else exists as an option, off by default and still in testing. Every change is logged with who made it and when.
What if two agents fit the same request?
One steps in, never two at once. When two of your own agents overlap, the choice isn't refined yet: we're testing it. Meanwhile, you pick by hand in the who-answers menu.
Which job would you explain first?
Think of whatever your business answers the same way twenty times. For most small business owners, that's the first agent.
Start free