A review written by the people who built the thing is worth exactly as much as its worst paragraph. So here is the worst paragraph first, and then five more like it, before anything good gets said.
What we are bad at
One channel, and no plans for a second. Instagram comments and DMs. That is the whole product. No WhatsApp — it is not built, it is not shipped, and if WhatsApp carries revenue for your business then this review is over and you should buy something else. No Messenger. No Telegram.
No flow builder. There is no branching canvas, no drag-and-drop conversation designer, nothing you can screenshot for a client deck. An automation here is a trigger, some keywords, a public reply, a private reply and a few advanced toggles. If your funnel needs a conditional tree six steps deep, we cannot draw it and we will not pretend a list of fields is the same thing.
No sequences or broadcasts. We reply to people who contacted us. We do not run multi-step nurture campaigns or send scheduled messages to a list.
A short integration list. Leads live in the product and you can get them out. What we do not have is a long shelf of one-click connectors into stores, CRMs, ad platforms and email tools. If your automation must hand off into three other systems automatically, you will be doing that by hand or not at all.
No scheduling or publishing. You still need a second tool for the calendar, and we are not planning to become that tool.
A cold start on voice. The voice profile is built from replies you actually wrote. A brand-new account with no reply history has nothing to build from, so the first drafts read flat and generic — the exact thing the feature exists to avoid. It gets good after you paste in a body of your own answers, and pasting them in is work you have to do.
We refuse a lot, and some of those refusals will annoy you. Ten checks run before every send. Dedupe means the person who commented twice gets one reply. Quiet hours mean a 2am question waits. A narrow keyword means a genuine customer sometimes gets nothing from the automation at all. Tools that always send feel more responsive on the demo, and on a bad day ours will have skipped something you wanted sent.
A small ecosystem. There is no agency network trained on us, no library of community templates, no shelf of YouTube tutorials made by someone else. When you are stuck at 11pm, you are reading our docs, not a forum thread from three years ago.
The free tier is genuinely small. A hundred credits, no card. That is roughly a week for a small account. People read it as stingy and they are not entirely wrong — it is sized to answer a question, not to run a business on.
Who should not buy this
Buy something else
We might fit
The third item on that left column is the one people miss. Our pricing meters generated replies, which is cheap when most of your inbox repeats and expensive when it does not. Count two weeks of your own sent replies. If four answers cover most of them, our shape saves you money. If they genuinely do not, a flat per-contact plan will cost you less and we would rather you worked that out now.
What we did spend the engineering on
Two things, narrowly.
The first is that replies are built from writing you actually produced, rather than from a tone dropdown. Not "friendly, professional, concise" — your own sent messages, which is where the identifying habits live. How you open. Whether you use full stops in a one-line answer. Whether "haan ji, available hai" is a thing you would type or something a model made up about you.
The second is refusal. A generated reply has to be grounded in a source you supplied. If it is not, or the model is unsure, or it is slow, your own template goes out instead of the draft. Some replies go to a review queue rather than to the customer.
The interesting part of an automation is not what it sends. It is what it declines to send, and whether it tells you why.
Beyond that: templated replies are free and unlimited, credits move only when the model writes something new, packs start at ₹499, and takeover stands the automation down the instant you reply to a thread by hand.
The parts that are neither good nor bad, just true
Everyone on the official Graph API operates in the same box. Comments can be answered for seven days, DMs within twenty-four hours of the person's last message, and nobody can message a stranger. Any review — ours included — that presents those as product strengths is padding.
How to prove this review wrong
Do not migrate anything. Leave what you have running, point us at one account and one surface, paste in your four most common answers, then read twenty replies as the recipient on a phone. Comment twice from a second account to check dedupe. Comment at midnight to check quiet hours. Reply by hand mid-thread to check takeover.
If the replies are ones you would have sent, the rest is arithmetic. If they are not, nothing on the feature list matters.
For the concessions laid against a larger product, the head-to-head is more specific. For how the refusal behaviour actually works, grounding has its own explanation. And what the free plan covers is the cheapest way to disagree with everything above.