TalkOmni

AI customer assistant that acts, not just chats

Customer Communication

SaaS

Artificial Intelligence

Visit WebsiteSee on Product Hunt

Hi Product Hunt 👋 I have spent the last months building assistants for real businesses, and the same thing kept happening. The bot would answer a question beautifully, the customer would say "great, so can you check my order then?", and the bot would say some version of "please contact our support team". That is the moment the whole thing stops being useful. So TalkOmni is built around the part that comes after the answer. It acts. The assistant reads your own documents to answer, and calls your own backend to do things: look up an order, book an appointment, open a ticket, create a record. Three actions are built in and need no code. For your own systems, you give us one HTTPS base URL and a secret. Every call is a POST signed with HMAC-SHA256, and you see it in a log with the request, the response, the status and the duration. Reads run freely. Writes have to be whitelisted, and their parameters are validated server-side before we ever call you, so the model carries intent and never a price or an ID it should not have decided. It knows when to stop. When a case needs a person, the conversation moves to a human in the shared inbox with the full history, and the bot goes quiet. When the conversation is resolved or goes idle, the assistant takes it back automatically. No customer ever repeats themselves, and your agent never fights the bot for the same reply. One brain, every channel. WhatsApp, Instagram, Messenger, Telegram, your own mailbox, a chat widget on your site, and a voice agent that can talk in the browser or answer your phone line. Same knowledge, same actions, same inbox, same history. A few things I decided on purpose: - No outbound marketing blasts. This is an inbound assistant. WhatsApp marketing is restricted in a lot of markets and I did not want to build a product that gets its customers banned. - Not a CRM. It captures leads and pushes them out to your pipeline over a webhook, and stops there. - Answers are grounded in the documents you upload, not in general model knowledge, so it does not invent a return policy you never wrote. The stack, since this crowd always asks: Go on the backend, Next.js on the panel, Postgres with pgvector for retrieval, Claude for the agent loop and tool calling. We dogfood it. The chat bubble in the corner of talkomni.com is the product answering from our own docs, and it will happily tell you when it does not know. For Product Hunt: 60 days free instead of 14. No card, no code, just sign up today and the extended trial is on your account. I am here all day. I would genuinely like to hear the thing you tried to make a chatbot do that it refused to do, because that list is my roadmap.