AI on WhatsApp in 2026: What Meta Allows and What to Do
Meta banned general AI assistants on WhatsApp and will charge per message from October 2026. What options remain and why Telegram is the way out.
AI on WhatsApp in 2026: What Meta Allows and What to Do
Since January 15, 2026, WhatsApp no longer supports general-purpose AI assistants: Meta changed its terms and ChatGPT, Copilot, Perplexity, and Luzia disappeared from the platform. Only business bots with a defined function (customer support, orders, bookings) are still allowed through Meta’s official API. And from October 1, 2026, every service response will also carry a per-message cost.
If you are a solo founder and want a personal AI assistant on your phone, the practical route in 2026 is Telegram: its official Bot API is free and still supports general assistants. That is exactly what Tessera Founder includes: a managed Hermes instance (the open-source agent by Nous Research, OpenClaw-style) connected to your Telegram, with AI included and no token counter, for €55 per month ($55 in the Americas).
What Meta banned and why ChatGPT disappeared from WhatsApp
Meta changed the WhatsApp Business API terms in October 2025 to ban general-purpose AI chatbots, with an effective date of January 15, 2026, as reported by TechCrunch. A “general-purpose” chatbot is one that runs on a language model, supports open conversation on any topic, and is not tied to a specific business process.
Meta’s stated rationale is that the WhatsApp Business API exists to help businesses provide customer support and send relevant updates, not to act as a distribution platform for third-party chatbots. In practice, that leaves Meta AI as the only general assistant available inside the app.
The consequence arrived on schedule. OpenAI removed 1-800-ChatGPT from WhatsApp on January 15, 2026, and Copilot, Perplexity, and Luzia followed the same path. That is why today you cannot “connect ChatGPT to WhatsApp”: it is not a technical problem, it is a contractual prohibition by Meta.
What is still allowed are bots with a defined function: support, order tracking, bookings, or appointments. If that is your use case, the official WhatsApp Business API remains valid; unofficial QR bridges never were, and they end in a number ban.
From October 2026, WhatsApp charges for every service message
The second change affects cost. Meta announced on July 1, 2026 that, starting October 1, 2026, service messages (free-form replies within the 24-hour window) are charged per message with no volume discount, according to its pricing documentation. Those replies had been free since late 2024.
Two details sting especially if you run an AI assistant. The charge applies to third-party AI tools but not to Meta AI, and the exact per-country rates will be published before September 1, 2026. Only the 72-hour window for free entry points (Click-to-WhatsApp ads from Facebook or Instagram) remains exempt.
Run the numbers with a real personal assistant: if it replies 40 times a day, that is roughly 1,200 responses per month going through the counter. And that is on top of the AI bill if your provider still meters tokens. We analyze the full change, including what is known about rates, in WhatsApp will charge for service messages.
Telegram: the channel that remains open for AI assistants
The short answer: Telegram still allows general AI assistants and its official Bot API is free, with no per-message cost. Creating a bot takes minutes with BotFather and requires no business verification.
| Channel | General AI assistant? | Per-message cost | Risk |
|---|---|---|---|
| Telegram (official Bot API) | Allowed | €0 | Low |
| WhatsApp (Meta’s official API) | Banned since Jan 2026; business bots only | Per message from Oct 2026 | Compliant |
| WhatsApp (unofficial QR bridge) | Banned | ”Free” | Number ban |
| Discord and Slack (official APIs) | Allowed | €0 | Low |
That is why personal agents like OpenClaw or Hermes increasingly live on Telegram. The channel adds no message counter and Meta does not decide which assistant you can use. Discord and Slack remain open too, which matters if your work already lives there; OpenClaw-style agents speak several channels, so the assistant is not locked to one app.
WhatsApp still makes sense in one case: customer support with defined flows, especially if your market is LATAM, where WhatsApp is the native business channel. Free regional assistants like Zapia boast millions of users, but “free” means someone else controls both the channel and the assistant.
Having your own assistant without self-hosting
Hermes is an open-source personal agent (MIT license) by Nous Research, OpenClaw-style: it lives in your chats and works for your business from there. The tutorials tell the truth: you need to run it in a private environment, which usually means renting a VPS, configuring it, and maintaining it.
The numbers surprise you where you least expect. The server is the cheap part, between $4 and $24 per month depending on the provider; the real expense is tokens, because an agent like Hermes carries between 15,000 and 20,000 tokens of tool definitions in every request. The full breakdown is in how much OpenClaw really costs.
The fear of “burning tokens like crazy” is reasonable, and most managed agent hosting services do not solve it: they host the software and the variable AI bill remains yours. We compare that market in OpenClaw alternatives.
Privacy matters as much as price
For many founders privacy is the first filter, not the last: the assistant will read emails, quotes, and customer data. With your own Hermes instance the setup is private, and your business history stays in your workspace instead of on a consumer app’s servers.
At Tessera Founder inference runs on European servers, with GDPR as the framework and no dependency on a messaging platform’s terms. It is the same criterion we explain in where to host LLM inference with EU data residency.
Tessera Founder: managed Hermes with AI included
Tessera Founder solves both halves for €55 per month: we operate your private Hermes instance on European servers and connect it to your Telegram, with AI included at a flat rate and no token counter. The only limit is throughput, 20 queries per minute, with 32K context and thinking mode without caps.
What about WhatsApp? When your use case involves business flows with defined steps, we configure it with Meta’s official API, which is what their terms allow, and we explain the October 2026 per-message cost before it arrives. For the everyday general assistant, the recommended channel is Telegram.
The comparison against a piecemeal stack is straightforward. Between ChatGPT Plus (around €20), automation tools (€30-50), VPS plus tokens (€15-70), a paid community (€30-50), and training (€45-90), the typical total is €160-300 per month. The plan also includes the founders community on Discord (spaces in Spanish and English), one live training session per month, and a weekly podcast.
There are 25 founding spots at €55 forever, with a first-month money-back guarantee and cancel-anytime terms. It is designed to run your business, not to serve your product’s end users at scale; when you grow, there is an upgrade path.
How to decide: three questions before choosing a channel
First: is your assistant for you or for your customers? If it is personal and general-purpose, WhatsApp is ruled out by terms and Telegram is the channel. If it is a support bot with a defined function, official WhatsApp remains valid.
Second: who pays for the tokens? An agent that converses all day multiplies requests, and every Hermes request carries its tool overhead. With a flat rate, the expense does not depend on how busy your month is.
Third: who maintains it? If you do not want to touch a VPS, choose a managed option. If you enjoy administering servers, self-hosting is a solid choice; the common trap is the middle ground, paying for managed hosting while still facing a variable token bill. Managed with the AI included is the only setup where both the channel and the model stop metering you.
FAQ
Can I connect ChatGPT to WhatsApp in 2026?
No. OpenAI removed ChatGPT from WhatsApp on January 15, 2026 due to Meta’s terms change, as did Copilot, Perplexity, and Luzia. General assistants remain available on channels like Telegram.
Is replying via WhatsApp with the official API still free?
Within the 24-hour window yes, but only until September 30, 2026. From October 1, Meta charges each service message per unit, with per-country rates published before September 1. Template messages, the pre-approved notifications, already had per-message pricing and keep it.
Is it safe to add AI to WhatsApp with a QR bridge?
No. Unofficial bridges violate WhatsApp’s terms and the documented outcome is a number ban. If you need WhatsApp, use Meta’s official API with a defined-function bot.
What do I need to have an AI assistant on Telegram?
A bot created with the official Bot API, which is free, and an agent like Hermes connected to a language model. You can set it up yourself on a VPS or get it managed with AI included, as with Tessera Founder. The bot is linked to your Telegram account and you can revoke it at any time from BotFather.