Manus, the AI agent platform Meta acquired for over $2 billion in December, went live on Telegram on February 16 with a full-featured agent that runs multi-step tasks directly inside chat. Every subscription tier gets access. The setup takes about a minute: scan a QR code, link your account, start delegating.
The choice of Telegram is the interesting part.
Why not WhatsApp?
Meta owns WhatsApp. Meta owns Manus. And yet the first messaging integration lands on a platform run by a completely separate company. The Decoder noted that the Meta-Manus deal is still under review by Chinese regulators, which may have forced the team's hand. There's another reading, though: Telegram is a lower-stakes sandbox. If an always-on AI agent misbehaves, or burns through credits in ways users don't expect, better to have that happen on someone else's platform first.
That caution might already be justified. TestingCatalog reported that Telegram suspended Manus's agent account shortly after an earlier launch attempt on February 14. Neither Telegram nor Meta commented publicly on the suspension. The agent appears to be back up now, but the incident underscores how fragile these integrations remain.
What the agent actually does
According to Manus's blog post, this isn't a stripped-down chatbot. The Telegram agent runs the same task engine as the web app, handling research, data processing, report generation, and PDF creation through multi-step orchestration. Users can send voice messages, images, and documents. The agent transcribes audio, figures out what you want, and returns results in the conversation.
Two model options: Manus 1.6 Max for complex reasoning tasks, and Manus 1.6 Lite for quick answers. You can also configure response style, from terse to verbose. "This is not a lightweight chatbot add-on," the company wrote, which is exactly what a company launching a lightweight chatbot add-on would say. But the feature list backs up the claim, at least on paper.
The cost question nobody's answering
Manus runs on a credit-based pricing model, and agent-style workflows are credit-hungry by nature. An always-available agent in your messaging app encourages longer, more frequent sessions. That's great for engagement metrics, less great for users who discover their credits evaporated on a Tuesday morning because their agent was doing "weekly meeting prep" in the background.
The blog post doesn't address credit consumption for Telegram tasks specifically. Manus co-founder Tao Zhang posted on X that WhatsApp, LINE, Slack, and Discord support is coming within 30 days, along with native Windows and Mac apps that would let Manus operate users' PCs directly. Ambitious timeline. The 30-day window for five additional platform integrations plus desktop apps suggests either a very large engineering team or some aggressive expectation-setting.
For context, Manus claims to have processed over 147 trillion tokens and powered 80 million virtual computers since launch, per its Meta announcement. Those are big numbers, though they tell you more about compute spend than about whether individual tasks actually worked well.
The competitive landscape is getting crowded. SiliconANGLE reported that OpenClaw has been doing similar things, but requires more technical setup. Meta itself has been spotted testing OpenClaw integrations in Meta AI, which raises the question of whether Meta is hedging its bets on which agent framework wins, even among its own products.
WhatsApp integration, whenever it arrives, will be the real test. That's where Meta's billions of users live, and where credit economics and agent reliability will face actual scale. Telegram is the rehearsal.
