How Telegram Is Becoming the Default Interface for AI Agents

TL;DR

• Telegram offers the most complete platform for AI agents today. 1B+ MAU, mature Bot API, Mini Apps for full UI, native payments via Stars and TON, and streaming responses shipped in March 2026.
• 183 AI bots with MAU >= 10K operate on Telegram, totaling 44.3M combined MAU (March 2026). The sum includes significant user overlap: the same person may be using multiple AI bots & it is counted multiple times. These are not necessarily unique users.
• BotFather hit 7.3M MAU -- more than doubling from 3.5M in early January. Developers are rushing to build bots at an accelerating pace.
• Cocoon is a decentralized AI compute network on TON: 6 independent GPU operators, 82K TON lifetime volume ($237K), 132K inference jobs per month and growing.
• The gap: no top-20 AI bots use TON on-chain yet. But the infrastructure is early-stage with rapid development: 14 Fast Grant winners, MCP adoption across the ecosystem, and x402 machine-to-machine payments already in progress.

Why Telegram?

Telegram has 1B+ monthly active users. Here's why it matters for AI agents.

Three signals confirm the thesis:

Durov is building for AI. In March 2026, Telegram shipped streaming responses for bots, which is purpose-built for AI assistants. From the announcement: "All Telegram chatbots can now stream responses to users in real time, which is great for AI assistants." Threaded conversations for bots followed, enabling multi-topic chats. The Bot API also supports polling mode for rapid prototyping. No webhooks, no HTTPS, no server required.

BotFather is a developer portal now. At 7.3M MAU, BotFather more than doubled from 3.5M in early January (2x growth in under two months). That's more active users than most SaaS developer platforms. Framework trends reflect the shift: grammY (JavaScript) is up 10x, aiogram (Python) up 2x.

Manus chose Telegram first. Meta acquired Manus for $2B+ in December 2025. Two months later, Manus launched AI agents on Telegram, even before WhatsApp, Slack, or Discord. Unlike chatbots, Manus performs multi-step tasks: apartment search, hotel booking, website building. As The Decoder noted, Meta's own subsidiary chose Telegram over WhatsApp for its agent launch, partly because EU regulations restrict third-party AI assistants on WhatsApp, and partly because Telegram's bot API is more developer-friendly. When a $2B AI company picks your platform first, that's external validation.

AI Bot Market on Telegram

As of January 2026, 183 AI bots with MAU >= 10K operate on Telegram, totaling 44.3M combined MAU. The top 10 bots account for over 80% of that total. Classic power law distribution.

Methodology note: Bot-level MAU is updated to March 2026. Market-wide aggregates (183 bots, 44.3M combined MAU, historical trend) reflect the latest complete March 2026 snapshot. Source: SwanRate bot analytics (800K+ bots tracked). 44.3M is the sum of individual bot MAUs with significant user overlap, and not unique users.

Top 10 AI Bots (March 2026)

Market Segments

We classified the largest AI-related Telegram bots and here is the distribution:

Payments: How AI Bots Monetize on Telegram

Current State

Most AI bots monetize through Telegram Stars or traditional fiat card payments. No top-20 AI bot accepts TON directly on-chain. Only @syntxaibot (Syntx AI) mentions cryptocurrency payments.

Stars as TON Settlement Rail

Even without direct TON integration, AI bots are already part of the TON economy. Here's why:

Telegram Stars are the native in-app payment method for bots. When developers earn Stars, they withdraw through Fragment and receive Toncoin on-chain:

User buys Stars (Apple/Google IAP) → Bot earns Stars → Developer withdraws via Fragment → Receives Toncoin on-chain

The numbers are significant (TON on-chain data):

• 81.7M TON distributed to bot and channel developers all-time ($192M)
• 3,223 unique developer recipients as of February 2026, growing monthly
• Current run rate: 8M TON per month ($12-13M)
• Stars purchase volume growing: 6.2M TON in January to 8.0M TON in February (+29%)

This makes TON the de facto settlement layer for Telegram's bot economy. Telegram reported $870M revenue in H1 2025 (+65% YoY), with bot and Mini App monetization as a primary growth driver.

As Coinbase CEO Brian Armstrong put it: "Very soon there are going to be more AI agents than humans making transactions. They can't open a bank account, but they can own a crypto wallet." TON is uniquely positioned here -- it's the only blockchain already embedded in a messenger with 1B+ users, where agents and payments coexist natively.

TON Pay SDK (February 2026)

A new on-chain payment primitive launched in February 2026, enabling direct TON and jetton payments inside bots. No Telegram Stars intermediary, no commission to Telegram. No AI bot has integrated yet, but three Fast Grant winners are building x402 machine-to-machine payment protocols on top of it.

The opportunity: as TON Pay matures, AI bots could shift from Stars (which carry a 15-30% commission depending on the purchase channel) to direct on-chain payments, which lead to better margins for developers.

Cocoon: Decentralized AI Compute on TON

What It Is

Cocoon (Confidential Compute Open Network) is a decentralized marketplace for AI inference. GPU providers earn Toncoin for running AI workloads inside Intel TDX Trusted Execution Environments. Data stays encrypted even from the operators processing it. The project is open-source on GitHub.

Pavel Durov announced Cocoon at Blockchain Life in October 2025. It launched on TON mainnet in December 2025. GPU providers can onboard via cocoon.org.

Telegram is both – the builder and the first client! Cocoon powers AI Summaries, translation features, and image understanding inside Telegram.

Growth Metrics

"AI Jobs" = individual inference requests (summarization, translation, image understanding, etc.) processed by GPU workers in TEE enclaves. Source: Cocoon Explorer, TON on-chain data.

Key observations:

• Total lifetime volume: 82K TON ($237K) processed through the network
• Jobs plateaued but revenue grew: fewer jobs at higher value -> the network is maturing
• Client count tripled: from 7 to 17 in one month -> real developer adoption
• Supply side expanding: 6 independent GPU operators running up to 60 workers, with new operators onboarding each month

Mira: First Third-Party Consumer

@mira is the first third-party app to consume Cocoon compute. Built by TOP - The Open Platform (team behind Telegram Wallet). Mira offers a "Private mode" powered by Qwen3 running on Cocoon's encrypted infrastructure.

Mira reached 262K MAU in February 2026, growing from 40K in December 2025. As a Cocoon consumer, Mira validates the core premise: third-party apps paying Toncoin for decentralized, privacy-preserving AI inference.

The Pattern

Cocoon is early-stage infrastructure. Telegram is simultaneously the builder and the anchor customer, establishing a pattern where AI compute becomes a TON-native service, paid in Toncoin. As more apps integrate (like Mira) -> the demand side diversifies. This is not unusual for infrastructure projects. For example AWS started as internal Amazon tooling before becoming a platform.

Developer Ecosystem

Fast Grants: 40 Submissions, 14 Winners

The "Agent Tooling Fast Grants" contest (February 12-24, 2026) attracted 40 submissions. 14 projects won $1K each (pool expanded from $10K to $14K due to quality).

Agent SDKs and Tools (4 winners)

• Teleton Agent -- autonomous AI agent platform for Telegram + TON, 100+ tools, persistent memory
• AI sdk ton-tools -- TON tools for the Vercel AI SDK
• TON OpenClaw Skill -- Claude Code skill for TON blockchain
• TON API Skill -- Claude Code skill for TON API

Infrastructure and Security (6 winners)

• Ton AI Audit -- LLM-powered smart contract auditing
• AgentGuard -- on-chain wallet guardrails (spend limits, whitelists)
• TonPay 402 -- x402 protocol for machine-to-machine payments
• x402-payments -- automatic payments for premium content and APIs
• Telehost (GitHub) -- agent hosting platform
• TON Agent Platform -- no-code agent building

Communication and Discovery (2 winners)

• TAMP -- on-chain agent registry + agent-to-agent discovery
• Mesh Network -- agent coordination with on-chain reputation

Gifts Ecosystem (2 winners)

• Gift Asset MCP -- Telegram Gifts data for AI agents
• Giftstat AI Bots (GitHub) -- Gifts analytics

Patterns

Four clear signals emerge from the grant winners:

1. MCP is popular, but not unopposed. Nearly every project exposes or consumes Model Context Protocol & it's becoming a common interoperability layer for AI agents. However, there's a growing counter-argument in the AI dev community: agent skills (self-contained, declarative tool definitions) may be better suited for many use cases because they don't pollute the agent's context window with tool schemas. The ecosystem is still converging on standards.
2. x402 payments are emerging. Two winners are building machine-to-machine payment protocols on TON -- enabling agents to pay each other without human intervention.
3. Claude Code skills for TON. Two winners build TON-specific skills for Claude Code, meaning TON is entering AI developer toolchains directly.
4. Agent-to-Agent protocols. TAMP and Mesh Network are building discovery and coordination layers -- the beginning of multi-agent infrastructure.

Notable Projects from the Broader Ecosystem

• @ton/mcp -- official TON WalletKit MCP server. Agents can manage wallets, send jettons and NFTs, interact with DeFi.
• Teleton -- 100+ tools, persistent memory, plugin architecture for autonomous agents
• OpenClaw -- full TON blockchain access via Claude Code
• Veritas -- Visual Security Oracle for scam detection by trading agents
• CleanAds AI -- autonomous escrow and brand-safety oracle for Telegram ads

On-Chain Reality Check

None of these projects have verified mainnet contracts yet. The framework layer is being built; on-chain usage is the future. This is the honest state of AI on TON: tooling first, applications next.

The Gap: TON in the AI Crypto Landscape

The AI agent crypto market peaked at $14B in January 2025 and currently sits around $4B. TON is notably absent from the major frameworks:

TON's Unique Position

TON is the only blockchain natively integrated with a 1B-user messenger. No other chain has this distribution advantage. Messaging + payments + blockchain + AI compute (Cocoon) + streaming responses. As of March 2026, no other platform combines all of these.

But having the stack and having adoption are two different things.

What Needs to Happen

1. A native AI agent framework for TON, or deep integration of TON into existing frameworks (GOAT SDK, AgentKit)
2. Marquee AI agents using TON on-chain (even one or two high-profile bots transacting on TON would shift the narrative)
3. Developer onboarding at scale: MCP tools exist
4. Machine-to-machine payments: x402 on TON is being built by 3 teams already (this could become TON's wedge

Conclusion: The Stack Is Ready, The Apps Aren't

What exists today:

• Distribution: Telegram with 1B MAU, BotFather at 7.3M MAU (2x in two months), Manus choosing Telegram first
• Compute: Cocoon processing 125K+ AI inference jobs per month across 6 independent GPU operators
• Payments: Stars with TON settlement (81.7M TON distributed to developers), TON Pay SDK for direct on-chain payments
• Developer tools: MCP ecosystem, WalletKit, Claude Code skills, x402 payment protocols
• Platform features: Streaming responses (March 2026), threaded bot conversations, 2GB file uploads

What's missing:

• AI applications that actually use TON on-chain: Currently zero of the top 20 AI bots transact on TON.
• A native AI agent framework for TON: The ecosystem has tooling but no cohesive framework.
• Diversified Cocoon demand: Telegram is the primary client.

The Narrative

Telegram is becoming the default interface for AI agents. TON is Telegram's native blockchain. The infrastructure: compute, payments, developer tools, streaming -- is being built. The application layer is the next frontier.