Telegram testing 'Cocoon AI' on iOS

TestingCatalog flagged an active iOS build that incorporates a Cocoon‑powered assistant for text improvements, automated fixes and in‑app translations during Telegram’s internal testing. (testingcatalog.com) Telegram’s official January 2026 release added AI summaries for channel posts and Instant View pages that it says are generated by open‑source models running on the Cocoon network, and the same update completed a Liquid Glass visual redesign for iOS. (telegram.org) Cocoon itself launched as a decentralized “Confidential Compute Open Network” on the TON blockchain on November 30, 2025, with Telegram positioned as the network’s first large‑scale client. (cointelegraph.com) The Cocoon GitHub and documentation describe Trusted Execution Environment (TEE) worker images, attestation and a marketplace where GPU owners earn Toncoin for serving models, signalling enclave‑based inference instead of standard cloud model hosting. (github.com) Reporting from developer and industry outlets indicates Telegram has already routed lightweight tasks such as message translation through Cocoon while planning to put heavier summarization and conversational assistants on the network in subsequent rollouts. (atomicwallet.io) Telegram published and third‑party coverage show Mira as a deployed Cocoon‑powered assistant inside Telegram in early February 2026, demonstrating movement from experimental iOS tests toward live assistant features. (top.co) Telegram’s user scale—announced by CEO Pavel Durov as over 1 billion monthly active users in March 2025—means any iOS Cocoon integration tested at scale will affect an immense installed base and platform telemetry. (techcrunch.com) Public repos and developer docs make clear mobile teams will need to wire client SDKs into Cocoon’s dispatch/attestation flows, manage network latency and battery costs for iOS inference calls, and coordinate model‑routing policies between local, Cocoon TEE, and server paths during phased rollouts. (github.com)