Ask it to find a decision, catch up on a conversation or prepare a reply. The agent reads Telegram’s messages directly and works through the client’s own actions.
An agent inside Telegram. Not a screen-reading workaround.
It works with chat identities, messages and native actions—not screenshot coordinates.
Example conversation
Because the agent is embedded in the client, it can request a chosen chat’s messages directly—with their senders and identities. You do not have to copy the conversation into another assistant.
“Catch me up on Launch team. What needs an answer from me?”
Example brief
Your reply: Mira needs your approval of the final copy. Read the source ↓
Update: Jules uploaded the assets. Waiting: Sam will run QA after approval.
The conversation behind the brief
Jules · Sep 5, 09:10“The assets are uploaded.”
Mira · Sep 5, 09:12“Can you approve the final copy before we publish?”
Sam · Sep 5, 09:14“I’ll run QA after approval.”
How it works
find_contact resolves “Launch team” to its Telegram chat ID. Ambiguous matches need a choice.
read_recent(target, n=50) reads a bounded history window with text, senders and message identities through the native client.
The agent separates updates from requests and writes the brief from those messages.
This example reads a recent message window. The agent writes the brief; the native client supplies the conversation data.
The native agent uses propose_message(target, text) to prepare a pending draft. You inspect the recipient and final wording in the client.
Accepting the native proposal starts delivery through Telegram. This page only illustrates the flow.
propose_message is an internal action. An external MCP agent uses send_message; per-message review requires a client that supports form elicitation. The example assumes review is enabled.
03 / Return to the conversation
Find the deadline you agreed on.
Search the selected chat, then check the later confirmation against the original proposal.
Example conversation
“What deadline did we finally agree on for the launch?”
Example answer
September 12. The later confirmation replaces the September 10 proposal. Read the confirmation ↓
Two messages, one changed decision
Jules · Sep 3, 10:20 · proposal“Let’s target September 10.”
Mira · Sep 5, 11:05 · confirmation“Confirmed: September 12. QA needs the extra time.”
How it works
find_contact resolves “Launch team” to the selected chat.
search_chat(target, query) looks for “final launch deadline” in that chat’s indexed memory.
read_around(target, date) reads the surrounding history for the date found. The agent compares the proposal with the confirmation.
Search finds candidate passages; reading the surrounding conversation is what lets the agent check a changed decision. The source links above stay inside this example.
Mimicry Telegram · v0.3-beta
Download the desktop client.
Public beta builds are available for Windows x64, macOS on Apple silicon and Linux x64.
