Scrocle

Telegram agent walkthrough · 2026

How a Telegram agent stack is wired

Laboratory

Role
Independent engineer
Stack
Python, FastAPI
Status
Laboratory
Live
scrocle.cloud/demo/multi-agent

Problem

People ask for a Telegram bot that remembers context, routes to the right worker, and survives a model outage. The shape of that stack is easy to talk past. This page is a walkthrough of the stages, not the running bot.

Constraint

No Telegram token, no LangGraph process, no LiteLLM proxy, no Postgres. Session memory is a dict. Failover only changes labels. The UI says so.

Architecture

  1. Inbound text
  2. Session dict
  3. Intent keywords
  1. Specialist template
  2. Fake chat UI
Typed message in. Fake session. Keyword intent. Canned specialist reply. Fake Telegram pane.

Decisions

  • Show the shape, not a fake production stack

    The stages are the ones I would actually build. The demo does not pretend those processes are running.

  • Keyword routing is enough for a click-through

    A buyer can see where SQL, research, memory, and actions sit. A real supervisor is a later job.

Click a scenario. You will see the path a real Telegram agent stack usually takes. You will not get a bot you can message.

If you need the production version, that is a webhook, a real supervisor, a model proxy, and a store that survives a restart. Say so in the brief.

Outcome

Demo. Live at scrocle.cloud/demo/multi-agent. Not a shipped bot.

What was handed over

  1. How to run the FastAPI simulator locally
  2. That nothing is sent to Telegram
  3. What a production webhook and store would add

Open the live demo

Next project Jobs Engine Rank public listings against a profile