Development time ~1–2 weeks
AI Agent
and asks me where the decision is mine — 134 days live
Development time ~1–2 weeks
and asks me where the decision is mine — 134 days live
AI Agent
About the Project
I have dozens of small jobs that need attention, not expertise: remind a client, check a status, hand a task to the right bot, follow up three days later. A regular chatbot answers one question and forgets. The real worry with an AI agent is different: what if it does something that can’t be undone?
I built myself a personal AI agent, Jes, that lives in Telegram and takes this routine off my plate. It understands text and voice notes, remembers earlier conversations, handles jobs on a schedule and hands tasks to other bots, while anything risky waits for my approval
Similar solution: AI Agents
What You Get
If you have dozens of small jobs and keep them all in your head, an AI agent like this takes the routine off you: it reminds, checks, hands tasks to your tools and bots and comes to you only for a choice or an approval. You write or talk to it in plain language in Telegram, and anything risky, costly or public happens only after your yes. The timeline is one to two weeks: setup, memory, approval rules, schedule and testing on your data. I give an exact estimate once I see which tools and bots you already have
How it works
- 01I type or send a voice note
- 02Jes recalls past conversations and my profile
- 03Runs 10 scheduled jobs on its own
- 04Hands tasks to 7 other bots
- 05Anything risky, costly or public — only after my yes
- What
- the copywriter’s post
- Where
- my Telegram channel
- Status
- waiting for my yes
How It Works
01
Text or voice
I send a voice note, and speech recognition turns it into a command
02
Memory that lasts
It keeps context from earlier conversations and my profile
03
Scheduled jobs
10 recurring tasks get done on their own
04
Teamwork
It hands tasks to 7 other bots and reports back only when there is a result
05
Two AI models
If the main one fails, the backup continues automatically
06
Safety rules
Deleting data, spending money, messaging clients or publishing always waits for my approval
07
It learns
After my corrections it writes new rules for itself
Tools
-
Hermes Agent
open-source Python core: memory, schedule, skills -
Telegram Bot API
talking to the agent by text and voice -
OpenAI Codex and a second model
the main AI model and an automatic backup -
Whisper via Groq
speech recognition for voice notes -
Supabase and Vercel
API integration for the website -
Linux server
a 24/7 service that restarts itself after a crash
Result
In service for 134 days. 165 tasks handed to other bots since July, 10 jobs running on schedule, and every risky step still goes through me
Anything risky, costly or publicI want the same
the agent does only after my word