Short answer
A supervised Hermes-style agent can read your calendar, draft inquiry replies, and notify you on Telegram for approval before anything sends. Built for power users who want speed with a human gate.
After this lesson: You will understand the moving parts of a calendar-aware, Telegram-supervised agent and whether you should build one.
Who this is for
You already respond fast manually. You want fewer context switches between cook, car, and inbox.
You are comfortable with setup steps, API keys, and testing in a sandbox before clients see output.
If that is not you, stay on ChatGPT drafts from the AI overview lesson. That is enough for most chefs.
What you are building
Hermes (or a similar open personal agent) runs as an assistant that can call tools: read calendar free-busy, summarize an inquiry email, draft a reply in your voice.
Telegram becomes your pocket approval UI: you get a draft, tap approve or edit, then it sends from your normal channel.
Calendar access lets the agent suggest meeting windows you can actually honor, not fantasy slots.
What is really happening
The win is not full autonomy. The win is shrinking time from inquiry to professional first touch while you are hands-on in a kitchen.
Why this works
Human-in-the-loop agents respect what selling personal chef work requires: judgment, tone, and real availability. Telegram approval is friction on purpose. The two seconds you spend tapping approve are cheaper than rebuilding trust after a bot promised Tuesday when you are booked through Thursday.
Setup outline (high level)
Install and configure Hermes (or your chosen agent runtime) on a machine or host you control.
Connect calendar read access (Google Calendar or similar) with least privilege: free-busy, not edit, until you trust the flow.
Connect Telegram for notifications and approve or reject buttons on each draft.
Write a system prompt with your non-negotiables: no pricing without you, no sample menus, book in-home meetings, link to your qualify checklist language.
Route inquiry sources (email forward, form webhook, or manual paste) into the agent as events.
Test with fake inquiries until ten drafts in a row look like you.
- Start read-only on calendar.
- Require explicit approve on every outbound client message.
- Log what the agent proposed for later tuning.
- Rotate API keys if you share the host.
Risks to respect
Auto-send without approval can quote wrong scope or double-book.
Client PII in logs needs hygiene if you self-host.
Over-automation can sound robotic if you stop editing prompts.
You still own every promise the client receives.
What to do
Complete the AI overview lesson and three manual prompts first.
Only add an agent when manual drafts feel easy.
Ship version one: inquiry alert, draft reply, Telegram approve, you send.
Add calendar suggestions in version two after version one is boringly reliable.
FAQ
- Is Hermes the only option?
- No. Any agent stack with tool use, calendar read, and a human approval step fits the same pattern.
- Can it book meetings automatically?
- Not at first. Suggest slots you approve, then you or the client confirms. Auto-booking is version three, not day one.
- What if I am not technical?
- Use ChatGPT drafts. Hire a tech-savvy friend for a one-time setup, or wait until ChefLeads ships simpler tooling.
- Does this replace ChefLeads inquiries?
- No. It helps you respond to inquiries you already receive, from any source.
Related: Respond fast to inquiries