AI agent deployment

A Hermes agent that finds the best deals and contacts the seller for you

The agent scans premium-car and real-estate listings, compares prices with the market median, scores deals 0–100 and posts them to Telegram. On request it runs a deep analysis, contacts the seller and presents the offer.

Hermes Agent · Telegram · scheduler · e-mail · market analysis

In short

An agent built on the open-source Hermes Agent (Nous Research) that works like a deal analyst: it periodically scans listings, drops weak offers, calculates how far the price is from the median of comparable offers and posts the best finds to topic channels on Telegram — premium cars and real estate separately. On a chat command it runs a full analysis of a specific offer, writes to the seller, watches for the reply and sums everything up for the client.

deal messages in the “Premium cars” channel
162
score for every deal
0–100
sample deal: price vs median of 8 comparable offers
−28%
tools · skills · MCP server of the agent
21 · 40 · 1
topic channels: premium cars, real estate, system
3

Figures from the agent dashboard and the Telegram channel on the day the screenshots were taken (November 2026). Offer details and chat history are blurred.

Sound familiar?

What this solves

Going through hundreds of listings a day eats hours.

The agent monitors sources on a schedule, drops offers that miss the criteria and only shows the ones worth a look.

You can’t tell whether a price is a bargain or a trap.

Each offer is compared with the median of similar listings (year, fuel, gearbox, mileage), and for imports the agent adds estimated import costs.

The best offers are gone before anyone calls.

On command the agent messages the seller right away, watches for the reply with a scheduled job and reports back on Telegram.

How it works

Step by step

  1. 01

    Source monitoring

    A scheduled job runs searches with saved criteria: makes, years, budget, location, property type.

  2. 02

    Normalising data

    The agent extracts the key parameters and normalises them so offers from different sites can be compared.

  3. 03

    Market pricing

    The price is compared with the median of similar offers. The agent records how many offers were used and how closely they matched.

  4. 04

    Scoring and posting

    The best deals get a 0–100 score and land in the right Telegram channel with an ID and a button to the listing.

  5. 05

    Analysis on demand

    Type “analysis” and the offer ID — the agent checks details, history and risks, then recommends the next step.

  6. 06

    Contact and presentation

    The agent writes to the seller, waits for the reply and prepares a summary for the client with arguments and negotiation questions.

Implementation details

The core is Hermes Agent — an open-source, MIT-licensed agent with cross-session memory, a skills system, a task scheduler and a messaging gateway. So we didn’t build everything from scratch: we wrote skills specific to this process (search, listing normalisation, market pricing, deal formatting) and connected tools — a browser, e-mail and an MCP server.

The agent talks to the team on Telegram. Deals land in separate topic channels and technical messages in a “System” channel, so nothing clutters daily work. Every deal has an ID, so talking to the agent feels natural: “analysis 65535” or “write to the seller”.

Pricing is based on the median of comparable offers, and the agent always states how many listings were compared and how closely they matched. For imports it adds estimated import costs and flags them as to be verified — the agent supports the decision, it doesn’t make it.

Outbound actions (messages to sellers) can require approval or run automatically by rules — the permission setup decides. Every operation is logged. Read more about the framework in our article on Hermes Agent, and about agents vs chatbots in AI agents for business.

01

Integrations

  • Hermes Agent
  • Telegram
  • Harmonogram (cron)
  • E-mail
  • Przeglądarka (browser-use)
  • MCP
  • LLM
02

What you need

  • Search criteria and offer sources (in line with each site’s terms)
  • A server or VPS for the agent (Docker)
  • A Telegram bot and channels
  • A mailbox for contacting sellers
  • An API key for the chosen language model
03

How we tailor it

  • Other markets: machinery, equipment, rentals, land
  • Your own deal-scoring formula
  • Daily or weekly report on Telegram or by e-mail
  • Saving deals to a CRM or spreadsheet

FAQ

Frequently asked questions

Does the agent message sellers on its own?

It can — either after your approval or automatically by agreed rules. Every message sent is logged and the seller’s reply lands on Telegram.

How does the agent decide a deal is good?

It compares the price with the median of similar offers (year, fuel, gearbox, mileage — or size and location) and states how many listings were compared, then gives a 0–100 score.

Does it work for real estate too?

Yes — in this deployment the agent has a separate real-estate channel. Criteria and pricing then rely on size, location and property type.

Is using listing sites legal?

We design data sources to respect each site’s terms — preferring official APIs, partner feeds and publicly available data used in permitted ways.

How long does a deployment take?

Usually 2–4 weeks to a first working version for one offer type, then extending it with more categories and contact rules.

Get a quote

Want this in your business? Tell us what it should do.

Describe your process and the tools you use. We reply within one business day with an implementation plan.

What do you need?

Goal, deadline, links to inspiration — the more specifics, the more accurate the quote.