Your inbox answers itself. In their language.
Production LLM agents working your inbound on Telegram, WhatsApp, Instagram and Discord. They wake a human only when the deal earns it.
Not a mockup
This is the dashboard, right now.
Real exchanges from the same deployment, replayed with every identifier stripped.
One agent, seventeen days
The number that matters is the one where nobody answered.
What happens at 3am
A lead messages you in Arabic on a Friday night.
- 01
It answers in Arabic
In the language they opened in, at the length a person would use, inside a minute.
- 02
It qualifies before you spend a minute
Deposit intent, account status, which regulator, whether they are already with a competitor. Tyre kickers get a real answer and stop there.
- 03
It chases the ones who go quiet
Most inbound dies because nobody followed up on day two. Follow up runs on its own schedule, paced like a person.
- 04
It wakes you only when the deal earns it
You get the thread with context attached and the handoff point marked. Nine times in ten that call never has to happen.
The agent is the part you can see.
Most of the work is the plumbing behind it. Scraping and enrichment, CRM writes, document pipelines that turn unstructured reporting into queryable data, and scheduled jobs that reconcile what nobody wants to reconcile by hand.
Python, FastAPI, PostgreSQL, pgvector, Redis, Docker, Linux, n8n. Claude through the Messages API, the Agent SDK and MCP, alongside OpenAI, Gemini, DeepSeek and local models on Ollama when the data cannot leave the building.
I also run five trading systems of my own against MetaTrader 5, Interactive Brokers, Bybit and Alpaca. That is where the paranoia about silent failure comes from.
Fixed scope, fixed price, no discovery phase.
The morning after
Put one on your inbound for a week.
Then read the transcripts and decide. That is a faster answer than any call I could give you.