Hermes Agent: the open-source agent that runs my scanner on a schedule
Nous Research's Hermes Agent is a self-improving AI agent with memory, skills, cron and a Telegram gateway. Here's how it became the operator behind my 3WT squeeze-scanner pipeline.
Repo: NousResearch/hermes-agent · MIT · plus nesquena/hermes-webui for a browser UI
Most "AI agent" demos answer a question and stop. What I needed was the opposite: something that runs every weekday evening whether I'm watching or not, remembers what happened yesterday, and reports back on my phone. That's what Hermes Agent turned out to be good at.
What it is
Hermes Agent is Nous Research's open-source, self-improving agent. The parts that matter for real work:
- Persistent memory it curates itself, with search across past sessions.
- Skills it creates after complex tasks and refines as it uses them.
- A built-in cron scheduler that runs natural-language automations unattended.
- One gateway for messaging — Telegram, Discord, Slack, WhatsApp, Signal or the CLI.
How it runs the 3WT pipeline
My squeeze scanner is a collection of plain Python scripts. Hermes is the operator around them:
- Weekdays after the close, a Hermes cron job runs the scan across ~3,700 tickers, imports results into the signal tracker and sends a compact report to Telegram.
- Sunday early morning, another job refreshes the stock universe.
- When something fails, the failure lands in the same Telegram chat as the results, so I see it.
The scripts stay deterministic and testable. Hermes handles scheduling, delivery and the conversation around the results — "why did this ticker drop off?" is a message, not a terminal session.
Memory that compounds
Hermes sessions are synced nightly into gbrain for retrieval, so the agent can look up past decisions — why a filter was added, which backtest killed an idea. A sync script exports only new or changed sessions, so it's safe to run every day.
The Web UI
Hermes WebUI is a small, self-hosted browser interface for the agent — plain Python and vanilla JavaScript with no build step. It's handy when a Telegram thread isn't enough room to read a long report.
What I'd watch out for
- Give it boring tools. The agent is best at orchestrating scripts you can test, not at being the logic itself.
- Keep secrets out of prompts. Load them from the environment the scripts already use.
- Read the cron dashboard. An unattended agent is only as good as your habit of checking it hasn't quietly stopped.
If you want an agent that does a job every day rather than a demo once, Hermes is worth an afternoon.
Some links in these notes are affiliate links. If you buy through one, I may earn a commission at no extra cost to you. I only link to tools I use or would recommend.