This is a submission for the Hermes Agent Challenge

What I Built

Well we are always in a hurry in the morning catching up on that favourite mocha and toast in our hand. Realising the need to give time to the news at that time of the day sounds unrealistic for the professionals in rush and struggling to make time for themselves even. Recognising a typical physical newspaper to read and mug up the information to find insights would be time-consuming and AI news apps to the least demand subscription and display detailed summary.What do I bring among you all is Daily Digest Agent a self-hosted, fully autonomous AI agent that researches the news on your chosen topics every morning and delivers a structured briefing straight to your inbox.
This is not a usual news digest tool which has a fixed prompt and hardcoded URLs. Hermes Agent runs a real multi-step loop — it plans its own searches, decides which articles are worth reading, fetches and extracts full article content, spots cross-topic patterns, and writes a clean digest. Then it emails it to you. Every single day, automatically, for free.The agent also gets smarter over time. After each run, it updates a skill memory log — tracking which topics returned rich results and which ran dry. By run 5, it's already adjusting its own search strategy without any input from you.
The problem it solves is information overload. Rather clicking on multiple tabs your mail is the ultimate stoppage to get news highlights compiled by an agent that knows your interests and not which is optimizing based on the clicks.

Demo

How is mail displayed!

Code

=== Daily Digest Agent starting — Friday, May 29 2026 ===
Hermes iteration 1/20
  Tool call: web_search → "artificial intelligence news May 29 2026" → 10 results
Hermes iteration 2/20
  Tool call: web_fetch  → reading techcrunch.com/...
Hermes iteration 3/20
  Tool call: web_search → "Indian startup funding round May 2026" → 10 results
...
Digest saved to output/digest_Friday_May_29_2026.md
Email sent to

My Tech Stack

• Hermes Agent (NousResearch) — agentic loop architecture
• OpenRouter (auto routing) — LLM backbone, free tier
• Python 3.10+ — core language
• httpx — async HTTP requests
• Google News RSS — real-time news, no API key needed
• trafilatura — article text extraction
• Gmail SMTP — email delivery
• Windows Task Scheduler / cron — daily automation

How I Used Hermes Agent

I repeat it's an engine not a wrapper, A multistep agentic loop inspired by Hermes agent core architecture.
Tool Calling-Hermes receives two tools (web_search and web_fetch) and autonomously decides which queries to run, which articles are worth reading, and when it has gathered enough to write the digest. No hardcoded steps.
Self-termination — The agent runs until it emits DIGEST_COMPLETE on its own. It decides when it's done.
GEPA-style skill memory — After every run, a skill log is updated with what topics were searched and what patterns worked. On the next run, this context is fed back to the agent — so it gets progressively sharper at finding relevant stories. After 5+ runs you can see it adjusting its own search queries.
Scheduled autonomy — Runs every morning via cron/Task Scheduler with zero human input. Fully autonomous end-to-end.

What makes this different from a simple LLM call is that the agent plans, executes, reads, reasons, and self-improves — exactly the capabilities Hermes Agent helps with.