<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>Forem: Alexandre Lasly</title>
    <description>The latest articles on Forem by Alexandre Lasly (@athenaios).</description>
    <link>https://forem.com/athenaios</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3934022%2Fc37fa2ad-2102-4fa4-8211-f64a8b0246d2.png</url>
      <title>Forem: Alexandre Lasly</title>
      <link>https://forem.com/athenaios</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://forem.com/feed/athenaios"/>
    <language>en</language>
    <item>
      <title>Bounty Watcher: An Autonomous AI Agent That Finds Paid Gigs While You Sleep</title>
      <dc:creator>Alexandre Lasly</dc:creator>
      <pubDate>Sat, 16 May 2026 00:26:25 +0000</pubDate>
      <link>https://forem.com/athenaios/bounty-watcher-an-autonomous-ai-agent-that-finds-paid-gigs-while-you-sleep-38l7</link>
      <guid>https://forem.com/athenaios/bounty-watcher-an-autonomous-ai-agent-that-finds-paid-gigs-while-you-sleep-38l7</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hermes-agent-2026-05-15"&gt;Hermes Agent Challenge&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;h2&gt;
  
  
  What I Built
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Bounty Watcher&lt;/strong&gt; is an autonomous agent pipeline running on Hermes Agent that scans freelance bounty platforms 24/7, filters out noise, deduplicates, and delivers only relevant opportunities — straight to Telegram.&lt;/p&gt;

&lt;p&gt;It covers two platforms:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Superteam Earn&lt;/strong&gt; — crypto/solana bounties with cash rewards&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;GitHub Issues&lt;/strong&gt; — open source bounties across the ecosystem&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The agent has been running in production for weeks on an Android phone via Termux. It costs nothing to operate beyond the LLM tokens used for occasional reasoning tasks — close to zero in this architecture.&lt;/p&gt;

&lt;h2&gt;
  
  
  Demo
&lt;/h2&gt;

&lt;p&gt;The agent delivers formatted bounty digests directly to Telegram every few hours:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;🏆 &lt;strong&gt;Superteam Bounties&lt;/strong&gt; — filters by AGENT_ALLOWED status, deadline proximity, and tech relevance&lt;/li&gt;
&lt;li&gt;🐙 &lt;strong&gt;GitHub Bounty Scan&lt;/strong&gt; — scores issues by keyword relevance, extracts bounty amounts from labels&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;When there are no new results, the script produces empty output and Hermes Agent sends nothing. Silent by design.&lt;/p&gt;

&lt;h2&gt;
  
  
  Code
&lt;/h2&gt;

&lt;p&gt;Full source available at: &lt;strong&gt;&lt;a href="https://github.com/AtlasNexusOps/hermes-bounty-watcher" rel="noopener noreferrer"&gt;AtlasNexusOps/hermes-bounty-watcher&lt;/a&gt;&lt;/strong&gt;&lt;/p&gt;

&lt;h3&gt;
  
  
  My Tech Stack
&lt;/h3&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Hermes Agent&lt;/strong&gt; — scheduling, delivery, and orchestration&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Bash + Python&lt;/strong&gt; — data fetching, filtering, deduplication&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;GitHub CLI&lt;/strong&gt; — issue search and metadata extraction&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Superteam Earn API&lt;/strong&gt; — bounty listings&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Telegram&lt;/strong&gt; — delivery channel&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Android/Termux&lt;/strong&gt; — production runtime (ARM64, always-on)&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  How I Used Hermes Agent
&lt;/h2&gt;

&lt;p&gt;Hermes Agent is the backbone of this project — not a wrapper, not an afterthought. Here is exactly how it powers every layer:&lt;/p&gt;

&lt;h3&gt;
  
  
  1. Cron Scheduling (cronjob tool)
&lt;/h3&gt;

&lt;p&gt;The &lt;code&gt;cronjob&lt;/code&gt; tool schedules both scanners on fixed cadences:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;hermes cron add &lt;span class="s2"&gt;"every 360m"&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--name&lt;/span&gt; &lt;span class="s2"&gt;"Superteam Bounties"&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--script&lt;/span&gt; ~/.hermes/scripts/superteam_bounties.sh &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--no-agent&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--deliver&lt;/span&gt; origin

hermes cron add &lt;span class="s2"&gt;"every 180m"&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--name&lt;/span&gt; &lt;span class="s2"&gt;"GitHub Bounty Scan"&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--script&lt;/span&gt; ~/.hermes/scripts/github_bounties.sh &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--no-agent&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--deliver&lt;/span&gt; origin
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The &lt;code&gt;--no-agent&lt;/code&gt; flag means scripts run directly (zero LLM tokens). Hermes handles retries, timezone management, and deterministic delivery routing.&lt;/p&gt;

&lt;h3&gt;
  
  
  2. Watchdog Pattern (script collects, script filters, agent delivers)
&lt;/h3&gt;

&lt;p&gt;This is the architecture that makes the project cost-efficient:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight plaintext"&gt;&lt;code&gt;Script → API fetch → Python filter → dedup cache → output
                                                        ↓
                                            Hermes Agent delivers to Telegram
                                            (silent when output is empty)
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Zero LLM cost&lt;/strong&gt; for data collection — scripts run in shell/Python&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deterministic&lt;/strong&gt; — same input always produces the same filtered output&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Silent when empty&lt;/strong&gt; — watchdog pattern: only notifies when there is something new&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  3. Relevance Scoring Engine
&lt;/h3&gt;

&lt;p&gt;The filtering logic uses curated keyword lists to surface only matching opportunities:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="n"&gt;RELEVANCE&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;solana&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;agent&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;defi&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;crypto&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;trading&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;blockchain&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;smart contract&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;usdc&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;usdt&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;API&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;web3&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;rust&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;python&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;typescript&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;nft&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;token&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;dex&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;amm&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;oracle&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;bridge&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;validator&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;staking&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;airdrop&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;spl&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;jupiter&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;raydium&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;sdk&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;CLI&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;cli tool&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;automation&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;bot&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;scraper&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;$&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;reward&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;prize&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;earn&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt;grants&lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
&lt;span class="p"&gt;]&lt;/span&gt;

&lt;span class="n"&gt;all_text&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;title_lower&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt; &lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt; &lt;span class="o"&gt;+&lt;/span&gt; &lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="s"&gt; &lt;/span&gt;&lt;span class="sh"&gt;"&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;join&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;labels&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;span class="n"&gt;score&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="nf"&gt;sum&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="mi"&gt;1&lt;/span&gt; &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;kw&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="n"&gt;RELEVANCE&lt;/span&gt; &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;kw&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;lower&lt;/span&gt;&lt;span class="p"&gt;()&lt;/span&gt; &lt;span class="ow"&gt;in&lt;/span&gt; &lt;span class="n"&gt;all_text&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h3&gt;
  
  
  4. Why Hermes Agent and Not a Regular Cron?
&lt;/h3&gt;

&lt;p&gt;A regular cron job would just dump raw API output into a chat. Hermes Agent provides:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Delivery routing&lt;/strong&gt; — &lt;code&gt;--deliver origin&lt;/code&gt; ensures output goes back to the right Telegram chat&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;State management&lt;/strong&gt; — the agent maintains the cron lifecycle (pause, resume, remove, list)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Memory&lt;/strong&gt; — persistent cache for deduplication across sessions&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Composability&lt;/strong&gt; — cron jobs can chain (&lt;code&gt;context_from&lt;/code&gt;), allowing future expansion (e.g., a summarization job that reads the scanner output)&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Script-only mode&lt;/strong&gt; — &lt;code&gt;no_agent: true&lt;/code&gt; lets us bypass the LLM entirely for mechanical data collection while still benefiting from the delivery infrastructure&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  5. Production Results
&lt;/h3&gt;

&lt;p&gt;After weeks of continuous operation:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Superteam&lt;/strong&gt;: 15-25 active bounties per scan, 3-5 with AGENT_ALLOWED status&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;GitHub&lt;/strong&gt;: 5-10 relevant issues per scan, filtered from 30+ raw results&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Signal-to-noise&lt;/strong&gt;: ~30% of fetched results pass the relevance filter&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Novelty detection&lt;/strong&gt;: 2-4 new bounties per day flagged as new&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  How to Deploy It Yourself
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;git clone https://github.com/AtlasNexusOps/hermes-bounty-watcher.git
&lt;span class="nb"&gt;cp &lt;/span&gt;hermes-bounty-watcher/scripts/&lt;span class="k"&gt;*&lt;/span&gt; ~/.hermes/scripts/

hermes cron add &lt;span class="s2"&gt;"every 360m"&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--name&lt;/span&gt; &lt;span class="s2"&gt;"Superteam Bounties"&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--script&lt;/span&gt; ~/.hermes/scripts/superteam_bounties.sh &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--no-agent&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--deliver&lt;/span&gt; origin

hermes cron add &lt;span class="s2"&gt;"every 180m"&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--name&lt;/span&gt; &lt;span class="s2"&gt;"GitHub Bounty Scan"&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--script&lt;/span&gt; ~/.hermes/scripts/github_bounties.sh &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--no-agent&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--deliver&lt;/span&gt; origin
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;






&lt;p&gt;&lt;em&gt;Built with Hermes Agent, running on an Android phone, delivering value every day.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagentchallenge</category>
      <category>showdev</category>
      <category>automation</category>
      <category>opensource</category>
    </item>
    <item>
      <title>Running Hermes Agent on Android: A Production AI Agent in Your Pocket</title>
      <dc:creator>Alexandre Lasly</dc:creator>
      <pubDate>Sat, 16 May 2026 00:25:30 +0000</pubDate>
      <link>https://forem.com/athenaios/running-hermes-agent-on-android-a-production-ai-agent-in-your-pocket-18nb</link>
      <guid>https://forem.com/athenaios/running-hermes-agent-on-android-a-production-ai-agent-in-your-pocket-18nb</guid>
      <description>&lt;p&gt;&lt;em&gt;This is a submission for the &lt;a href="https://dev.to/challenges/hermes-agent-2026-05-15"&gt;Hermes Agent Challenge&lt;/a&gt;&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;em&gt;No server. No cloud bill. Just a phone, Termux, and an AI agent that ships code.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;When I first heard about Hermes Agent, the open-source agentic framework from Nous Research, I did what any reasonable developer would do: I tried installing it on my laptop. It worked. Then I asked myself — what if I could run it &lt;strong&gt;on my phone&lt;/strong&gt;, always on, always ready?&lt;/p&gt;

&lt;p&gt;Turns out you can. And it is surprisingly practical.&lt;/p&gt;

&lt;h2&gt;
  
  
  The Setup: Android + Termux
&lt;/h2&gt;

&lt;p&gt;Termux gives you a functional Linux environment on Android. With it, you get bash, Python, Node.js, git — everything you need. Installing Hermes Agent is a one-liner:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;pkg &lt;span class="nb"&gt;install &lt;/span&gt;python git
pip &lt;span class="nb"&gt;install &lt;/span&gt;hermes-agent
hermes setup
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The setup wizard walks you through picking a provider (I use DeepSeek), configuring tools, and connecting messaging platforms. Within 10 minutes, I had a fully operational AI agent running on a Samsung phone.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why Run an Agent on Your Phone?
&lt;/h2&gt;

&lt;p&gt;Three reasons:&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Always-on.&lt;/strong&gt; Your phone is already on 24/7. No separate device, no VPS, no cloud bill.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Messaging-native.&lt;/strong&gt; Hermes Agent connects to Telegram natively — you interact with it like a contact. Wake up, check your messages, ask your agent what happened overnight.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Capable.&lt;/strong&gt; Despite running on ARM64 with limited RAM, Hermes performs real agentic work: multi-step reasoning, tool orchestration, file manipulation, git operations, web search, and cron scheduling.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  How an AI Agent Let Me Work on an Entire GitHub Portfolio — From My Phone
&lt;/h2&gt;

&lt;p&gt;This is where Hermes Agent truly shines. Over the past weeks, I built and deployed an entire professional portfolio site — portfolio design, HTML/CSS edits, Git versioning, GitHub Pages deployment, README documentation, and even a separate open-source project — all through my agent, without touching a laptop.&lt;/p&gt;

&lt;p&gt;The agent did not just execute commands. It &lt;strong&gt;owned the workflow&lt;/strong&gt;:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;
&lt;strong&gt;Repository creation&lt;/strong&gt; — &lt;code&gt;gh repo create&lt;/code&gt;, initialize, push to main&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Content edits&lt;/strong&gt; — semantic find-and-replace via the &lt;code&gt;patch&lt;/code&gt; tool, not fragile line numbers&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Commits&lt;/strong&gt; — meaningful commit messages in French and English, contextual to the change&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Deployment&lt;/strong&gt; — push to main → GitHub Pages auto-deploys. Live in seconds.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Cross-repo work&lt;/strong&gt; — jumping between &lt;code&gt;atlasnexusops.github.io&lt;/code&gt; (portfolio) and &lt;code&gt;hermes-bounty-watcher&lt;/code&gt; (tool project) without losing context&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The agent maintained persistent memory of the repo structure, file paths, and even my stylistic preferences ("use purple gradients, not green"). It remembered that I prefer French commit messages for the portfolio and English for open-source repos. That is something no CLI can do.&lt;/p&gt;

&lt;h3&gt;
  
  
  A real session: from idea to live site in minutes
&lt;/h3&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Me:&lt;/strong&gt; Remplace bot par IA agentique professionnelle dans le portfolio&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Hermes:&lt;/strong&gt; &lt;em&gt;[searches for bot across the repo, finds line 223 in index.html, patches the exact string, commits with "Remplace bot par IA agentique professionnelle", pushes to main]&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Hermes:&lt;/strong&gt; Fait. C est en ligne.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Zero copy-paste. Zero SSH. Zero context switching. The agent had persistent knowledge of the repo structure, found the exact string, applied a semantic patch (not fragile line numbers), committed with a meaningful bilingual message, and pushed to main. GitHub Pages deployed within seconds.&lt;/p&gt;

&lt;p&gt;Later that same session:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;&lt;strong&gt;Me:&lt;/strong&gt; Maintenant on supprime les coordonnees du footer et tu peux mettre mon nom en footer Alexandre Lasly&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Hermes:&lt;/strong&gt; &lt;em&gt;[navigates to footer section, removes contact links, adds name, commits, pushes]&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Hermes:&lt;/strong&gt; Fait.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;h3&gt;
  
  
  The whole GitHub surface area
&lt;/h3&gt;

&lt;p&gt;Over multiple sessions, the agent managed:&lt;/p&gt;

&lt;div class="table-wrapper-paragraph"&gt;&lt;table&gt;
&lt;thead&gt;
&lt;tr&gt;
&lt;th&gt;Action&lt;/th&gt;
&lt;th&gt;Tool&lt;/th&gt;
&lt;/tr&gt;
&lt;/thead&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td&gt;Create repo + push&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;terminal&lt;/code&gt; (gh CLI)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Edit HTML/CSS&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;patch&lt;/code&gt; (semantic replace)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Write new files&lt;/td&gt;
&lt;td&gt;&lt;code&gt;write_file&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Search codebase&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;search_files&lt;/code&gt; (ripgrep)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Read with context&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;read_file&lt;/code&gt; (paginated)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Git commit + push&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;terminal&lt;/code&gt; (git)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Schedule cron jobs&lt;/td&gt;
&lt;td&gt;&lt;code&gt;cronjob&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Deliver updates&lt;/td&gt;
&lt;td&gt;
&lt;code&gt;send_message&lt;/code&gt; (Telegram)&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td&gt;Remember preferences&lt;/td&gt;
&lt;td&gt;&lt;code&gt;memory&lt;/code&gt;&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;

&lt;p&gt;Every GitHub operation — from init to deploy — went through the agent. No laptop. No IDE. Just Telegram messages on a phone.&lt;/p&gt;

&lt;h2&gt;
  
  
  Cron: The Secret Weapon
&lt;/h2&gt;

&lt;p&gt;Hermes Agent is cron system deserves special attention. You define a job once, and it runs on schedule — fully autonomous:&lt;br&gt;
&lt;/p&gt;

&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;hermes cron add &lt;span class="s2"&gt;"every 6h"&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--script&lt;/span&gt; ~/.hermes/scripts/superteam_bounties.sh &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--no-agent&lt;/span&gt; &lt;span class="se"&gt;\&lt;/span&gt;
  &lt;span class="nt"&gt;--deliver&lt;/span&gt; origin
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;The &lt;code&gt;--no-agent&lt;/code&gt; flag means the script runs directly (zero LLM tokens) and its stdout is delivered verbatim. Silent when there is nothing to report.&lt;/p&gt;

&lt;h2&gt;
  
  
  Lessons Learned
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt;
&lt;strong&gt;Memory is underrated.&lt;/strong&gt; Hermes Agent is persistent memory means it remembers your preferences, project structure, and past fixes.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Skills are living documentation.&lt;/strong&gt; When you solve a complex workflow, save it as a skill.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;No-agent crons are efficient.&lt;/strong&gt; For repetitive data collection, skip the LLM entirely.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;The phone is not a toy.&lt;/strong&gt; An Android phone becomes a production-grade AI workstation.&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Agents own workflows, not just commands.&lt;/strong&gt; The difference between a CLI wrapper and an agent is memory, planning, and context retention across sessions.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Try It Yourself
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight shell"&gt;&lt;code&gt;pkg &lt;span class="nb"&gt;install &lt;/span&gt;python git &lt;span class="nt"&gt;-y&lt;/span&gt;
pip &lt;span class="nb"&gt;install &lt;/span&gt;hermes-agent
hermes setup
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;p&gt;Connect a Telegram bot, schedule a cron, send a message — you will have an AI co-worker in your pocket within the hour.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Hermes Agent is open source. Runs anywhere you can install Python — including the phone in your hand right now.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>hermesagentchallenge</category>
      <category>ai</category>
      <category>android</category>
      <category>opensource</category>
    </item>
  </channel>
</rss>
