The full Hermes track — not a feature tour but a climb up the delegation ladder. Nine hands-on lessons take you from a first install (cloud or a local LM Studio model) through giving the agent a memory of you, reaching it from your phone, letting it act under your approval, writing its own skills, and running standing errands — ending with one real always-on assistant doing a job for you. Plus a command + delegation cheat sheet and starter skills to drop in.
A remodelling-company owner built a roofing lead-finder with Hermes instead of paying for an expensive lead-gen platform.
→A pipeline for finding local jobs at the cost of tokens, not a monthly SaaS bill.
Try it yourself
Find homeowners in [my service area] who have posted about roof damage or remodelling needs, save the leads to memory, and send me a daily Telegram summary.
Hermes surfaces an email-finding tool (Hunter.io via Composio) inside an outreach workflow.
→Prospecting and first-draft outreach handled by the agent instead of by hand.
Try it yourself
Given this list of target companies, find a likely contact email for each, draft a short personalised intro, and put it all in a table for me to review before anything sends.
A skill set covering CRM, invoicing and project-management workflows so you run the back office by talking to the agent.
→The admin side of a small business handled from a chat window.
Try it yourself
Save a new contact, log a call note, create an invoice, and update a project status — all in one conversation — and remember my CRM layout for future sessions.
An open dashboard to manage fleets of agents, dispatch tasks, and track costs across them.
→Multi-agent work becomes observable and budgetable instead of a black box.
Try it yourself
Build me a control-room dashboard for my agent fleet — list every running agent, let me dispatch tasks to each, and show a live cost tally so I can see what each agent is spending.
A plugin records all tool and API calls to SQLite with Grafana dashboards profiling token spend across Telegram, WhatsApp and cron. Needs the token-audit plugin installed first.
→Finance-grade cost and usage visibility before you scale agents to a team.
Try it yourself
Start recording all my tool and API calls to SQLite and set up a Grafana dashboard showing daily spend by channel.
A main agent plans, a coder sub-agent implements, a QA sub-agent tests, and failures loop back to repair before it ships — a one-person build team on cheaper models.
→A solo founder gets a plan-build-test-fix loop that normally needs a small team.
Try it yourself
Act as a one-person build team for [describe your feature] — plan it yourself, implement it with a coder sub-agent, test it with a QA sub-agent, loop back on failures, and message me on Telegram when it ships.
A community skill that turns any startup idea into eight investor-ready documents — market analysis, competitor map, pitch deck — plus a landing page. Needs the startup-architect skill installed first.
→A fundraising starter pack drafted in one prompt, ready for you to sharpen.
Try it yourself
Turn this idea — [your idea in one sentence] — into a full investor-ready kit: market analysis, competitor map, pitch deck outline, and a landing page draft.
A builder switched to Hermes and shipped five small apps in a single day, leaning on the agent's self-learning to speed up each build.
→Idea-to-launch compresses to hours when the agent remembers your stack between builds.
Try it yourself
I want to ship [your app idea] today — plan the simplest working version, build it, write a reusable skill for any repeating patterns, and tell me when each milestone is done so I can keep the momentum.
A main agent with cross-project memory spawns a sub-agent per project or Slack channel, runs on a VPS, and sends a daily WhatsApp report.
→One coordinating brain that remembers every project — and reports proactively.
Try it yourself
Act as my Chief of Staff — keep cross-project memory, spin up a sub-agent for each of my projects, and send me a WhatsApp summary every evening at 7pm covering progress, blockers, and what needs my attention.
After ~10 days the agent internalises a developer's review preferences — which files to read first, patterns to flag, how to format output.
→Code review that fits your team’s taste because it actually remembers it.
Try it yourself
Study my repo at [path] — learn which files matter most, note patterns to flag in reviews, and remember my preferred output format. After ten sessions, save everything you have learned as a codebase-review skill.
A second agent monitors Hermes’s workflows in real time, catches breaks as they happen, and fixes them mid-run.
→Long autonomous runs that self-correct instead of silently failing.
Try it yourself
Spin up a monitor sub-agent that watches my main workflow in real time, catches any tool failure or loop, fixes it mid-run if it can, and pings me on Telegram if it needs my input.
A multi-agent orchestration layer that splits a goal into research, planning and execution agents.
→A repeatable way to throw several agents at one objective without chaos.
Try it yourself
For this goal — [your goal] — split the work into three specialist agents: one for research, one for planning, and one for execution. Coordinate them and report back with the final output when done.
An AI-native SDLC built on a code graph so the agent understands the whole repo before it changes anything.
→Bigger changes land safely because the agent reasons over the full codebase.
Try it yourself
Index my entire repo into a code graph, then implement [your feature or bug fix] — read the graph to understand dependencies before touching anything, and show me a diff for review before committing.
A three-actor email orchestration on DBOS + PostgreSQL + S3 with the Gmail API, running in production for weeks.
→Proof Hermes holds up as real backend infrastructure, not just a demo.
Try it yourself
Build a three-actor email pipeline on PostgreSQL: one actor fetches new Gmail threads, one drafts replies in my voice, one sends approved drafts — save it as a skill and run it every 30 minutes.
Merge to main → an image builds → a watcher restarts the agent automatically, so the agent ships its own updates.
→Your agent improves on a normal git workflow, hands-off.
Try it yourself
Set up a CI/CD loop for this repo — on every merge to main, build a new image, restart the agent automatically, and message me on Telegram confirming the version that is now live.
An MCP server that exposes nine Hermes tools (terminal, file I/O, web search, memory, skills) to any MCP client like Claude Desktop. Needs the Hermes MCP server skill installed first.
→Your favourite chat app gets Hermes’s hands without leaving its window.
Try it yourself
Connect to my MCP client and expose your terminal, file I/O, web search, and memory tools so I can use them from any MCP client without switching apps.
A consultant ran a production Hermes for 297+ days straight, automating over $100K of client value.
→Evidence a self-hosted agent can be durable, revenue-generating infrastructure.
Try it yourself
Take over [my recurring client deliverable — weekly reports / data pulls / outreach sequences], run it automatically every week, and log completed work to memory.
A shared agent reachable over WhatsApp, with a different use case per family member, for a one-time ~$200 cost.
→Several paid AI subscriptions collapse into one self-hosted agent.
Try it yourself
Set up a shared WhatsApp connection and remember a different use case per family member — [grocery lists for one, homework help for another] — so everyone gets a personalised assistant.
A personal Hermes running a strong cloud model via OpenRouter on a $10/month Hetzner box.
→An always-on personal agent for the price of a couple of coffees.
Try it yourself
Walk me through deploying yourself as a systemd service on a $10/month Hetzner VPS using a strong cloud model via OpenRouter, then message me on Telegram once you are running.
A managed Hermes-hosting service — bring an API key and a Telegram bot, they run the box.
→A self-hosted-style agent for non-technical owners who don’t want a server.
Try it yourself
Guide me through connecting my API key and Telegram bot token to a managed Hermes host so I have a personal agent running with no server setup on my part.
Two Telegram community groups served by one gateway — same personality and memory, but specialised per group.
→Run multiple audience channels from a single agent without them bleeding together.
Try it yourself
Connect to my two Telegram group chats, remember a distinct personality and topic focus for each, and use the same memory backend so context carries across both groups.
A starter where the agent summarises yesterday’s progress across your projects each morning and asks what to prioritise, then recaps at day’s end.
→A self-running operating rhythm for a one-person business.
Try it yourself
Every weekday at 9am, summarise where each of my projects stands and ask me what to prioritise. At 6pm, recap what moved and flag anything stuck.
Hermes Small biz
Course starter
Turn receipts into a clean expense log
A starter where you forward photos of receipts on Telegram and the agent extracts vendor, amount and date into a running spreadsheet.
→Bookkeeping that happens as you spend, not in a Sunday-night panic.
Try it yourself
When I send you a receipt photo on Telegram, read the vendor, total and date and append a row to my expenses sheet. Each Friday, send me the week’s total by category.
Hermes AI-native SME Founder
Real build
Run the company as a four-agent team
A 24/7 setup runs four Hermes agents — PM, Dev, DevOps and Content — sharing five MCP servers and 34 tools, coordinated by a watchdog.
→A blueprint where each function has an always-on agent doing the routine work.
Try it yourself
Set up a four-agent team for my company with a PM, Dev, DevOps and Content agent sharing the same MCP tools, and message me on Telegram when each agent completes a task.
A printing-factory operations manager wrote a skill that auto-categorises tasks and compresses completed work into summary cards so nothing gets lost.
→A non-tech SME turning a real operational pain — forgetting things — into a trusted AI workflow.
Try it yourself
Create a skill that auto-categorises my team’s incoming tasks by type and compresses completed ones into weekly summary cards saved to a shared folder.
A community skill that makes Hermes an autonomous SRE — detecting, diagnosing and healing production infra, and learning from every incident. Needs the hermes-incident-commander skill installed first.
→On-call that gets smarter over time, combining memory, cron, messaging and sub-agents.
Try it yourself
Monitor my main service, auto-diagnose the next alert, attempt a fix, and message me on Telegram with the outcome.
A stack underneath Hermes adds tamper-proof audit logs, prompt-injection blocking, memory encryption, cost tracking and EU AI Act exports.
→A path to running agents in a regulated company without crossing compliance lines.
Try it yourself
Add tamper-proof audit logging and prompt-injection blocking to my current agent setup, then export an EU AI Act compliance report for my last 30 sessions.
Hermes on a local Kubernetes cluster for isolation, producing a daily cybersecurity + AI briefing.
→Enterprise-grade isolation with a useful daily output to justify it.
Try it yourself
Deploy to my local Kubernetes cluster in an isolated namespace and schedule a daily cybersecurity and AI briefing delivered to my Telegram every morning at 7am.
Claude handles chat and research while Hermes does the execution — email, browsing, forms — 24/7 on a mini PC.
→A clean division of labour: one brain to think, one to do.
Try it yourself
Handle all my execution tasks — email, browsing, form submissions — triggered by my Claude chat sessions, and remember my workflow preferences across both agents.
Connecting Hermes to a managed long-term memory layer (retain / recall / reflect) for reliable recall at scale.
→Org knowledge that persists and is retrievable beyond a single machine.
Try it yourself
Connect to a managed long-term memory backend, remember my top five client requirements, then retrieve them accurately in a new session to confirm reliable recall.
A security skill that protects against common LLM threats with recommended policies — important before you let a team agent touch real systems. Needs the security-hardening skill installed first.
→A safer baseline for an agent that has real access.
Try it yourself
Apply the security-hardening skill’s recommended policies to my current setup and report which LLM threat vectors are now blocked.
A starter where the agent holds your team's playbooks as skills, answers "how do we do X here?" in Slack, and writes a new skill whenever someone teaches it something.
→The knowledge that usually lives in three people’s heads becomes a shared, growing agent.
Try it yourself
Watch our team Slack. When someone explains how we do a recurring task, save it as a skill. When anyone asks "how do we do X?", answer from those skills and cite who taught it.
Hermes Founder AI-native SME
Real build
A full agent stack in one command
A community installer that stands up Hermes with 29 pre-configured plugins in a single command.
Try it yourself
Run the Evey community installer to stand up Hermes with 29 pre-configured plugins in a single command on my machine. Needs the evey-setup skill installed first.
A GEPA + DSPy loop that optimises the agent's skills, prompts and tool text into PR-ready improvements. Needs the GEPA + DSPy self-evolution plugin installed first.
Try it yourself
Analyse my last week of sessions and generate PR-ready skill and prompt improvements using the self-evolution loop.
A deployment pattern running Hermes as a Fly Machine with a volume-backed data dir for ~$15/month.
Try it yourself
Generate a Fly.io deployment config for yourself with a volume-backed data directory, then walk me through deploying it as a Fly Machine for roughly $15 per month.
A community plugin that watches sessions and turns repeated workflows into reusable skills automatically. Needs the hermes-skill-factory plugin installed first.
Try it yourself
Watch my sessions over the next week and automatically package any repeated workflows into reusable skills.
A community skill that delegates tasks across Hermes, Codex and Claude through the Agent Client Protocol. Needs the hermes-agent-acp-skill installed first.
Try it yourself
Break my current task — [task] — into sub-tasks and delegate the coding parts to Codex and the reasoning parts to Claude via ACP.
A built-in skill for reading and writing Airtable bases through its REST API.
Try it yourself
Turn on the Airtable skill, then read my contacts base, add a new record for a lead I just met — [name, company, notes] — and message me on Telegram when it’s saved.
A built-in maps skill for geocoding, points of interest, routes and timezones — handy for any field-service business.
Try it yourself
Turn on the maps skill, then find the five nearest [plumbers] to [job-site address], calculate drive time from my depot, and message me the ranked list on Telegram.
A team bot with per-user sessions, an isolated terminal and an allowlist — one agent the whole team can message. Needs the team-telegram skill installed first.
Try it yourself
Configure the allowlist with my team’s Telegram usernames [paste list] and give each member their own session with a shared terminal.
A documented pattern running authorised security assessments with Kali tools, scope confirmation and approval gates. Needs the pentest skill installed first.
Try it yourself
Confirm the target scope [paste scope], then run an authorised assessment against my staging environment [URL] and produce a findings report.
Hermes learns your style from past posts and scripts, remembers your preferred phrasing and emojis, and drafts LinkedIn posts or tweets that sound like you.
→On-brand drafts that stay consistent because the agent remembers your voice.
Try it yourself
Read my last 20 posts and learn my voice. When I paste a rough idea, draft three posts in my style and save anything I tweak so you sound more like me next time.
The agent proposed the workflow itself: pull your X lists and bookmarks, structure them, and feed them into NotebookLM to produce a podcast episode.
→Your saved reading turns into a listenable show — a pipeline the agent designed.
Try it yourself
Pull my X bookmarks and list items from the past week, structure them into topics, and generate a NotebookLM-ready source document I can turn into a podcast episode.
Three-times-daily news triage sorted into Discord channels by urgency, auto-linked to your current video projects.
→You always know what’s fresh enough to keep a video relevant.
Try it yourself
Schedule a news triage three times a day, sort stories into my Discord channels by urgency, and automatically link each story to whichever of my current video projects it relates to.
A media artist extended a TouchDesigner skill with reference docs so the agent helps build generative, interactive pieces. Needs the TouchDesigner skill installed first.
→Creative-coding assistance for live visuals and installations.
Try it yourself
Build a generative visual patch that reacts to audio input and explain each node you add.
A starter where the agent proposes next week’s posts from your recent work and what’s trending, then schedules drafts for your review.
→You approve a calendar instead of staring at a blank one.
Try it yourself
Every Friday, look at what I published this week and what’s trending in my niche, propose next week’s 5 posts with hooks, and save drafts for me to approve.
Hermes Small biz Creator
Built-in skill
Generate a client deck
A built-in skill that creates and edits PowerPoint (.pptx) presentations.
Try it yourself
Take these bullet points and build a 10-slide client proposal deck with a clean title slide and a summary at the end.
Built-in songwriting skills that draft lyrics and matching Suno AI music prompts.
Try it yourself
Use the suno skill to write upbeat lyrics and a matching Suno prompt for a 30-second intro jingle for my podcast [podcast name], then save both to a text file.
A daily agent on a lab workstation pointed at a local model summarises each morning’s new PubMed alerts and learns which topics you actually care about.
→Zero-cost and fully local — abstracts and queries never leave the lab, and it gets more relevant the longer you use it.
Try it yourself
Every weekday at 8am, search PubMed for new papers on my saved topics, summarise the 5 most relevant, message me the digest, and weight future digests toward the ones I open.
Hermes Scientist
Real build
Autonomous experiment bookkeeper
Hermes-Lab: give it a search space and an evaluation method and it schedules runs, tracks every result, and suggests what to try next. Needs the Hermes-Lab skill installed first.
→The bookkeeping of an ML sweep runs itself; you just decide the next direction.
Try it yourself
Here is my search space and evaluation function — schedule runs, track every result in memory, and message me on Telegram with your next suggested experiment after each batch.
ChEMBL, AlphaFold, OpenFDA and QSAR workflows wired into Hermes skills to run early drug-discovery queries from a laptop. Needs the drug-discovery toolkit skill installed first.
→Heavyweight computational-chemistry workflows become a chat you can run anywhere.
Try it yourself
Using ChEMBL and OpenFDA, find approved compounds similar to [your target molecule], run a QSAR filter, and save the shortlist as a skill I can re-run weekly.
Walk Hermes through cleaning one messy instrument-export CSV; it writes a reusable skill the first time and applies it automatically next week.
→Teach a fiddly procedure once; the agent owns it from then on and gets faster every run.
Try it yourself
Help me clean this instrument CSV step by step. When we're done, save what we did as a reusable skill so you can clean the next export the same way without me re-explaining.
Hermes Scientist
Real build
Vectorless RAG over your own papers
A retrieval setup (PageIndex) where the agent understands a document's structure and reasons over it with tools instead of chopping it into vector chunks. Needs the PageIndex skill installed first.
→Answers that respect how a paper is actually organised, with fewer hallucinated links.
Try it yourself
Index these papers from my folder and answer [your research question] by reasoning over their structure with tools — no vector chunks.
A data scientist ported an entire Python weather stack (MetPy, Herbie, cfgrib, WRF-Python) to Rust and built two Hermes plugins around it. Needs the weather-stack plugin installed first.
→A slow, fragile scientific pipeline becomes a fast, agent-drivable tool.
Try it yourself
Run my MetPy/Herbie pipeline on today’s GFS data, save the output plots to my results folder, and schedule this every morning at 6am.
A research stack of seven MCP tools plus a self-hosted SearXNG search to replace a paid research API.
→Deep web research on your own infrastructure, squeezing the free tiers instead of paying per query.
Try it yourself
Build me a private research engine using SearXNG plus web, PDF, and citation tools — save it as a skill called deep-research so I can call it with any question and get a cited summary.
Watches your field, picks the real signals, writes a brief, suggests angles, and delivers via Discord, Slack, Notion, email, Obsidian or markdown.
→One self-improving morning brief instead of ten tabs you never get to.
Try it yourself
Each morning, scan my field's top sources, pick the 5 developments that matter, write a short sourced brief, and post it to my Slack. Learn which items I react to.
A community skill that turns Hermes into an autonomous research agent for surveying and summarising a topic. Needs the literature-research-agent skill installed first.
→A first-pass literature scan that runs while you do something else.
Try it yourself
Survey and summarise the last two years of papers on [your topic], save the summary to memory, and message me the key findings on Telegram.
Reads Hacker News daily and emails a curated summary of what's worth your attention.
→The signal from a noisy feed, on a schedule, without the doomscroll.
Try it yourself
Every morning at 7am, read Hacker News, filter for science and engineering stories relevant to [your field], and email me a curated brief with your picks and one-line summaries.
A built-in research-paper-writing skill that structures an ML paper for a conference from your results.
Try it yourself
Use the research-paper-writing skill to draft a conference-style ML paper from my results, structuring it with abstract, related work, methodology and evaluation sections.
Built-in skills to run local inference (vLLM for throughput, llama.cpp for GGUF) so the whole group queries one private model.
Try it yourself
Use the vllm skill to spin up a local inference server for [my chosen model], then share the endpoint URL with my team so everyone queries one private instance.
A built-in skill that runs Python step-by-step against a live Jupyter kernel, keeping state between cells.
Try it yourself
Use the jupyter-live-kernel skill to load my dataset, run an exploratory analysis step-by-step in a live kernel, and keep the variable state ready for follow-up queries.
A built-in skill to pull, push and manage Hugging Face models and datasets from the agent.
Try it yourself
Use the huggingface-hub skill to pull the latest checkpoint for my fine-tuned model, evaluate it on my test split, and push the updated weights back to my repo.
A built-in skill that monitors blogs and RSS/Atom feeds and flags what is new and relevant.
Try it yourself
Use the blogwatcher skill to monitor my five favourite ML blogs and Nature RSS feeds, then message me on Telegram whenever a post matches my research keywords.
A community plugin exposing National Weather Service models and NEXRAD radar imagery to the agent. Needs the hermes-weather-plugin skill installed first.
Try it yourself
Pull the latest NWS forecast and NEXRAD radar imagery for [my study region] and summarise current conditions.
Hermes drove TurboQuant to optimise an MLX model (Qwen3.5-9B) for local inference on a Mac.
Try it yourself
Drive TurboQuant to quantize [my target model] for Apple Silicon, benchmark inference speed before and after, and save the optimised MLX weights locally.
A single question fanned out across Reddit, X, YouTube, Hacker News and Polymarket, then synthesised.
Try it yourself
Fan out my research question '[my topic]' across Reddit, X, YouTube, Hacker News and Polymarket, then synthesise the key signals into a one-page brief.
A community project running a Mars rover simulation through ROS2 and Gazebo, driven by Hermes. Needs the Hermes-mars-rover community skill installed first.
Try it yourself
Run a navigation scenario in the ROS2/Gazebo Mars-rover simulator and log the rover’s path to a CSV.
A community project fine-tuning vision-language-action models so the agent improves a robot over time. Needs the hermes-embodied community skill installed first.
Try it yourself
Run one training iteration on my robot demonstration data using the vision-language-action fine-tuning skill and report the improvement metric.
A personal knowledge base on a small VPS that the agent — not you — maintains, with a chat-bot interface and a static site.
→Your notes compound over time instead of rotting, because something actively tends them.
Try it yourself
Set up a self-maintaining research wiki on my VPS — watch my notes folder, update wiki pages when I add or edit files, and expose a chat interface so I can query it on Telegram.
Structured markdown notes in a synced Obsidian vault act as durable memory that survives context resets (a widely-upvoted community pattern).
→The agent’s knowledge lives in files you own and can read, not a black box.
Try it yourself
Use my Obsidian vault at [path] as your long-term memory — read existing notes at session start, write a structured summary note after each conversation, and sync it so my vault stays current.
A knowledge substrate that builds a qualitative "map" over existing Obsidian/vimwiki notes and sessions.
→Your scattered notes become a navigable structure instead of a flat pile.
Try it yourself
Read my Obsidian notes at [path], build a semantic map of the concepts and how they connect, save it as a skill, and let me query it by asking things like 'what links to [topic]?'
Pulls Whoop fitness data locally and exposes Hermes tools so you can ask custom questions over your own biometrics.
→Your health data stays on your machine while still being queryable like a dataset.
Try it yourself
Connect to my Whoop data locally, build tools for HRV, strain, and recovery trends, then schedule a morning message on Telegram each day with yesterday’s key metrics and any anomalies.
Hermes autonomously wrote "Job Scout" — searches job boards, scores listings against your CV, researches companies, tracks applications. It wrote and tested all the code.
→A bespoke tool, coded by the agent on request — and a clean demo of Hermes building software.
Try it yourself
Write a Job Scout skill: search LinkedIn and Indeed for roles matching my CV, score each listing, research the companies, and message me on Telegram with the top five matches each morning.
A "Meal Manager" plugin that tracks fridge inventory in natural language and scores suggestions by availability and recency — a clean pattern for any light inventory job. Needs the Meal Manager plugin installed first.
→A reusable template for low-friction inventory the agent keeps in its head.
Try it yourself
Here’s what’s in my fridge: [list]. Suggest three dinners ranked by ingredient availability and how long each item has been sitting.