What it actually does
Most agents forget everything the moment a session closes. Hermes is built around the opposite premise: a learning loop. It creates skills from things it has actually done, refines them during use, nudges itself to persist what it learned, searches its own past conversations, and builds a deepening model of who you are over time.
The second design decision is that it does not live on your laptop. Hermes runs on a $5 VPS, a GPU cluster, or serverless infrastructure that costs almost nothing when idle — and you talk to it from Telegram, Discord, Slack, WhatsApp, Signal, or the CLI through a single gateway process. Send a voice memo from your phone and it keeps working on a cloud VM.
Model choice is deliberately unopinionated. Nous Portal, OpenRouter, OpenAI, or your own endpoint — switch with `hermes model`, no code changes and no lock-in. The terminal interface is a real TUI with multiline editing, slash-command autocomplete, interrupt-and-redirect, and streaming tool output.
At around 228,000 stars it is one of the most starred agent projects on GitHub, which also means a large issue queue and a fast-moving surface.
Who it's for
- 01
Founders who want an agent that is reachable from a phone, not tethered to an open laptop
- 02
Small teams that need long-running work to continue after they close the lid
- 03
Anyone tired of re-teaching an agent the same context every session
- 04
Operators who want to switch model providers without rewriting their tooling
Where it earns its keep
- A persistent assistant that accumulates real knowledge of your business over months
- Long-running research or monitoring jobs that report in when something changes
- Running an agent from Telegram while travelling, with no laptop involved
- Consolidating several one-off scripts into one agent that learns which to use
- Avoiding provider lock-in while you evaluate which model actually performs for your work
Use it, or skip it
Reach for it when
- You want continuity — an agent that is meaningfully better in month three than month one
- The work outlives a single session and needs somewhere to run
- Reaching the agent from chat apps matters more than a polished desktop UI
- You expect to change model providers and want that to be a config change
Skip it when
- You need a stable, slow-moving dependency. A project this large and this fast-moving will churn
- Your compliance posture cannot accommodate an agent with persistent memory of business conversations
- You want a supported commercial product with an SLA rather than a self-hosted one
- Your use case is a single deterministic task — a cron job and a script will be cheaper and more predictable
10 automations
Ideas, not tutorials. Each one is work a team does by hand today.
- 01Founders
Founder's morning briefing
Overnight, pull revenue, signups, support volume, and anything unusual, then send a single Telegram message before the first meeting.
- 02Marketing
Always-on competitor watch
Monitor competitor pricing, changelogs, and job postings continuously, and message you only when something material changes.
- 03Founders
Voice-memo task capture
Send a voice note between meetings; the agent transcribes it, works the task on its VM, and reports back when it is done.
- 04Operations
Skill accumulation from repeated asks
When you ask for the same report three times, let it write itself a skill so the fourth request runs instantly.
- 05Sales
Deal-desk escalation
Watch for deals that stall past a threshold and ping the rep on Slack with the specific next action from past conversation history.
- 06Sales
Inbound lead qualification
Enrich each new signup, score fit against your ICP, and drop the qualified ones into a channel with the reasoning attached.
- 07Engineering
On-call summariser
During an incident, keep a running plain-English timeline and post the summary to WhatsApp for people not at a keyboard.
- 08Marketing
Content repurposing loop
Turn each published post into channel-specific variants, learning from engagement which framings work for your audience.
- 09Finance
Model cost comparison
Run the same recurring job across two providers for a week and report which produced better output per dollar.
- 10Support
Customer memory across channels
Search its own conversation history so that when a customer name comes up, the full backstory arrives with it.
Want one of these running by Friday?
LimeDock builds these as real workflows inside your stack — deployed to your cloud, wired into your Slack and CRM, with the code in your repo. You pay a build fee and your own API keys, nothing else.
Source
Repository stats were read from the GitHub API and reflect the last time we refreshed this entry. The editorial breakdown above is LimeDock’s own analysis — we are not affiliated with NousResearch.
https://github.com/NousResearch/hermes-agent