Hermes Agent by Nous Research is an open-source, self-hosted agent that learns skills as it goes. Rerun gives you self-improving agents without the server to run them on.
Last reviewed , against each product’s public pricing and docs.
Pick Hermes Agent if you want MIT-licensed code, a choice of terminal backends and full control of the runtime. Pick Rerun if you want agents that learn and run 24/7 without provisioning, patching or watching a server, with a dashboard and approvals your team can use.
Rerun does what they do, without the terminal.
Terminal install, local or on a serverYour machine, a VPS or a sandbox you runUpdates and uptime are on youTerminal UI and logsBuilt for one personDeveloper, terminal, configFree software, plus model and hostingFour differences that change your day to day, shown on the real product.
Hermes Agent runs where you install it, a laptop, a VPS or a sandbox, and updates and uptime are on you. Rerun agents run on a private Box that is provisioned, isolated and updated for you.
A payment just failed
Margo is chasing it
New email from a lead
Jules is drafting a reply
Every morning at 7:00
Sam sorted your inbox
Rerun agents also build skills and memory from experience, and you see what they learned and every action they took on a live dashboard.
## Procedure
1. Pull unpaid invoices from Stripe daily
2. Send a first reminder on day 7 past due
3. Escalate to the account owner after 2 tries
4. Log every touch in the CRM thread
## Exceptions
- Never chase an invoice under $50
- Read references/tone.md before escalating
## Billing
- Pays around day 30, sometimes later
- A PO number is required on every invoice
- Invoices go to ap@acme.co, not to Dana
## People
- Dana Reyes runs ops, answers within a day
- Escalations go through their COO
## Voice
- English, formal, no emoji
- Sign off as the founder, not an assistant
- Keep it under 120 words
## Timing
- Follow up twice, then stop
- Send whenever the draft is ready
Agents stop before anything consequential and ask in the app, and any teammate on the workspace can answer.
Connect 190+ apps or any MCP server, and run on our models, your subscription, your API key or a local model.
No migration project. Start with the one task that costs you the most, and let it prove itself.
Give the agent the job in plain words. It writes its own instructions and skills.
Keep the model provider you like, and connect your tools in one click.
The agent now runs on its Box with its own schedule, and you stop maintaining the machine Hermes ran on.
Hermes Agent is free and MIT-licensed. You pay for model usage and for wherever it runs, a local machine or a remote backend, and you maintain it. Rerun is one plan from $24 a month with the private Box, model credit and unlimited runs included.
Everything you need to put agents to work
Add agents, Boxes, seats or model credit whenever you need them.
7-day free trial · cancel anytime
3 agents
Three agents running side by side, each with its own job and tools.
1 Box
Your own private computer in the cloud, where your agents live. Only yours, and your data stays inside it.
1 seat
Your own workspace, with extra seats whenever your team grows.
Bring your own AI
Optional: your own key, subscription or local model.
190+ apps
All your tools ready to connect, plus any MCP or API.
Unlimited executions
No run quotas. Your agents work around the clock.
Agent & Skill library
Ready-made agents and skills, just drag and drop.
Looking for a Hermes AI alternative? We rank and compare Hermes Agent, OpenClaw forks, Claude Code, Lindy and Rerun on security, control, and governance, and explain why a human-in-the-loop agent wins.
Hermès Agent is free and open-source, so why do real bills range from $8 to $200+ a month? A full breakdown of hosting, tokens, and the hidden ops costs, plus the transparent managed alternative.
An honest, head-to-head comparison of Hermes Agent and OpenClaw across architecture, memory, security, setup, and cost, plus when self-hosting either agent is the wrong call.
Rerun does what they do, without the terminal.
Read the comparisonRerun does what they do, without the terminal.
Read the comparison
Rerun does what they do, on the model you choose, with its own plan.
Read the comparisonRerun does what they do, without building the Zap.
Read the comparisonRerun does what they do, without wiring the scenario.
Read the comparisonRerun does what they do, without wiring the workflow.
Read the comparisonYes. Rerun gives you agents that learn and run around the clock, without installing or hosting anything. Each agent lives on your private Box in the cloud.
Yes. Rerun agents build skills and keep memory between runs, and they get sharper each time you correct them. You can read what they learned from the dashboard.
The software is free. The real cost is the model usage, the machine or sandbox it runs on, and the time spent maintaining it. Rerun bundles the Box and model credit into one plan from $24 a month.
Rerun runs on Claude, ChatGPT and more, and you can bring your own key or custom endpoint, or a local model.
No. You create and manage agents in the browser.
© 2026 Rerun. All rights reserved.
