What is Hermes Agent?
Hermes Agent is an open-source personal agent from Nous Research, the lab behind the Hermes family of open models. It runs as a terminal app or as a long-lived gateway process that you message from Telegram, Discord, Slack, WhatsApp, Signal or email, and it works with any model provider, including OpenRouter, OpenAI, your own endpoint or Nous's own Portal.
Its selling point is a learning loop: after a complex task it can write a reusable skill, keep notes in a curated memory, and search its past sessions with SQLite full-text search. It also schedules jobs with a built-in cron, spawns subagents for parallel work, connects to MCP servers and can import settings, memories and skills from OpenClaw, Claude Code, Codex and Cursor.
Commands can run on the host or in one of six terminal backends: local, Docker, SSH, Modal, Daytona and Vercel Sandbox. The repository was created in July 2025, is MIT-licensed and ships often: v0.22.5 came out on 23 October 2026, one of at least 30 releases since late May.
Is Hermes Agent worth it?
Hermes Agent has become the default open-source answer to the always-on assistant, and its scale is now both its strength and its main risk. The repository has about 257,000 GitHub stars, and it also carries roughly 17,200 open issues and 34,000 open pull requests, a backlog no maintainer team can triage closely.
The design is sound for an agent that lives on a server rather than a laptop. One gateway process serves six messaging platforms, cron jobs deliver reports unattended, and the Daytona and Modal backends hibernate the agent's environment when idle, so a VPS-plus-API-key setup costs little between tasks. Skills follow the agentskills.io format, which means what Hermes learns is a folder of files you can read, edit or delete rather than an opaque memory.
The security documentation is more candid than most. By default a helper model rates each shell command, auto-approving low-risk ones and asking a person about the rest; unanswered prompts are refused after 300 seconds, and cron and unattended modes deny by default. Messaging is closed until you pair a user with an eight-character code that expires in an hour. The weak points are stated too: the local backend has no isolation, credential-path protection is described as defence in depth rather than a boundary, and hooks dropped into ~/.hermes/hooks load without a prompt.
Release cadence cuts both ways. At least 30 releases in five months bring fixes quickly, but configuration keys and commands move, so a gateway serving a team should be pinned and upgraded deliberately. Compared with Goose or OpenHands, Hermes is less a coding tool and more a resident assistant, and it is better at that job than either.
Where it earns its place
A solo founder running a two-product SaaS
Get a morning summary of signups, errors and support email in Telegram
Run Hermes on a small VPS with the Docker backend, add a cron job in plain language and pair your Telegram account. Software is free; cost is the VPS plus model tokens, or a $20/month Nous Portal Plus plan that bundles $22 of credits.
A platform engineer at a 30-person agency
Give the team a shared assistant in Slack that can run scripts without touching production hosts
Use the Daytona or Modal backend so commands run in a sandbox that sleeps when idle, keep approvals in smart mode and set an explicit user allowlist. The catch is upgrade discipline: pin a release and test each new one before rolling it out.
A researcher training tool-calling models
Generate agent trajectories for fine-tuning
Hermes includes batch trajectory generation and compression for training data, so the same harness used day to day produces datasets. Compute and model costs are yours; the MIT licence puts no limits on using the output.
Before you commit
- Setup
- One-line installer for Linux, macOS and WSL2 (curl | bash), a PowerShell installer for native Windows and an APT repository for Termux on Android; hermes setup runs a wizard and can import from OpenClaw, Claude Code, Codex or Cursor.
- Integrations
- Any OpenAI-compatible provider plus OpenRouter and Nous Portal; Telegram, Discord, Slack, WhatsApp, Signal and email; MCP servers; 40+ built-in tools; local, Docker, SSH, Modal, Daytona and Vercel Sandbox backends.
- Data and privacy
- Self-hosted: conversations, memory and skills live under ~/.hermes, and prompts go to whichever model provider you choose. Docker containers start with all capabilities dropped and no host environment variables; the security page does not mention telemetry. MIT licence.
- Cost at scale
- The agent is free; spend is model tokens and hosting. Nous Portal bundles models and hosted tools at $20, $100 or $200 a month with $22, $110 or $220 in credits, and Docker sandboxes default to 1 CPU, 5 GB of memory and 50 GB of disk each.
- Limits
- A very large open issue and pull request backlog; the local backend is not isolated; YOLO mode skips approvals except a hard blocklist; frequent releases change commands and config.
What's good
- One gateway for Telegram, Discord, Slack, WhatsApp, Signal and email
- Skills and memory stored as readable files
- Six terminal backends, including hibernating Modal and Daytona sandboxes
- Smart approval mode that fails closed after 300 seconds
What's not
- About 17,200 open issues and 34,000 open pull requests
- Local backend runs commands on the host with no isolation
- Hooks in ~/.hermes/hooks load without a prompt
- Fast release cadence means frequent config changes
Best for
Developers and small teams who want a self-hosted assistant that runs scheduled jobs and answers in their chat apps.
Hermes Agent alternatives
Hermes Agent: common questions
- Is Hermes Agent worth it?
- Hermes is the open-source agent to put on a $5 server if you want one that remembers you, with a security model you need to read before you turn on YOLO mode. We rate it 4.3 out of 5.
- How much does Hermes Agent cost?
- Free under the MIT licence; you pay for models and hosting. Optional Nous Portal: Free, Plus $20/month, Super $100/month, Ultra $200/month.
- What are the downsides of Hermes Agent?
- About 17,200 open issues and 34,000 open pull requests; local backend runs commands on the host with no isolation; hooks in ~/.hermes/hooks load without a prompt; fast release cadence means frequent config changes.
- Who is Hermes Agent best for?
- Developers and small teams who want a self-hosted assistant that runs scheduled jobs and answers in their chat apps.
- How much does Hermes Agent cost for a team or in production?
- The agent is free; spend is model tokens and hosting. Nous Portal bundles models and hosted tools at $20, $100 or $200 a month with $22, $110 or $220 in credits, and Docker sandboxes default to 1 CPU, 5 GB of memory and 50 GB of disk each.
- Is Hermes Agent safe for company data?
- Self-hosted: conversations, memory and skills live under ~/.hermes, and prompts go to whichever model provider you choose. Docker containers start with all capabilities dropped and no host environment variables; the security page does not mention telemetry. MIT licence.
- What are the best Hermes Agent alternatives?
- The closest alternatives we have reviewed are goose, OpenHands and claude-mem.
Answers checked Oct 26, 2026.
Make Hermes Agent? Show this rating on your site
