Hermes got popular fast. A small, auditable, open-source agent runtime that runs in your terminal and talks to any LLM via the OpenAI-compatible API. If you've read a few "local agents in 2026" threads, you've seen it mentioned. If you're evaluating a workspace-grade answer to the same problem, you've probably seen Froots too.
This is the straight comparison. We ship Froots, so there's a bias you should be aware of — we still think the recommendation below is honest.
TL;DR
| Pick Hermes | Pick Froots | |
|---|---|---|
| Shape | CLI daemon + SDK | Desktop app + CLI |
| Best for | Scripting, CI, headless servers, devs | Daily knowledge work, notes, teams |
| Notes editor | No | Yes (block editor, graph, search) |
| Built-in agents | Define your own | Four, plus write-your-own |
| License | Apache-2.0 | BSL (app) / AGPL-3.0 (sync server) |
| Price | Free | Free Seedling; $14.99/mo Pro; $29/seat Teams |
If your default move is vim agent.yml you'll love Hermes. If your default move is ⌘N new note you'll love Froots.
Where they're the same
Both runtimes hit the same marks, so you can skip these when comparing:
- Local-first. Your data does not leave your machine unless you configure it to.
- Bring your own model. Anthropic, OpenAI, Google, local (Ollama / LM Studio / llama.cpp / mlx), any OpenAI-compatible endpoint.
- MCP (Model Context Protocol) native. You can plug in MCP servers for tool use.
- Tool sandboxing with per-tool policies. Deny, allow, prompt — all three supported in both.
- Plain-file outputs. Nothing is stored in a proprietary binary format.
If a thread tells you one of these is a Hermes-exclusive or Froots-exclusive, it's outdated.
Where they diverge
Hermes is a runtime; Froots is a workspace
Hermes is one binary (hermes). You give it a config, it runs agents. You wire it into your editor, your terminal, your CI — wherever you want agents.
Froots is a full Tauri desktop app that contains an agent runtime. You can't "wire Froots into VS Code" the way you can Hermes, because Froots wants to be the surface, not a backend.
This is the biggest single difference. Pick based on whether you want plumbing or a place.
Hermes has a smaller surface
Hermes does one thing: run agents. No editor, no calendar view, no inbox, no graph, no widgets. This is a feature, not a limitation — Hermes is easier to audit, port, and integrate. The entire source fits in a weekend's reading.
Froots is ~300K lines when you include the frontend, the Rust backend, and the sync server. It does more because it's doing more. The tradeoff is that it's harder to audit line-by-line, easier to use on day one.
The agent philosophy
- Hermes agents are defined in YAML: name, system prompt, tools, model, temperature. You write many small ones and compose them.
- Froots agents are defined in markdown (frontmatter + prose). Four come built-in (Lime = Researcher, Yuzu = Scribe, Clem = Maker, Cherry = Gardener). You write more by dropping markdown files into
Agents/.
Hermes's philosophy: agents are code, keep them terse. Froots's philosophy: agents are characters, give them voice. Neither is wrong.
Routines
Hermes doesn't have "routines" as a first-class concept — you compose agents and schedule them via cron or systemd timers. Fine for sysadmins, a wall if you're not one.
Froots ships a sentence-to-routine compiler: highlight "every weekday at 8am, pull my top three Linear issues and summarize them," press ⌘⇧R, and you get a runnable graph with a cron trigger, a Linear call, an agent step, and an output. Non-coders build real automations this way.
Inbox, widgets, mobile
Hermes: none of these. Not part of the scope.
Froots: all three. Unified inbox spans iMessage, Gmail, Slack, Telegram, Signal, WhatsApp, X, Discord. Native OS widgets on macOS / Windows / iOS. iOS and Android apps (coming Q3 2026).
Deployment story
Hermes is great at headless. hermes serve --config prod.yml and you have a long-running agent that you can hit over HTTP or a Unix socket. Drop it in Docker. Run it on a $5 VPS.
Froots is a desktop app first; there's a CLI (froots command) and a self-hosted sync server, but it's not built for "put it behind nginx." If you want agents on servers, run Hermes. If you want agents on your laptop, run Froots.
When to pick each
Pick Hermes if:
- You work in a terminal more than a GUI.
- You want the smallest, most auditable runtime possible.
- You're deploying agents to servers, not to humans.
- Your compliance team needs Apache-2.0 specifically.
- You already have a note-taking setup you love and don't want to move.
Pick Froots if:
- You want notes + agents + automations in one window.
- You want agents with persistent memory across sessions.
- You want a sentence-to-routine compiler.
- You want to bring non-technical teammates along.
- You want mobile, widgets, and a real inbox.
The short version
Hermes is the best CLI-native agent runtime we've tried. We've used it in production; we recommend it if its shape is your shape.
Froots is the best workspace-native agent surface we know how to build. It's more app, less runtime — and if you're the kind of person who lives in a notes app, it's the one you should pick.
Try Froots free at froots.ai/download.