Hermes Agent alternative · Compared honestly · Public Alpha

A Hermes Agent alternative whose memory works in your browser and your editor.

Start where you left off.

Choose Hermes Agent if you want a free, MIT-licensed agent with no account that writes its own skills, runs on a $5 VPS or serverless backends, and answers on 29 listed messaging platforms. Choose Vodou, the Hermes Agent alternative on this page, if the memory needs to be where you work: Vodou types it into ChatGPT, Claude and 20 other AI sites in Chrome, loads it into Claude Code and Cursor through hooks, and shares it with editors through an MCP host. Vodou needs a free account, costs $10 a month on your own model keys after a 60-day trial, and its engine is proprietary. Vodou imports Hermes memory files and Hermes-format skills.

Download Vodou See the comparison

Imports Hermes memory and skills22 AI sites in ChromeClaude Code · Cursor hooks60-day free trial

TL;DR

Vodou and Hermes Agent both keep an agent's memory under your control; Hermes Agent is free, MIT-licensed, needs no account and had 246,000+ GitHub stars on Sep 17, 2026, while Vodou costs $10 a month on your own keys after a 60-day trial and requires a free account.
Hermes Agent's built-in memory is two capped files — MEMORY.md at 2,200 characters and USER.md at 1,375 — plus FTS5 search over past sessions and 8 optional provider plugins; Vodou's memory is a local SQLite store searched with full-text, vector and reranked retrieval, where every fact keeps its source.
Vodou types its memory into 22 AI sites in Chrome and loads it into Claude Code and Cursor through hooks; Hermes Agent reaches Claude Code, Cursor and Codex as an MCP server that lends them its messaging tools.
Vodou imports Hermes memory with vodou-core mem import hermes ~/.hermes/memories and converts a Hermes SKILL.md with vodou-core skill import.

Definition

What is Hermes Agent?

Hermes Agent is an open-source, self-improving AI agent from Nous Research, released under the MIT license. It creates skills from experience and refines them in use, keeps a curated memory and a searchable session history, and runs anywhere from a laptop to a $5 VPS or a serverless backend, reachable from a terminal, Hermes Desktop or 29 listed messaging platforms.

Hermes Agent is not the Hermes language models; it is the agent runtime Nous Research builds around any model — Nous Portal's 300+ models, OpenRouter, OpenAI or your own endpoint. It installs with one command on Linux, macOS, WSL2 and Termux or natively on Windows, had 246,000+ GitHub stars on September 17, 2026, and shipped v2026.9.14 on September 14. It also runs scheduled jobs, a kanban board of workers and parallel subagents.

Hermes keeps memory in ~/.hermes/memories/ as MEMORY.md, the agent's notes, and USER.md, your profile, loaded into every session as a snapshot; when a write would exceed the limit, Hermes returns an error instead of silently dropping entries. Conversations live in a SQLite database with FTS5 search, and eight provider plugins, including Honcho, Mem0 and Supermemory, add more. Vodou imports the two Markdown files directly.

The problem

Why do people look for a Hermes Agent alternative?

The alternative guides and Reddit threads give three reasons: running Hermes means operating a server-style, terminal-first runtime; the self-improvement loop can rewrite skills you edited and grade its own work too kindly; and its built-in memory is deliberately small. Kilo's analysis of about 1,300 Reddit comments also points to Hermes Agent's shorter track record than OpenClaw's. Vodou answers the memory reason; any agent you run yourself, Vodou included, still has to be run.

A runtime to operate

Hermes is built to run on a server and be driven from a terminal or chat app; one alternatives guide puts it as “running it means owning it.” Hermes Desktop for macOS and Windows narrows that gap. Vodou also runs on your own machine, so it changes where the upkeep happens, not whether there is any.

Self-improvement you can't fully steer

Hermes writes and patches its own skills after hard tasks. Reddit users praise the learning loop, but some report it rewriting skills they had edited by hand, or reporting success on a task that failed. Vodou skill steps are fixed in the SKILL.md until someone edits it.

Memory that is small by design

MEMORY.md holds about 800 tokens and USER.md about 500, so Hermes curates hard and leans on session search or a plugin for the rest. That keeps prompts short; it also means detail has to be searched for rather than remembered.

Capabilities

How is Vodou different from Hermes Agent?

Vodou is built around a memory your other tools share. Hermes Agent centres on one agent that learns — its own skills, a compact memory and session search inside Hermes; Vodou centres on a local memory that the Vodou Bridge extension, Claude Code and Cursor hooks, an MCP host and your messaging channels all read and write. Memory is stored and searched on your computer; with a hosted model, what you send goes to that provider.

Your memory inside ChatGPT and Claude

Press Ctrl+B on any of 22 AI sites in Chrome and the Vodou Bridge extension types the relevant memory into the chat box, visible and editable before you send; inject is manual by default, so nothing is sent on your behalf. Chats on those sites flow back into the same memory.

Claude Code and Cursor start with context

Claude Code gets memory from four Vodou hooks — at session start, on every prompt, and saved back at session end. Cursor loads memory at session start, and every prompt refreshes a context file of matching memories that the Cursor agent reads.

One MCP host with a vault per editor

Vodou's MCP host on 127.0.0.1:8787 serves Claude Desktop, Cursor, VS Code, Windsurf and Zed, each with its own token, memory vault and rate limit. A 30-day audit log shows what each client read.

Skills with fixed steps and stopping points

Vodou skills are SKILL.md files with an AGENT_ACTIONS block that the workflow driver runs in order, pausing where the skill asks for your choice. A skill does the same thing every run until someone edits it, and you can add as many as you need.

Memory with sources, corrections and undo

Vodou stores memory in a local SQLite database searched with full-text, vector and reranked retrieval. Every fact keeps where it came from; correct one once, pin it, hide what came from one website, or roll back an import.

Local or hosted models, your keys

Run Ollama, LM Studio or the bundled llama.cpp and nothing you say leaves the machine, or bring keys for any model provider and switch per conversation. With a hosted model, what you send — including memory extraction — goes to that provider; Vodou never meters those keys.

Comparison

Vodou vs Hermes Agent: how do they compare?

Hermes Agent wins on price, license, account-free setup, community size, messaging platform count, self-written skills and where it can run — a VPS, Docker, SSH or serverless backends. OpenClaw, the agent Hermes is most compared with, shares most of those strengths. Vodou's column is about reach: the same local memory in 22 browser AI sites, Claude Code, Cursor and an MCP host with a vault per client, skills that run in a fixed order, and an importer for Hermes memory. Vodou is a public alpha.

→Scroll the table sideways to see 2 columnsall 3 columns

CAPABILITY	VODOU public alpha	HERMES AGENT	OPENCLAW the agent Hermes is most compared with
What it is	local memory layer + agent harness	self-improving agent (Nous Research)	self-hosted personal assistant (OpenClaw Foundation)
Price	$10/mo on your own keys after a 60-day free trial	free; optional Nous Portal subscription for models and hosted tools	free — “no paid tier, hosted service, or token”
Account required	yes — a free Vodou account licenses the engine	no (Nous Portal sign-in only if you use it)	no
License	open-core: gateway, skills, extension, MCP servers Apache-2.0; engine proprietary	MIT	MIT
GitHub stars (Sep 17, 2026)	public alpha — early community	246,000+	389,000+
Self-improving skills	no — skill steps are fixed in SKILL.md and run in order; imports Hermes-format skills	yes — creates skills after hard tasks and improves them in use; agentskills.io compatible	no — ClawHub community registry + plugins
Memory store	Markdown logs + SQLite: full-text, on-device vectors, reranker; provenance; correct, pin, undo	MEMORY.md (2,200 chars) + USER.md (1,375 chars); FTS5 session search; 8 memory provider plugins	MEMORY.md, USER.md, daily notes; SQLite keyword + vector hybrid; OpenAI embeddings by default
Memory in browser AI chats	22 sites via the Vodou Bridge Chrome extension	no chat-site extension	no chat-site extension
Memory in Claude Code / Cursor	hooks: Claude Code (4 hooks); Cursor (session start + per-prompt context file)	no hooks; hermes mcp serve lends Claude Code, Cursor and Codex its messaging tools	no hooks; runs Claude and Codex as harness plugins instead
MCP	client for any server; host on 127.0.0.1:8787 with per-client token, vault, rate limit, audit	client (stdio, HTTP, OAuth); hermes mcp serve exposes 10 messaging tools	client (openclaw mcp add); openclaw mcp serve shares channel conversations
Messaging channels	Telegram, Slack, Discord, WhatsApp, Signal, iMessage, Teams, Google Chat, voice, web — plus any channel package you install	29 listed, incl. Telegram, Discord, Slack, WhatsApp, Signal, Email, SMS, Teams, Matrix, iMessage via BlueBubbles	32 listed, incl. WhatsApp, Telegram, Slack, Discord, iMessage, Signal, Teams, Matrix, LINE, WeChat
Apps	web console and CLI; no phone apps — reach it from your phone through a channel	Hermes Desktop (macOS, Windows), terminal UI, CLI	native apps for macOS, iOS, Android, Windows, Linux + Control UI
Models	any provider on your keys; Ollama, LM Studio, bundled llama.cpp	Nous Portal (300+ models), OpenRouter, OpenAI, your own endpoint	hosted and local providers; Claude and Codex as harness plugins
Scheduled and background work	60-second scheduler, automations, kanban board (worker processes opt-in)	cron with delivery to any platform, kanban board, subagents	cron jobs, background tasks, sub-agents
Where tools run	your computer; fail-closed permission engine and approval gates; no container backends	7 terminal backends: local, Docker, SSH, Singularity, Modal, Daytona, Vercel Sandbox	the host by default; Docker sandbox you can turn on
Where it runs	your computer; binds to 127.0.0.1 only	a laptop, “a $5 VPS, a GPU cluster, or serverless”	a laptop, or one Gateway shared by a team
Moving in	imports Hermes and OpenClaw memory files; adapts Hermes SKILL.md files	hermes claw migrate: OpenClaw settings, memories, skills, API keys	imports Codex, Claude Code and Hermes memory
Research tooling	—	batch trajectory generation and compression for training models	—
OS	macOS, Windows, Linux	Linux, macOS, WSL2, native Windows, Termux	macOS, Linux, Windows; iOS and Android apps

Sources checked (2026-09-17); competitor columns are as of that date.

Hermes Agent — github.com/NousResearch/hermes-agent (README, MIT, stars, release v2026.9.14) · hermes-agent.nousresearch.com · messaging · memory · profiles · MCP · security · kanban · ACP
OpenClaw — github.com/openclaw/openclaw (README, MIT LICENSE, stars) · channels · memory · MCP · sandboxing · cron jobs
Why people switch — Kilo, 1,300 Reddit comments analyzed · Respan, Hermes Agent alternatives

Vodou column checked against the product on 2026-09-17.

Fair to both

When should you choose Hermes Agent instead of Vodou?

Choose Hermes Agent when you want an agent that improves itself, lives on a server you can reach from anywhere, and costs nothing but model tokens. Vodou is the better fit only when the memory must be usable in your browser AI chats, Claude Code or Cursor, and you accept a free account, a monthly price after the trial and a proprietary engine.

You want free and MIT-licensed with no account — Nous Portal is optional.
You want the agent to write and refine its own skills after hard tasks.
You want it always on: on a $5 VPS, in Docker, over SSH or on Modal or Daytona serverless backends. Vodou runs on your own computer and binds to 127.0.0.1.
You need a platform Vodou has no first-party adapter for, such as Matrix, Email, SMS, DingTalk, Feishu or WeCom.
You train models: Hermes generates and compresses tool-calling trajectories for research.

How it works

How do you move from Hermes Agent to Vodou?

Vodou reads Hermes memory files directly: install Vodou, preview with vodou-core mem import hermes ~/.hermes/memories --dry-run, then run it without the flag. MEMORY.md and USER.md are copied into Vodou's local memory and indexed, and vodou-core skill import converts a Hermes SKILL.md; session history, cron jobs, channel settings and API keys are not moved.

STEP 1
Install Vodou and sign in.
Install Vodou on macOS, Windows or Linux and sign in with a free Vodou account; the 60-day trial needs no credit card. Point Vodou at Ollama or add a key for your model provider. Hermes Agent can keep running while you compare the two.
STEP 2
Preview the memory import.
Run vodou-core mem import hermes ~/.hermes/memories --dry-run and Vodou lists the files it would write. Pass the folder explicitly: current Hermes keeps MEMORY.md and USER.md in memories/, and a named profile keeps them in ~/.hermes/profiles/<name>/memories.
STEP 3
Import memory, then skills.
Run the command again without --dry-run. MEMORY.md and USER.md land as MEMORY-imported.md and USER-imported.md, never overwriting Vodou's own, with HTML comments and invisible characters stripped and injection-looking lines flagged for review. Then run vodou-core skill import on a Hermes skill folder to convert its SKILL.md into a Vodou skill.
Deep dive: AI memory · Skills
STEP 4
Reconnect and verify.
Connect Telegram, Discord, Slack or another channel in Vodou's Messaging settings and search the imported memory in the Vodou Console. To start again, vodou-core mem import-undo <job-id> removes every Hermes and OpenClaw import.

FAQ

Frequently asked questions about Hermes Agent alternatives

Is Hermes Agent free?

Yes. Hermes Agent is free and open source under the MIT license; you pay your model provider, or optionally subscribe to Nous Portal for 300+ models and hosted tools under one plan. Vodou is not free forever: it is free for 60 days with no credit card, then $10 a month on your own model keys, and it requires a free Vodou account.

Is Hermes Agent open source?

Yes. Hermes Agent's code is on GitHub at NousResearch/hermes-agent under the MIT license. Vodou is open-core: the gateway, skills, Vodou Bridge extension and MCP servers are Apache-2.0 at github.com/VodouAI/OS, while the Rust engine that does routing and memory is proprietary and licensed to you through a free account.

Is Hermes Agent safe?

Hermes ships real guards: a default smart approval mode in which an auxiliary model auto-approves low-risk commands, denies dangerous ones and asks you about the rest; a hard blocklist; pairing codes for new chat users; and hardened Docker containers. It still runs commands on real machines, so review its settings. Vodou takes a different line — loopback-only, a fail-closed permission engine and approval gates — and needs the same care.

Hermes Agent vs OpenClaw: which is better?

Neither wins outright. Hermes Agent is the better fit for an agent that learns — it writes its own skills and runs on a VPS, Docker, SSH or serverless backends — and it imports an OpenClaw setup with hermes claw migrate. OpenClaw is the better fit for a control plane: 32 listed channels, phone apps and team deployments. Vodou imports memory from both and adds that memory to your browser, Claude Code and Cursor.

Is Hermes Agent good for coding?

It can code, and it runs inside VS Code, Zed and JetBrains editors through the Agent Client Protocol, but Hermes is a general agent first; Claude Code, Codex and OpenCode are built for coding. Vodou does not replace any of them: its hooks give Claude Code and Cursor a searched local memory, so your coding agent starts with the context your other tools already learned.

Is Hermes Agent local?

It can be. Hermes Agent installs on your laptop and works with a local model endpoint, but it is designed to live anywhere — a $5 VPS, Docker, SSH, or serverless backends such as Modal and Daytona. Vodou runs only on your own computer, listens only on 127.0.0.1, and can use Ollama, LM Studio or its bundled llama.cpp, so nothing you say leaves the machine.

Is there a Hermes Agent alternative for Windows?

Several, though Hermes Agent itself now runs natively on Windows through a PowerShell installer or Hermes Desktop for Windows 10/11, and OpenClaw has a Windows installer too. Vodou installs natively on Windows 10/11 x64 as well as macOS 13+ and Linux, and brings its memory into Chrome, Claude Code and Cursor on the same machine.

Is Hermes Agent a harness?

Yes. Hermes Agent is an agent harness: it wraps a model in a loop with tools, memory, skills, scheduling and approvals. Vodou is a harness too, with a different centre — skills whose steps run in a fixed order, an MCP host other clients attach to, Claude Code and Cursor hooks, and one local memory they all share. Vodou's agent harness page at vodou.ai/agent-harness compares the two in more depth.