Hermes Agent review, pricing and verdict

Choose Hermes Agent if you want an agent you host yourself that learns your recurring work and reaches you in Telegram, Slack or Discord. Skip it if you want a zero-setup tool or an editor copilot: you manage a model provider, updates and a fast release cadence.

Hermes Agent is an MIT-licensed, self-hosted AI agent from Nous Research that keeps persistent memory, writes its own reusable skills and answers through many messaging platforms. Its GitHub repository holds about 253,900 stars, and the October 2026 release added a desktop setup chat and one-click local models.

Maker: Nous Research · Autonomy: semi autonomous · Maturity: GA

Underlying models: Model-agnostic: Claude Fable 5.1, Claude Sonnet 5.5, GPT-6, Gemini 4 Argon, DeepSeek V4, Qwen3.8 (local), Any OpenAI-compatible endpoint

About Hermes Agent

Hermes Agent comes from Nous Research, the open-source AI lab that launched the agent on February 25, 2026. In October 2026 the company announced a $90 million Series B at a $1.5 billion valuation, led by Robot Ventures with participation from NVIDIA, Microsoft, Samsung, Union Square Ventures, Y Combinator and Menlo Ventures, according to its law firm's announcement. The money is earmarked for enterprise work, including a product called Hermes for Businesses.

The agent itself is MIT-licensed and installs with one terminal command or as a desktop app for macOS and Windows. Version 0.22.0, released October 13, 2026, added a guided setup chat that runs on a free trial model, hosted app connectors, and one-click local models through llama.cpp, so a newcomer can run it with no account and no API key. It works with Nous Portal, OpenRouter and any compatible endpoint, and the release notes list Claude Fable 5.1, Claude Sonnet 5.5 and Gemini 4 Argon as pickable models. Local runtimes such as Ollama, LM Studio and llama.cpp are supported, and the local catalog includes DeepSeek V4 Flash.

Day to day, Hermes keeps conversations, memories and skills between sessions, runs scheduled jobs, and talks to you through a gateway that its docs say covers more than twenty platforms, including Telegram, Discord, Slack, WhatsApp, Signal and Microsoft Teams. Terminal work can run on seven backends, among them Docker, SSH, Modal and Daytona. Its MCP support and a plugin catalog of 600 entries extend it further.

Hermes is not the right pick for everyone. If you mainly write code in an editor, Claude Code is built for that job and Hermes explicitly is not an IDE copilot. If you want another self-hosted agent to compare, OpenClaw and Eigent cover similar ground with different trade-offs. Hermes also moves fast: v0.22.0 made the Windows desktop app MSIX-only and defaults to pruning sessions inactive for 90 days, so self-hosters should read release notes before updating.

Pricing

The Hermes Agent software is free under the MIT license. You pay your model provider, or run local models for nothing. The optional Nous Portal subscription, read on the vendor page on 2026-10-15, has four plans: Free ($0, free models only, standard rate limits, $0 monthly credits), Plus ($20 a month with $22 of credits), Super ($100 a month with $110 of credits) and Ultra ($200 a month with $220 of credits). Paid plans carry a 10% bonus on credits, 200+ models and hosted tool usage, with rollover caps of $10, $50 and $100. Hermes Business uses one pooled team balance with per-member spend caps and publishes no price; Enterprise (on-premises or private cloud) is by contacting Nous. Compare other [self-hosted agents](/hub/agents/openclaw) before you commit.

Key Features

• Closed Learning Loop: Per the vendor docs, the agent curates its own memory, nudges itself to persist knowledge, writes skills from experience and improves them during use, so repeat work needs less instruction.
• Persistent Memory Across Sessions: Conversations, memories and skills are stored in your Hermes data directory and survive restarts; cross-session recall uses full-text search with LLM summarization.
• Many Messaging Platforms: One gateway connects Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Email, Microsoft Teams, Google Chat and more, per the docs (more than twenty in total), with scheduled jobs delivered to any of them.
• Model-Agnostic Runtime: Works with Nous Portal (hundreds of models), OpenRouter, OpenAI or any compatible endpoint; the latest release added Claude 5.5, GPT-6 and Gemini 4 Argon to the picker.
• One-Click Local Models: Since the October 2026 release Hermes installs and maintains llama.cpp for your hardware (CPU, CUDA, Metal, Vulkan or ROCm) and recommends the best catalog model that fits your GPU and RAM, with no account or key.
• Seven Terminal Backends: Run commands locally or inside Docker, SSH, Daytona, Singularity, Modal or Vercel Sandbox, with serverless backends that hibernate when idle.
• MCP, Connectors and a 600-Plugin Catalog: Connect MCP servers and hosted apps such as Linear and Notion, and install from a catalog of 600 reviewed plugins pinned to specific commits.

Strengths

• MIT license and a very large open-source community on GitHub; the star and fork counts sit in the benchmarks table below.
• Runs fully on your own hardware with local models and no account, which suits privacy-conscious users and offline setups.
• Wide reach from one agent: a multi-platform messaging gateway, a desktop app, a CLI and scheduled jobs share the same memory.
• Optional Nous Portal pricing ladder with a credit bonus on paid plans, instead of a forced subscription.

Weaknesses

• Releases are fast and sometimes breaking: v0.22.0 made the Windows desktop app MSIX-only (Windows 11 22H2 or newer), made the macOS build Apple Silicon only by default and moved managed installs to Python 3.14.
• Sessions inactive for 90 days are pruned automatically by default in the latest release, so users who want a permanent history must switch that setting off.
• It is not an IDE copilot: the docs say so directly, and there is no inline editor integration.
• You still supply and pay for a model provider; the free Nous Portal plan covers free models only with $0 monthly credits.
• Hermes Business and Enterprise carry no public price, so teams cannot budget without talking to Nous.

Visit Hermes Agent Official Site