Hermes Agent
github.com/nousresearch/hermes-agent- Category
- Developer Tools
- Rank
- No. 24Tools index
- Listed in
- #14 Find agent skills
- Pricing
- Open Source
- Platform
- cli · web
- Type
- TOOL
- Builder
- @NousResearch
- GitHub
- 243.4k stars
- Latest release
- v2026.9.7
- Date
About
A self-improving AI agent with a built-in learning loop that creates skills from experience, improves them during use, and builds a deepening model of users across sessions. It runs anywhere from a $5 VPS to GPU clusters and works with any LLM provider.
What it does
Hermes Agent is a terminal-first assistant that can also receive conversations through messaging services. It runs tools, schedules unattended jobs, delegates parallel work, preserves conversation state, and routes different communities or threads into isolated profiles. Users select models and tool providers through configuration rather than changing code.
Why it's ranked here
The breadth is unusually practical: a capable terminal interface, remote messaging, scheduling, isolated profiles, resumable sessions, and several execution backends share one system. The repository also documents failure handling and operating tradeoffs with rare specificity. That scope brings substantial setup and dependency weight, but the operational design is more than a thin chat wrapper.
What's good
Conversation controls include history, multiline editing, interruption, retries, context compression, and streaming tool output. Messaging sessions survive restarts and distinguish hard resets from recoverable interruptions. Profile routing isolates memory, persona, tools, and sessions by community or thread. Scheduled work can deliver through connected platforms, while hosted scheduling supports sleeping between genuine job fires.
Tradeoffs
Hermes is a large system with Python, Node.js, messaging adapters, optional providers, platform-specific dependencies, and several deployment paths. Standard wheel and source-distribution builds are deliberately blocked outside Nix, so normal package installation expectations do not apply. Continuous micro-compaction is opt-in because it adds a model call, summarizes older assistant output sooner, delays turn completion, and invalidates prompt-cache prefixes.
How to use it well
Choose Hermes for a long-running personal or community agent that needs terminal access, messaging continuity, scheduled work, and isolated profiles. Start with the default batch compression and add optional providers only when needed. It is not a lightweight Python library for embedding into another application, and it does not remove the need to configure provider credentials or a portal subscription.
Technical notes+
The Python package surface is declared in pyproject.toml, requires Python 3.11 through 3.13, and separates provider, messaging, deployment, and development extras. setup.py blocks wheel and sdist creation outside a Nix build while preserving editable development installs. package.json defines a private npm workspace spanning desktop, TUI, web, and JavaScript tests, with Node.js 22.22.0 or newer. docs/profile-routing.md specifies weighted route selection and per-profile isolation. docs/micro-compaction.md describes rolling summary markers, database compaction, failure skipping, and prompt-cache costs. docs/chronos-managed-cron-contract.md documents authenticated one-shot scheduling, compare-and-set fire claims, and re-arming.
Observed
- License
- MIT
- Languages
- Python core with TypeScript and JavaScript UI workspaces
- Install surface
- Shell installer for Linux, macOS, WSL2, and Termux; PowerShell installer for native Windows; Docker and Nix distributions are also named
- Packaging
- Wheel and source-distribution builds are blocked outside Nix; editable development installs remain supported
- Interfaces
- Interactive CLI and TUI, messaging gateway, dashboard HTTP surface, and gateway API server
- Messaging platforms
- Telegram, Discord, Slack, WhatsApp, Signal, and email are listed as gateway entry points
- Execution backends
- Local, Docker, SSH, Singularity, Modal, Daytona, and Vercel Sandbox
- Runtime requirements
- Python 3.11 through 3.13; Node.js 22.22.0 or newer for npm workspaces
Read from README.md, setup.py, package.json, pyproject.toml, docs/streaming-tts.md, docs/profile-routing.md, docs/micro-compaction.md, docs/billing-lifecycle.md, docs/session-lifecycle.md, docs/relay-connector-contract.md, docs/rca-ssl-cacert-post-git-pull.md, docs/chronos-managed-cron-contract.md.
What it can do
Create new skills from user interactions and experiences
User conversations and task patterns → Executable skills and capabilities
Build and maintain persistent user models across sessions
User behavior, preferences, and interaction history → Personalized user profiles and context
Improve existing skills through continuous use and feedback
Performance data and user feedback on skill execution → Enhanced and optimized skills
Deploy and run on various infrastructure configurations
Infrastructure specifications (VPS, GPU clusters, etc.) → Running AI agent instance
Process messages across multiple messaging platforms
Messages from various platforms and channels → Platform-appropriate responses and actions
Schedule and automate tasks based on learned patterns
User scheduling preferences and task requirements → Automated task execution and reminders
Integrate with different LLM providers for text processing
Text prompts and LLM provider configurations → Generated text responses and completions
Intel on Hermes Agent
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.