🩺 Vitals
- 🟢 Last active: 2026-08-05
- 📦 Latest release: v2026.8.3 (2026-08-03)
- 🐞 Open issues: 28033
- 🌟 Stars: 225.7k
What do these metrics mean?
- Last active: when code was last pushed, as of our last check. The dot is green when that was recent, grey otherwise. A long gap can mean a tool is finished and stable, not only unmaintained.
- Latest release: the most recent tagged, packaged version the maintainers published. Not every healthy project tags releases.
- Open issues: unresolved reports and requests. A high number is normal for a popular project and is not a warning on its own.
- Stars: how many people bookmarked the project on its forge. A rough popularity signal, not a measure of quality.
🏗️ Profile
- Official: hermes-agent.nousresearch.com
- Source: github.com/NousResearch/hermes-agent
- License: MIT
- Deployment: Docker | Native Desktop | Local Terminal
- Data Model: SQLite / Local State
- Jurisdiction: USA 🇺🇸 (Nous Research)
- Compliance (SaaS): N/A (Undisclosed)
- Compliance (Self-Hosted): Self-Hosted (User Managed)
- Complexity: Medium (3/5) - One-click desktop install for individuals; Docker plus LLM API keys and platform tokens for server deployment
- Maintenance: Low (2/5) - 187k+ stars, near-daily releases, active contributor base
- Enterprise Ready: Low (2/5) - New web admin dashboard, but still no SSO, audit logs, or commercial SLA
1. The Executive Summary
What is it? Hermes Agent is an autonomous, self-improving AI agent built by Nous Research. Its defining capability is a closed learning loop: the agent creates its own skills, stores them in persistent memory, and improves over time without retraining the underlying model. It deploys natively across 23+ messaging platforms (Slack, Discord, Telegram, WhatsApp, Signal, Microsoft Teams, Email, and more) and supports parallel subagent delegation, scheduled automations, and a curated MCP (Model Context Protocol) tool catalog. Tasks execute across five backends (local, Docker, SSH, Singularity, and Modal) with container hardening and namespace isolation. As of June 2026 the original terminal/CLI client is joined by a native desktop app for macOS, Windows, and Linux, adding streaming tool output, a file browser, voice I/O, and a web admin dashboard. All state is stored locally; no data leaves the operator's infrastructure unless an external LLM API is called.
The Strategic Verdict:
- 🔴 For Regulated Environments Requiring Audit Trails: Caution. No built-in RBAC, audit logging, or SSO. Autonomous skill creation means the agent's behaviour evolves over time. Compliance teams cannot certify a static behaviour profile. Requires significant internal governance engineering before production deployment.
- 🟢 For Innovation Teams Building Internal AI Assistants: Strong Buy. A single deployment covers every messaging channel the team already uses, and the new desktop app lowers the adoption barrier for non-technical staff while the terminal/CLI client remains for operators and headless servers. The self-improving memory loop eliminates repetitive prompt engineering: the agent learns organisational context once and retains it across sessions.
2. The "Hidden" Costs (TCO Analysis)
| Cost Component | Copilot Studio (SaaS) | Hermes Agent (Self-Hosted) |
|---|---|---|
| Licence Fee | Per-agent/per-message pricing | $0 (MIT) |
| LLM Inference | Included (Azure OpenAI) | BYOK, or optional Nous Portal managed credits |
| Data Sovereignty | Microsoft Cloud | 100% local (your infrastructure) |
| Channel Coverage | Microsoft Teams-centric | 7+ platforms (Slack, Discord, Telegram, etc.) |
| Behavioural Lock-in | Copilot ecosystem | Model-agnostic (any LLM provider) |
3. The "Day 2" Reality Check
🚀 Deployment & Operations
- Installation: Three entry points serve different operators. Individuals can use the one-click native desktop installer (macOS, Windows, Linux) with in-app self-update. Server and headless deployments continue to use Docker Compose or direct Python execution via the terminal/CLI client. All paths require LLM API keys (supports OpenAI, Anthropic, local models) and platform-specific bot tokens for each messaging channel. A single instance can serve multiple channels simultaneously, and the desktop app can connect to a remote Hermes gateway over OAuth.
- Scalability: Designed for team-scale deployments. The agent's persistent memory and skill store grow over time. Scaling to organisation-wide deployment requires isolating agent instances per team or domain to prevent context contamination across business units.
🛡️ Security & Governance (Risk Assessment)
- Jurisdiction & CLOUD Act (USA 🇺🇸): Nous Research is a VC-backed US corporation: standard CLOUD Act exposure applies to any managed offering. However, Hermes Agent is entirely self-hosted with no telemetry or phone-home behaviour. The geopolitical risk materialises only when the operator routes inference outward, through US-based LLM APIs (OpenAI, Anthropic) or the new Nous Portal managed-credit service. Pairing Hermes Agent with a locally-hosted model and avoiding the Portal eliminates US jurisdiction exposure entirely.
- The Compliance Shift: No formal compliance certifications exist: no SOC 2, no ISO 27001. The new web admin dashboard improves operational visibility and a Promptware defence layer guards against prompt-injection attacks, but neither substitutes for audit logging or access controls. The deeper governance challenge is structural: autonomous skill creation means the agent's capabilities evolve through use, so compliance teams must treat Hermes Agent deployments as living systems requiring periodic behaviour audits, not static software installations. API key security, network egress controls, and LLM provider selection remain entirely the operator's responsibility.
- License Risk (MIT: No Traps): MIT licence imposes no copyleft requirements, no network use clauses, and no commercial restrictions. Enterprises can embed, modify, and redistribute Hermes Agent within proprietary systems without disclosure obligations. The new desktop app and admin dashboard ship under the same MIT licence. There is no open-core split. The optional Nous Portal sells managed inference credits, not gated features: persistent memory, MCP catalog, subagent delegation, and every messaging channel remain fully available in the self-hosted open source release.
4. Market Landscape
🏢 Proprietary Incumbents
- Microsoft Copilot Studio: The enterprise AI agent builder, tightly integrated with Teams, Power Platform, and Azure. Organisations migrate to Hermes Agent to escape per-message pricing, Microsoft ecosystem lock-in, and the inability to deploy agents across non-Microsoft channels.
- OpenAI Assistants API: The developer-first approach to building stateful AI agents with tool use and memory. All conversation data resides in OpenAI's cloud; Hermes Agent provides equivalent autonomous capabilities with full data sovereignty and model-agnostic inference.
🤝 Open Source Ecosystem
- Agent Zero: An autonomous code-execution and terminal agent focused on research and DevOps workflows. Where Agent Zero excels at sandboxed system-level automation, Hermes Agent specialises in persistent, cross-platform conversational deployment with self-improving memory.
- OpenClaw: An autonomous AI agent with deep OS access and zero telemetry. Both share the autonomous agent paradigm, but OpenClaw focuses on local system automation while Hermes Agent prioritises multi-platform messaging deployment and cumulative learning across sessions.