Hermes Agent from Nous Research has become 2026's fastest-growing self-improving AI agent. From 0 to 177,000 GitHub stars in just four months. That's more than the entire Khoj project pulled in over three years of development.
Hermes hit a nerve. Persistent memory, self-improving skills, and over 30 LLM providers in a single local Python agent. But what does the growth actually look like? Who's building with it? Which channels get used the most? Which models?
In this article, you'll find the latest numbers, data, and facts on Hermes Agent. All figures are current as of June 2026 and come from official sources (GitHub repository, Nous Research blog, hermesatlas.com) or verified third-party sources.
- Hermes Agent has around 177,000 GitHub stars in June 2026, growing by roughly 35,000 stars per month since March.
- Around 1,400 contributors, 28,600 forks, and more than 80 community projects in the Hermes ecosystem (per hermesatlas.com).
- Over 40 LLM providers and 9 channels supported, from CLI to Telegram to Microsoft Teams. Ollama unlocks local models with zero API costs.
1. What Is Hermes Agent?
Hermes Agent is Nous Research's open-source AI agent for self-improving workflows. You install it locally (with Python 3.11+), connect an LLM provider of your choice, and talk to it either through the command line or via one of nine multi-channel adapters (Telegram, Discord, Slack, WhatsApp, Signal, Email, Microsoft Teams, Home Assistant).
Three things set Hermes apart from other agents: persistent memory across sessions, self-improving skills that learn from experience, and the Tool Gateway with four built-in tools (web search, image generation, text-to-speech, and Cloud Browser).
2. GitHub Stars and Growth
Hermes Agent has been publicly available since February 2026. Here's how the growth has played out:
The growth curve is striking. From 57,200 stars on April 11 to 177,000 stars on June 2. That's roughly 120,000 stars in seven weeks. For context: OpenClaw currently grows by about 3,000 stars per week, and Hermes blew past that at 9,500 stars per week during its peak.
3. Star Growth vs. Competing Adaptive Agents
Hermes isn't the only player in the adaptive agent space. The chart below compares Hermes with the four most important direct competitors by GitHub stars (in thousands, as of June 2026):
gradually.aiHermes Agent's dominance is clear-cut. 177,000 stars is more than all four direct competitors combined (around 136,600 stars). That makes Hermes the market leader in the adaptive agents category.
For context, Open Interpreter from September 2023 is the oldest project on this list, and Khoj has been around since 2022. Hermes hit in four months what took Khoj three years.
4. Release Timeline and Major Versions
Hermes Agent has been developed at a blistering pace since February 2026. Here are the most important milestones:
Date | Version | Milestone | Details |
|---|---|---|---|
| Feb. 2026 | v0.1 | Initial Release | Erster öffentlicher Release von Hermes Agent durch Nous Research |
| März 2026 | v0.7 | Multi-Channel Gateway | Telegram, Discord, Slack und WhatsApp Channels integriert |
| April 2026 | v0.10 | Persistentes Memory | Erste stabile Version der selbstverbessernden Skills mit sessionübergreifendem Memory |
| 16. Mai 2026 | v0.14 | Stable Memory Backend | 808 Commits, 633 PRs. Stabilisierung des Memory-Backends. |
| 28. Mai 2026 | v0.15 | Velocity Release | Skill Bundles und Kanban-Board für Multi-Agent-Workflows. session_search 4.500x schneller. |
The release cadence is intense. The "Velocity Release" v0.15 on May 28, 2026, alone shipped 1,302 commits and 747 pull requests since v0.14. Highlights were Skill Bundles (one slash command loads an entire workflow) and Kanban as a real multi-agent platform with orchestrator auto-decomposition.
5. Supported LLM Providers
Hermes Agent supports over 30 LLM providers. Here are the twelve most important ones:
Provider | Category | Model Examples | Notes |
|---|---|---|---|
| Anthropic | Frontier | Claude Opus 4.7, Sonnet 4.6, Haiku 4.5 | OAuth für Pro/Max-Abos, API-Key direkt möglich |
| OpenAI | Frontier | GPT-5.5, GPT-5.4, GPT-5-Codex, GPT-5 nano | API-Key oder Env-Var, Pricing pro Token |
| Google Gemini | Frontier | Gemini 3.1 Pro, Gemini 3 Flash | API-Key, OAuth über Google Cloud |
| OpenRouter | Aggregator | 200+ Modelle proxied | Ein API-Key für alle Provider |
| Nous Portal | Aggregator | Nous-eigene Modelle plus 200+ via OpenRouter | Standard-Provider nach hermes setup --portal |
| Ollama | Lokal | Llama 3.3, Qwen 2.5 Coder, DeepSeek R1 | Komplett lokal, GPU-Offloading automatisch |
| DeepSeek | Frontier | DeepSeek V3, R1 | Günstige Cloud-API |
| MiniMax | Frontier | MiniMax-M1, M2 | Asiatisches Frontier-Modell |
| xAI | Frontier | Grok 4, Grok 4 Code | OAuth möglich |
| AWS Bedrock | Enterprise | Claude, Llama, Mistral | Über AWS-Credentials |
| Azure OpenAI | Enterprise | GPT-Modelle über Azure-Endpunkte | Azure Foundry Integration |
| Kimi (Moonshot AI) | Frontier | Kimi K2 | Asiatisches Frontier-Modell |
The range is impressive. From a free local setup with Ollama all the way to enterprise solutions via AWS Bedrock and Azure OpenAI. If you want an overview of current open-source LLMs, you'll find the most important options there.
5.1. Tool Gateway
Through the Nous Portal, Hermes ships four built-in tools that activate with a single OAuth login:
- Web Search: Structured search results available directly in the agent
- Image Generation: AI image generation via Nous Portal
- Text-to-Speech: Synthetic voices for voice channels
- Cloud Browser: Headless browser for web automation
6. Channel Integrations
Hermes Agent is multi-channel-first. Through the gateway setup (hermes gateway setup), you can connect the following platforms:
Channel | Type | Setup Effort | Cost |
|---|---|---|---|
| CLI | Terminal | Trivial | kostenlos |
| Telegram | Messaging | Niedrig (2 Min, BotFather) | kostenlos |
| Discord | Messaging | Niedrig (Discord-Bot-Token) | kostenlos |
| Slack | Messaging | Mittel (Slack-App) | kostenlos (Slack Free) / kostenpflichtig (Pro+) |
| Messaging | Hoch (Twilio Setup) | ab 0,005 USD pro Nachricht + Meta-Templates | |
| Signal | Messaging | Niedrig | kostenlos |
| Klassisch | Mittel (SMTP/IMAP) | eigener Mailserver oder SMTP-Anbieter | |
| Microsoft Teams | Enterprise | Hoch (Azure AD) | kostenpflichtig (Teams-Abo) |
| Home Assistant | Smart Home | Mittel | kostenlos |
CLI, Telegram, Discord, and Signal are the easiest entry points. WhatsApp needs Twilio, which makes it noticeably more involved and pricier. Microsoft Teams requires an Azure AD setup and is primarily aimed at enterprise environments.
7. Community and Ecosystem
Hermes has built an unusually active community in its first four months. The key numbers as of June 2026:
- 177,000+ GitHub stars (growth: ~35,000 per month)
- 28,600+ forks
- 1,400+ contributors since project launch
- 80+ community projects tracked on hermesatlas.com
- 1,302 commits in the Velocity Release cycle alone (12 days)
The high contributor count (around 1,400 in four months) is remarkable. For comparison: Khoj has around 110 contributors after three years. OpenClaw has just under 380 after eight months. Hermes blows past both.
The community project ecosystem is growing too. hermesatlas.com currently lists over 80 third-party projects built on Hermes, including skill collections, channel adapters, and specialized frontends.
8. Hermes Agent vs. OpenClaw vs. Open Interpreter
For the most important comparison question, the table below shows the key metrics for the three most popular agent frameworks as of June 2026:
Hermes Agent | OpenClaw | Open Interpreter | |
|---|---|---|---|
| GitHub Stars | 177,000 | 295,000 | 63,600 |
| Release Date | Feb. 2026 | Nov. 2025 | Sept. 2023 |
| Language | Python | Python | Python |
| License | MIT | MIT | AGPL-3.0 |
| Codebase Size | medium | ~124k lines | ~25k lines |
| Multi-Channel | 9 channels | 8+ channels | CLI-only |
| Persistent Memory | Yes | Plugin | No |
| Self-Improving Skills | Yes | No | No |
| Skill Marketplace | agentskills.io (smaller) | ClawHub (60k+ skills) | none |
| Skill Security Findings | n/a | 48% with findings | n/a |
It comes down to your use case. OpenClaw dominates in skill variety and absolute reach, but carries security risks from its open marketplace model. Open Interpreter is the best fit for natural-language code execution, but it's missing multi-channel and memory. Hermes Agent comes closest to combining both worlds, with multi-channel reach plus persistent memory and self-improving skills.
9. Security and Local Data Handling
Security is one of the main reasons Hermes has gotten so popular. Unlike OpenClaw, which shipped CVE-2026-25253 in January 2026 (a critical one-click RCE, CVSS 8.8), there are no known critical CVEs for Hermes so far.
The design principles are:
- No telemetry: Hermes doesn't collect usage data by default and sends nothing back to Nous Research.
- No cloud lock-in: All data stays local on your machine. Memory and skills included.
- Local models possible: With Ollama, Hermes can run fully offline, without a single token reaching external servers.
- Open standard skills: Skills use the open agentskills.io format and are portable and auditable.
One caveat. As soon as you use cloud LLMs (Claude, GPT, Gemini), your prompts go to the respective providers. For absolute privacy, you need Ollama plus a local model. You'll find a full cost breakdown in our article on Hermes Agent costs.
10. Market Position and Conclusion
Hermes Agent has shaken up the adaptive agent market in four months. 177,000 stars is a clear signal: demand for an open-source agent with persistent memory, multi-channel integration, and local data handling is huge.
Hermes is 2026's reference point in the self-improving agent space. If you want a personal agent that gets better over time, it's hard to look past Hermes. If you need the largest skill ecosystem, take a look at OpenClaw and its alternatives. If you want a broader overview of all Hermes alternatives, you'll find it in our comparison of the best Hermes Agent alternatives.
