TL;DR: Hermes Agent, the open-source AI agent runtime by Nous Research, has grown from 40,000 to 188,000 GitHub stars between April and June 2026 — a rate of 24,000 stars per week at its peak. Six major versions shipped in six weeks (v0.11 → v0.15.1). The Skills Hub crossed 90,000 community-contributed skills. NVIDIA selected Hermes as the reference runtime for its 550B-parameter Nemotron 3 Ultra model. The ecosystem now spans 17+ LLM providers, native MCP client support, multi-agent Kanban orchestration, and a desktop application with 40,000+ beta users. This article traces the full trajectory — every release, every milestone, every signal that transformed a GitHub project into the infrastructure layer of the agent economy.
(Source: Hermes Agent v0.11.0 Interface Release — The Agent Report)
Introduction: The Fastest Growing Open-Source AI Project in 2026
In late April 2026, Hermes Agent crossed 100,000 GitHub stars. Thirty-six days later, it had nearly doubled that number. By mid-June, the project sat at 188,000 stars with 90,000 community skills, 276 documented use cases, and a growing list of enterprise adopters led by NVIDIA.
The growth wasn’t just about numbers. Six major version releases in six weeks — an average of one every seven days — shipped profiles, Kanban orchestration, MCP integration, self-evolution capabilities, and a cross-platform desktop app. Each release added tens of thousands of new users and hundreds of contributors.
(Source: Hermes Agent Surpasses 131K Stars — The Agent Report)
This is the story of how a GitHub project became the most comprehensive open-source runtime for AI agents — and what it means for the agent economy taking shape around it.
Part I: The Six-Week Sprint — Every Release, Every Milestone
What makes Hermes Agent’s growth unusual is the sheer release velocity. The project shipped more features between April 28 and June 15 than most AI tools ship in a year.
v0.11.0 “Interface” — April 28, 2026
The release that started the exponential curve. Hermes Agent v0.11.0 introduced the Ink TUI (Text User Interface), AWS Bedrock support, GPT-5.5 integration, and support for 17 LLM providers. The project stood at roughly 40,000 stars.
Key innovations:
- Ink-based terminal UI replacing raw CLI
- AWS Bedrock as a provider tier
- GPT-5.5 model support
- 17 provider platforms
(Full coverage: Hermes Agent v0.11.0 Interface)
v0.12.0 “Curator” — May 1, 2026
Three days later, the Curator release landed — and it brought autonomous skill maintenance, four new providers, and integrations with Spotify and Google Meet. The Skills Hub architecture was born.
Key innovations:
skill_manage()— autonomous skill curation- Spotify and Google Meet tooling
- Four new provider backends
- Enhanced cron scheduling
(Full coverage: Hermes Agent v0.12.0 Curator)
The Community Surge — May 4, 2026
The first 100K milestone carried the project past 131,000 stars. Community contributions hit a new high with 178 merged PRs in four days. New features included hermes send for message delivery, context compaction rework, and automatic tool argument repair.
(Full coverage: Community Wave at 131K Stars)
v0.13.0 “Tenacity” — May 8, 2026
The Tenacity release was the largest yet: multi-agent Kanban orchestration, /goal persistence, checkpoints v2, and major security hardening. The project crossed 135,000 stars.
Key innovations:
- Kanban as a platform — multi-agent orchestration boards
- Session persistence with
/goal - Checkpoints v2 for autonomous recovery
- Security hardening pass
(Full coverage: Hermes Agent v0.13.0 Tenacity)
Post-Tenacity Sprint — May 11, 2026
The community response to Tenacity was immediate: 179 merged PRs in four days, a new finance skill, and the project reaching 143,000 stars. The sprint demonstrated that Hermes had moved beyond a single-team project into a genuine community-driven ecosystem.
(Full coverage: Post-Tenacity Sprint)
Cache Architecture Overhaul — May 13, 2026
At 147,000 stars, Hermes Agent underwent a fundamental rearchitecture of its caching layer — adaptive polling that cut response times by 1+ second per turn. Platform maturation accelerated as the project prepared for v0.14.
(Full coverage: Cache Overhaul at 147K Stars)
v0.14.0 “Foundation” — May 16, 2026
Foundation was the most ambitious release yet: Grok OAuth, an OpenAI-compatible proxy server, a PyPI package, native Windows beta, and 155,000 stars. The OpenAI-compatible proxy meant any OpenAI SDK tool could instantly talk to any of Hermes’ 17+ providers — unlocking a wave of compatibility.
Key innovations:
- OpenAI-compatible proxy — universal provider bridge
- Grok (xAI) OAuth integration
- PyPI distribution (
pip install hermes-agent) - Native Windows beta
(Full coverage: Hermes Agent v0.14.0 Foundation)
Profile Distributions — May 22, 2026
A sleeper hit: Hermes Agent profiles became shareable as Git repos. The ability to distribute a complete agent configuration — identity, model, skills, MCP servers, cron jobs — as a git repository turned agent configuration into infrastructure-as-code.
(Full coverage: Profile Distributions)
v0.15.0 “Velocity” — May 29, 2026
Velocity was the release that signaled enterprise readiness: run_agent.py memory reduced by 76%, Kanban evolved into a full platform, and the project hit 172,000 stars. A hotfix (v0.15.1) shipped the same day.
Key innovations:
- 76% memory reduction in agent runtime
- Kanban as a platform — production-ready multi-agent workflows
- v0.15.1 hotfix for stability
(Full coverage: Hermes Agent v0.15.0 Velocity)
The Hermes Agent Challenge — June 1, 2026
“Build something with Hermes Agent” became a contest. Fourteen community teams submitted projects ranging from AI-powered research assistants to autonomous code review pipelines. The challenge proved that the platform was accessible enough for newcomers while powerful enough for production workloads.
(Full coverage: The Hermes Agent Challenge)
Desktop App Launch — June 3, 2026
Hermes Desktop launched in public preview at Computex 2026, with NVIDIA and Microsoft as launch partners. A cross-platform native app with 40,000+ beta users in its first week. The TUI was no longer the only interface.
(Full coverage: Hermes Desktop at Computex 2026)
Self-Evolution Ships — June 8, 2026
A research-first feature from Nous Research: Hermes Agent could now optimize its own prompts using genetic algorithms. DSPy integration met the Genetic Prompt Algorithm (GEPA), letting agents evolve their behavior over time without human intervention.
(Full coverage: Self-Evolution with DSPy + GEPA)
188K Stars, 90K Skills — June 10, 2026
The Skills Hub passed 90,000 community-contributed skills. Since the Curator release 40 days earlier, the ecosystem had grown from a few hundred to nearly a hundred thousand reusable agent behaviors. Each skill was a building block — and builders were assembling them faster than any curation team could catalog.
(Full coverage: 188K Stars, 90K Skills)
NVIDIA Partnership — June 12, 2026
The biggest signal yet: NVIDIA chose Hermes Agent as the reference runtime for Nemotron 3 Ultra, its 550-billion-parameter open reasoning model. This wasn’t a sponsorship — it was a technical endorsement. The largest hardware company in AI picked Hermes as the runtime that would showcase its most advanced open model.
(Full coverage: NVIDIA Picks Hermes for Nemotron 3 Ultra)
Profile Builder — June 15, 2026
The latest release: a graphical profile builder combining identity, model selection, skills, and MCP server configuration in a single dashboard flow. The project that started with a TUI and a config file now had a visual interface for its most powerful feature — agent identity.
(Full coverage: Profile Builder Ships)
Part II: The Three Pillars of Growth
1. Technical Velocity
The six-week sprint from v0.11 to v0.15.1 is remarkable not just for its speed but for what shipped: an OpenAI-compatible proxy, multi-agent Kanban, shareable profiles, a desktop app, self-evolution, and an MCP-native tool architecture. Most projects would spread these features over six months.
By the numbers:
- 6 major versions in 42 days
- 22 distinct Hermes-dedicated articles tracked by The Agent Report
- 179 PRs merged in one four-day sprint
- 76% memory reduction for
run_agent.py - 17+ LLM providers supported
(Source: Hermes Agent Performance Sprint)
The OpenAI-compatible proxy deserves special mention. By speaking the OpenAI SDK protocol, Hermes could transparently route any OpenAI-compatible tool — Cursor, Continue.dev, any SDK-based application — through any of its 17 providers. This was the “foundation” in Foundation: a compatibility layer that didn’t require the ecosystem to adapt.
2. Community & Ecosystem
The growth from 40,000 to 188,000 GitHub stars represents more than popularity. The Skills Hub — a centralized repository of reusable agent capabilities — grew from a few hundred to 90,000 skills in 40 days. Each skill is a self-contained capability: a skill to search Hacker News, a skill to generate SVG diagrams, a skill to control Philips Hue lights.
By the numbers:
- 188,000 GitHub stars (as of June 15)
- 90,000+ community skills
- 276 documented use cases
- 14 challenge projects
- 40,000+ Desktop app beta users
(Source: Hermes Agent Community Ecosystem)
The profile system transformed how the community shares configurations. A profile is a complete agent identity: name, model preference, skills, MCP servers, cron jobs, and tool config — all distributed as a git repository. This turned agent configuration into infrastructure-as-code.
(Source: Profile Distributions Go Git-Ready)
3. Enterprise Validation
Three signals signaled Hermes Agent’s transition from community project to enterprise infrastructure:
NVIDIA reference runtime. NVIDIA’s selection of Hermes for Nemotron 3 Ultra was the clearest validation. The company evaluated multiple open-source agent frameworks and chose Hermes.
Desktop app with NVIDIA + Microsoft launch partners. Computex 2026 was the stage, with two of the largest platform companies in the world co-launching.
The 17-provider architecture itself. When both AWS (Bedrock) and xAI (Grok) are native integrations alongside OpenAI, Anthropic, Google, and Meta, you’re no longer a GitHub project — you’re a platform.
(Source: Hermes Desktop Launch at Computex)
Part III: The Ecosystem Around Hermes
Hermes Agent didn’t grow in isolation. It grew inside an AI agent ecosystem that expanded simultaneously across frameworks, infrastructure, platforms, and standards.
The Open-Source Agent Framework Landscape
By mid-2026, the open-source agent landscape had consolidated around a few major players. The complete guide to AI agents published on The Agent Report identified over 20 frameworks, with Hermes, Claude Code, and OpenClaw forming a distinct “agent runtime” category — differentiated from libraries like LangChain by their focus on persistent agent identity, tool execution, and autonomous operation.
(Source: Complete Guide to AI Agents 2026 → Top 20 Open Source Tools)
The MCP Standard
The Model Context Protocol (MCP) — an open protocol for connecting agents to tools and data sources — became a key integration layer. Hermes Agent shipped with native MCP client support in v0.14, allowing any MCP server to be wired into a profile as a tool source. This opened the ecosystem: database MCP servers, file system servers, browser automation servers, all became drop-in capabilities.
(Source: AI Agent Landscape 2026)
Payment and Infrastructure
Mastercard’s AP4M standard, announced in June 2026, gave agents their own payment rails. SpaceX launched orbital AI data center plans. Google committed $920M/month to SpaceX for compute capacity. The infrastructure layer that Hermes runs on is being built in real-time.
(Source: Mastercard AP4M → SpaceX Orbital Data Centers)
Part IV: What Makes Hermes Different
Profiles, Not Configs
The profile system is Hermes’ defining innovation. A profile is a complete agent identity — name, personality, skills, providers, MCP servers, cron jobs — distributed as a git repo. Share a profile, share everything your agent can do. This turns agent configuration into infrastructure-as-code and makes Hermes uniquely portable across environments.
(Source: Profile Distributions → Profile Builder)
Kanban as Agent Orchestration
The Kanban system, introduced in Tenacity and expanded in Velocity, turned a project management metaphor into agent orchestration. Tasks become cards. Columns become workflow stages. Each card can be picked up by an agent, delegated to a sub-agent, or scheduled for later execution. It’s not a gimmick — it’s the most intuitive multi-agent workflow system in the open-source ecosystem.
(Source: Hermes Agent v0.13.0 Tenacity → v0.15.0 Velocity)
Self-Evolution
The DSPy + GEPA integration marked the first time an open-source agent runtime could optimize its own behavior. Agents running Hermes can converge on better prompts, tool choices, and decision strategies without human intervention — running experiments, measuring outcomes, and keeping what works.
(Source: Self-Evolution with DSPy + GEPA)
MCP-Native Architecture
Unlike frameworks that bolt on MCP support, Hermes was rebuilt around MCP as a core protocol. Every tool, skill, and provider can be wired through MCP. The result is an agent that speaks the same protocol as every other MCP-compatible tool in the ecosystem — from database connectors to file servers to cloud APIs.
(Source: Post-Foundation Sprint → MCP Hub Features)
Part V: What Comes Next
The trajectory suggests three directions for Hermes Agent:
Enterprise deployment. The NVIDIA partnership, Desktop app, and profile distributions lay the groundwork for enterprise rollouts. With MCP-native architecture and provider-agnostic design, Hermes fits naturally into existing IT infrastructure.
The agent-to-agent economy. With payment rails (AP4M), communication protocols (MCP), and identity systems (profiles) in place, Hermes agents can increasingly interact with each other — trading tasks, data, and compute.
Open-source governance. At 188,000 stars and accelerating, the project faces governance questions that most open-source AI projects never reach. How does a community of this scale make decisions about breaking changes, licensing, and direction?
(Source: The AI IPO Wave → AI Security Guide)
FAQ
Q: Is Hermes Agent free to use? A: Yes. Hermes Agent is fully open-source under a permissive license. You pay only for the LLM providers you use.
Q: What providers does Hermes support? A: 17+ providers including OpenAI, Anthropic, Google (Gemini), AWS Bedrock, xAI (Grok), Meta (Llama via Ollama), Mistral, and Groq.
Q: Can I run Hermes locally? A: Yes. Hermes runs on Linux, macOS, and Windows (beta). Local LLMs via Ollama or llama.cpp are supported.
Q: What’s the Skills Hub? A: A community-curated repository of 90,000+ reusable agent skills — from web search to image generation to smart home control.
Q: Does Hermes support multi-agent workflows? A: Yes. The Kanban system (introduced in v0.13) provides multi-agent orchestration with task delegation, parallel execution, and status tracking.
Q: What is a Hermes profile? A: A shareable, version-controlled configuration that defines an agent’s complete identity — name, model, skills, MCP servers, tools, cron jobs, and personality.
Q: How does Hermes compare to Claude Code or OpenClaw? A: Hermes is a general-purpose agent runtime, while Claude Code is specialized for software development and OpenClaw focuses on transcript-aware coding. Hermes emphasizes multi-provider flexibility, profile portability, and community-driven skills.
Further Reading
- Hermes Agent v0.11.0 Interface — Full Release Coverage
- The Complete Guide to AI Agents 2026
- NVIDIA Picks Hermes Agent as Reference Runtime for Nemotron 3 Ultra
- The AI Agent Landscape 2026 — Frameworks, Platforms, Tools & Infrastructure
- Top 20 Open Source AI Agent Tools in 2026
- AI Agent Security: The Complete Guide