Hermes Agent v0.15 brings agent swarms, tool search, and faster autonomous workflows

Hermes Agent just received a major upgrade with the v0.15 “Velocity” update, and it’s a big step toward turning it into a full agentic operating system. The release focuses on speed, scalability, and smarter use of tools, while also adding powerful new features like agent swarms, a built-in MCP catalog, and support for new models and image generation.

What is Hermes Agent?

Hermes Agent is an open-source AI agent framework built by News Research under the MIT license. It’s designed as a persistent autonomous system that can run 24/7 on your own infrastructure.

Instead of just answering one-off prompts, Hermes is built to:

• Build long-term memory about your projects and preferences
• Create reusable skills over time
• Learn your workflows and improve the more you use it
• Coordinate tools, APIs, and external services to complete complex tasks autonomously

With the Velocity update, Hermes moves closer to acting like an AI operating system that can manage and orchestrate multiple agents and tools around your work.

Smarter tool search for huge tool stacks

As you connect more MCP servers, plugins, and external tools, each one exposes a schema that normally has to be loaded into the model’s context. When you have dozens or hundreds of tools, these schemas can quickly eat up your context window and slow everything down.

The new tool search system fixes this by using progressive, or “lazy,” loading of tools:

• Hermes no longer loads every tool definition up front
• Instead, it only loads a tool when the model actually needs it for the current task
• This keeps the context window clean and focused on reasoning, not tool definitions

The result is:

• Lower context usage
• Faster responses from the agent
• Better scalability as you connect larger MCP ecosystems
• More efficient multi-agent workflows

Core Hermes tools like file editing, terminal access, browser automation, memory, web search, and task delegation remain instantly available. Only external MCP and plugin tools are deferred until needed, so you still get responsiveness where it matters most.

Agent swarms for complex, multi-step tasks

One of the headline features in v0.15 is the new agent swarm system, integrated into the Hermes Kanban view. The Kanban acts as a visualization layer for everything happening across your sessions.

Previously, a single agent would try to handle an entire task end-to-end. With agent swarms, Hermes can now break a large goal into smaller objectives and distribute them across multiple specialized agents.

How agent swarms work

When you start a swarm, Hermes can:

• Decompose a big task into subtasks
• Assign each subtask to a specialized agent (researcher, builder, verifier, etc.)
• Coordinate their work and then synthesize the final result

For example, you might ask Hermes to build an AI SaaS landing page:

• Agent 1: Research competitors and gather examples
• Agent 2: Build the front-end layout and components
• Agent 3: Act as a verifier, review outputs, and ensure everything fits together

All of this can be monitored and managed through the Hermes web UI, which shows active sessions, tools being used, and the final output. In practice, this can even get close to cloning a real-world landing page layout, like Stripe’s, with minimal manual intervention.

Major codebase refactor for faster development

Behind the scenes, Hermes Agent received a huge internal cleanup. The core agent loop was reduced from over 16,000 lines of code to around 3,800 lines and split into multiple modules.

End users may not see this directly, but it has important benefits:

• Easier maintenance and debugging
• Faster shipping of new features
• Cleaner architecture for contributors and extension builders

This kind of refactor is critical for a fast-moving open-source agent platform that aims to support many tools, models, and workflows over time.

Built-in MCP catalog for safer, easier integrations

Instead of manually hunting down MCP servers from random GitHub repos and docs, Hermes now includes a built-in MCP catalog.

With the catalog, you can:

• Discover available MCP integrations directly inside Hermes
• Install and manage them from a centralized, curated list
• Reduce the risk of pulling in unsafe or unmaintained integrations

This makes it much easier to grow your tool ecosystem without constantly copy-pasting configs from external sources.

New model support: Qwen, Opus, and Creata 2

Hermes Agent v0.15 also expands its model support, giving you more flexibility depending on your budget and use case. If you want a deeper dive on model roles within Hermes (orchestrators, executors, auxiliaries), you may find this guide to top AI models for Hermes Agent helpful.

Qwen 3.7 Max

• Strong at web development and coding tasks
• Token-efficient and relatively affordable
• A great option if you want strong performance without the cost of the biggest proprietary models

Opus 4.8

• High-end model with strong capabilities
• More expensive than many alternatives
• Positioned as a premium choice, though still not seen as surpassing the very latest frontier models like GPT‑5.5 in overall capability

Creata 2 for image generation

Hermes now integrates Creata 2, a powerful image model focused on:

• High visual quality
• Fine-grained creative control
• Style transfer and mood boards
• Adjustable creativity settings

This means Hermes can act not just as a text and code agent, but also as an image generation provider that you can wire into your workflows.

Faster session search and better security

The Velocity update also brings a rebuilt session search system:

• Completely free to use
• Does not rely on a large language model
• Reported to be up to 4,500× faster for searching past Hermes sessions

This is especially useful if you rely on Hermes as a long-running assistant and need to quickly retrieve past work, decisions, or outputs.

On the security side, Hermes now includes:

• Prompt injection defenses to help protect against malicious or manipulated inputs
• Bitwarden and secrets manager integration, so sensitive keys and credentials can be handled more safely

New workflow and deployment integrations

Hermes Agent v0.15 also adds several quality-of-life features aimed at developers and power users:

• Skills bundles – package related skills together for easier reuse and sharing
• Netlify integration – streamline deployment workflows for web projects Hermes helps you build
• TUI session orchestrator – manage multiple live Hermes sessions from a single terminal window, and connect other agents without restarting Hermes

Cold-start performance has also been improved, so Hermes spins up faster, and OpenHands skill integration has been enabled, further expanding what the agent can do in coding and automation scenarios.

Getting started with the Velocity update

Hermes Agent is available across major operating systems, including native support for Windows. Once installed, you can update to the latest Velocity release directly from the command line using a simple update command.

From there, you can:

• Connect MCP servers and explore the new catalog
• Experiment with agent swarms for multi-step projects
• Try different models like Qwen 3.7 Max or Opus 4.8
• Wire in Creata 2 for image generation workflows

If you want to keep up with broader model and platform updates around Hermes, you may also be interested in our roundup of new AI models and next‑gen image systems.

Overall, Hermes Agent v0.15 is a substantial upgrade that makes the platform faster, more modular, and more capable of handling real-world autonomous workflows at scale.