Hermes Agent v0.15! Huge New Updates: Agent Swarms, Tool Search, NEW Models, & More!

Key Concepts

• Hermes Agent: An open-source, persistent autonomous AI system designed for long-term memory, skill building, and workflow evolution.
• Velocity Update: A major release for Hermes Agent focusing on performance, scalability, and multi-agent orchestration.
• Tool Search (Lazy Loading): A mechanism to dynamically load tool schemas only when required, preserving context window space.
• Agent Swarm: A system that breaks complex tasks into sub-tasks, distributing them among specialized agents.
• MCP (Model Context Protocol): A standard for connecting AI agents to external data sources, tools, and services.
• TUI (Terminal User Interface): A text-based interface for managing multiple agent sessions and orchestrating workflows.

---

1. Tool Search and Context Optimization

The "Velocity Update" introduces Tool Search, a progressive loading system designed to solve the issue of "context bloat."

• The Problem: Previously, loading dozens or hundreds of MCP (Model Context Protocol) tool schemas consumed significant portions of the LLM's context window, leading to faster token exhaustion and reduced reasoning capacity.
• The Solution: Hermes now employs a "lazy loading" approach. Core tools (file editing, terminal access, browser automation, memory, web search) remain active, while external MCP tools are deferred until the model explicitly requires them for a specific task.
• Benefits: Faster agent responses, improved scalability for large MCP ecosystems, and more room for the model to perform complex reasoning.

2. Agent Swarm System

The new Agent Swarm feature allows for the decomposition of complex projects into manageable sub-tasks.

• Methodology: Users initiate a swarm via the command Hermes canban swarm. The system automatically breaks a high-level prompt into specialized objectives.
• Example: When tasked with building an AI SaaS landing page, the system deployed:
  • A Research Agent to analyze competitors (e.g., Stripe).
  • A Frontend Agent to build the UI components.
  • A Verifier Agent to review outputs and synthesize the final product.
• Management: Users can monitor the swarm’s progress through the Hermes web UI, which tracks tool usage and session status in real-time.

3. Codebase Refactor and Performance

The development team performed a massive refactor of the Hermes codebase to improve maintainability and speed.

• Technical Detail: The core agent loop was reduced from over 16,000 lines of code to approximately 3,800 lines.
• Modularity: The code was split into multiple modules, facilitating faster feature shipping and easier community contributions.
• Cold Start: Significant improvements were made to the system's cold start performance, making the agent more responsive upon initialization.

4. Ecosystem and Integration Updates

• Built-in MCP Catalog: A centralized, safer repository for discovering and installing MCP integrations, replacing the need to manually hunt for repositories on GitHub.
• New Model Support:
  • Qwen 3.7 Max: Recommended for web development due to high token efficiency and cost-effectiveness.
  • Opus 4.8: Supported for high-end performance tasks.
  • Creata 2: Integrated for image generation, offering fine-grained creative control and style transfer.
• Platform Expansion: Hermes is now natively available on Windows, accessible via the command prompt. Existing users can update by typing Hermes update.

5. Rapid-Fire Quality of Life Improvements

• Session Search: A rebuilt search system that is reportedly 4,500 times faster and operates without requiring an LLM.
• Security: Added prompt injection defense and a Bitwarden secrets manager integration.
• Orchestration: A new TUI session orchestrator allows users to manage multiple live sessions within a single terminal window and switch between agents without restarting the system.
• Compatibility: Enabled integration with OpenHands skills.

---

Synthesis and Conclusion

The "Velocity Update" marks a significant shift for Hermes Agent, moving it closer to a fully realized "Agentic AI operating system." By prioritizing context efficiency through Tool Search and enabling complex, multi-agent workflows via the Agent Swarm, the project addresses the primary bottlenecks of autonomous AI: token management and task complexity. The combination of a refactored, modular codebase and a centralized MCP catalog suggests a focus on long-term sustainability and user-friendly adoption, positioning Hermes as a robust tool for developers and power users looking to automate complex, persistent workflows.