42-evey/hermes-plugins

Custom plugins for hermes-agent — goal management, inter-agent bridge, model selection, cost control

What it solves

This project provides a suite of 23 custom plugins for the hermes-agent framework, designed to transform a standard LLM agent into a more autonomous, observable, and self-improving system. It addresses common agentic challenges such as cost control, hallucination detection, prompt injection, and long-term memory management.

How it works

The plugins extend the capabilities of hermes-agent by adding specialized tools and logic for different operational categories:

  • Autonomy & Decision-Making: Implements a "brain" for self-directed behavior through planning, reflection, and a multi-model debate system for complex decisions.
  • Observability & Telemetry: Provides structured JSON logging, real-time MQTT event streams, and budget enforcement via Langfuse.
  • Quality & Safety: Uses self-correction loops, regex-based hallucination detection, and dual-layer prompt injection screening.
  • Learning & Memory: Manages experiential learning, adaptive memory decay, and nightly consolidation of facts into vector databases (Qdrant).
  • Communication & Integration: Facilitates bidirectional communication with Claude Code and manages autonomous goal tracking.

Who it’s for

Developers and AI researchers building autonomous agents that require high reliability, safety guards, and the ability to evolve their personality and knowledge base over time.

Highlights

  • Multi-Model Debate: A 3-model consensus system for high-stakes decision making.
  • Adaptive Memory: A system that scores and decays memories based on importance to prevent clutter.
  • Cost Guard: Real-time budget tracking and alerting to prevent unexpected API costs.
  • Self-Updating Identity: An evolving personality system based on user interactions.
  • Sandboxed Execution: Secure code execution with resource limits.

Related

  • Project
    NousResearch/hermes-agent

    A self-improving AI agent with a built-in learning loop that creates and refines skills from experience and maintains persistent user memory across sessions.

  • Project
    NeoLabHQ/context-engineering-kit

    A plug‑in marketplace for LLM‑based coding assistants (Claude Code, Gemini, Antigravity, etc.) that supplies token‑efficient, context‑engineering patterns—reflection loops, sub‑agent judges, specification‑first development, code‑review agents, and Git helpers—to make AI‑generated code more reliable and easier to integrate into CI pipelines.

  • Project
    joeynyc/hermes-hudui

    A browser-based monitoring dashboard for the Hermes AI agent that provides visibility into its memory, costs, and internal state.

  • Project
    0xsline/awesome-deepseek-harness

    A curated ecosystem of plugins, tools, and infrastructure for DeepSeek Harness (DSH) to extend agent orchestration, memory, and context management.

  • Project
    ClaudioDrews/memory-os

    A local memory operating system for Hermes Agent that provides permanent, multi-layered memory and context injection to prevent agents from forgetting across sessions.