Hermes Agent Plugin Development — Build Custom Extensions

Extend Hermes Agent with custom plugins, skills, and tools tailored to your workflows. Build integrations with your internal systems, create domain-specific automation, and package reusable capabilities that your agents can invoke on demand.

Start a project Read the guide

Plugin Architecture

Plugins live in ~/.hermes/plugins/ and register tools, hooks, and CLI commands. The PluginManager discovers them at startup — no core changes needed.

Tool Registration

Each plugin can register as many tools as needed via ctx.register_tool(...). Tools get full schema support, availability gating via check_fn, and error wrapping.

Lifecycle Hooks

Hook into pre_tool_call, post_tool_call, pre_llm_call, post_llm_call, on_session_start, and on_session_end to build observability, validation, and stateful plugins.

Skills System

Skills are markdown-driven capability packs with SKILL.md frontmatter, scripts, templates, and references. They inject domain knowledge and procedures into the agent's context on demand.

What We Build

Plugin Categories We Support

General Plugins

Tools, hooks, CLI commands. The most flexible category — anything that doesn't fit elsewhere goes here.

Memory Providers

Pluggable memory backends implementing the MemoryProvider ABC. Honcho, mem0, supermemory, and custom implementations.

Model Providers

Inference backends — OpenAI, Anthropic, Gemini, OpenRouter, custom endpoints. Register via ProviderProfile.

Context Engines

Custom context assembly and retrieval strategies. Plug into the agent's context pipeline for domain-specific enrichment.

Image Generation

Providers for DALL-E, Stable Diffusion, ComfyUI, and other image generation backends — tools agents can call to create visuals.

Kanban & Dispatch

Multi-agent work queues with SQLite-backed boards, assignees, reviews, and auto-dispatch. Workers spawned per-task with isolated contexts.

Development Approach

  1. 1Requirements gathering: what integrations, tools, or workflows does the client need? What existing systems must Hermes connect to?
  2. 2Architecture: choose plugin category, define tool schemas, design hook integration points, plan skill structure
  3. 3Implementation: write the plugin in Python, register tools and hooks, package skills with scripts and references, test with real Hermes startup
  4. 4Integration testing: verify the plugin loads correctly, tools appear in the agent's schema, hooks fire at the right times, skills inject properly
  5. 5Delivery & documentation: install the plugin, document its tools and usage, hand over maintenance guidance

Ideal For