This is an early release preview. You may encounter bugs.
Hermes Plugins logo
Unclaimed

Tool monitoring security coding

Hermes Plugins

23 Hermes Agent plugins covering autonomy, observability, cost control, safety screening, and memory

C 54/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
No votes yet

01 / About

What Hermes Plugins is.

Hermes Plugins is a set of 23 plugins for hermes-agent, written for an agent named Evey that runs continuously and unattended. The plugins extend a Hermes instance with self-directed decision-making, telemetry and budget controls, output checking, memory management, and channels to other agents and services. They install by copying the evey-* directories and a shared evey_utils.py helper into the Hermes plugins folder.

The plugins are grouped into five areas. Autonomy plugins add planning, reflection, a multi-model debate for hard decisions, and model routing with retries and a fallback chain. Observability plugins log every tool call as structured JSON, expose a single status check, stream events over MQTT, and enforce daily and per-task budgets through Langfuse. Quality and safety plugins critique outputs before sending, score outputs for hallucination, screen incoming email for prompt injection, and sandbox code execution. Learning and memory plugins score, decay, cache, and consolidate memories. Communication plugins bridge to Claude Code, manage goals, and run research and digest pipelines.

Features

  • evey-autonomy: autonomousdecide, autonomousplan, and autonomous_reflect tools for self-directed behaviour
  • evey-council: three models debate a decision and a consensus is extracted
  • evey-delegate-model: task-type detection, three retries per model, a four-model fallback chain, and a sensitivity filter that keeps private data on local models
  • evey-telemetry: structured JSON logs of tool calls, delegations, and errors with a telemetry_query tool
  • evey-cost-guard: daily and per-task cost tracking via Langfuse with warnings at 80% and alerts at 100% of budget
  • evey-mqtt and evey-status: a real-time MQTT event stream and a one-call status summary covering bridge, MQTT, costs, cron, goals, and time
  • evey-reflect and evey-validate: a self-critique pass with a low-cost model and hallucination scoring (regex plus LLM score 0–10)
  • evey-email-guard: prompt-injection screening using 20+ regex patterns and a local classifier
  • evey-sandbox and evey-session-guard: sandboxed code execution with resource limits and session lifecycle cleanup
  • Memory plugins: importance scoring with decay, nightly consolidation into MEMORY.md and Qdrant vectors, delegation caching, and a self-updating SOUL.md
  • evey-bridge: two-way communication with Claude Code through a file inbox/outbox and an MCP server
  • Workflow plugins: goal tracking, Telegram formatting, a research pipeline, daily digests, and delegation-result scoring

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

03 / Build

Build with Hermes Plugins.

Browse the catalogue for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get Hermes Plugins →

Browse the catalogue