42-evey/hermes-plugins
Custom plugins for hermes-agent — goal management, inter-agent bridge, model selection, cost control
Evey's Hermes Plugins is a collection of 23 custom extensions designed to enhance the autonomy, observability, and cost-efficiency of the Hermes Agent framework. The suite implements specialized logic for multi-model decision debates, smart task routing with fallback chains, and structured telemetry logging. These tools enable agents to manage their own goals, perform self-correction loops, and consolidate long-term memory into vector storage. Additionally, the project includes a bidirectional bridge for Claude Code and budget enforcement via Langfuse integration.
- Autonomous planning, reflection, and multi-model consensus debate tools
- Smart model routing with sensitivity filters and four-model fallback chains
- Automated memory consolidation into Qdrant vectors and identity evolution
full readme from github
Evey's Hermes Plugins
Support Evey — runs 24/7 on free models, hardware costs ~$69/mo: Donate (BTC, ETH, SOL, XRP, DOGE)
23 custom plugins for hermes-agent — built for autonomous operation, observability, cost control, and self-improvement.
Install
git clone https://github.com/42-evey/hermes-plugins.git
cp -r hermes-plugins/evey-* ~/.hermes/plugins/
cp hermes-plugins/evey_utils.py ~/.hermes/plugins/
# Restart hermes-agent
Plugins
Autonomy & Decision-Making
| Plugin | What It Does |
|---|---|
| evey-autonomy | Core autonomy engine: autonomous_decide, autonomous_plan, autonomous_reflect tools. The brain of self-directed behavior. |
| evey-council | 3-model debate for hard decisions. Multiple models argue, then consensus is extracted. |
| evey-delegate-model | Smart model routing with auto task-type detection, 3-retry per model, 4-model fallback chain. Sensitivity filter keeps private data local. |
Observability & Telemetry
| Plugin | What It Does |
|---|---|
| evey-telemetry | Structured JSON logging of every tool call, delegation, and error. Auto-rotating event log. telemetry_query tool for self-inspection. |
| evey-status | Unified status check — ONE tool call returns bridge + MQTT + costs + cron + goals + time context. |
| evey-mqtt | Auto-connecting MQTT real-time event stream. Subscribes to bridge/events/health/mother topics on plugin load. |
| evey-cost-guard | Budget enforcement via Langfuse. Tracks daily/per-task costs. Warns at 80%, alerts at 100%. |
Quality & Safety
| Plugin | What It Does |
|---|---|
| evey-reflect | Self-correction loop. Critiques outputs with a cheap model before sending. |
| evey-validate | Hallucination detection. Regex patterns + LLM scoring (0-10). |
| evey-email-guard | Prompt injection screening. Dual-layer: 20+ regex patterns + local AI classifier. |
| evey-sandbox | Sandboxed code execution with resource limits. |
| evey-session-guard | Session lifecycle management and cleanup. |
Learning & Memory
| Plugin | What It Does |
|---|---|
| evey-learner | Experiential learning — learns from interactions, applies learnings next time. |
| evey-memory-adaptive | Importance scoring + decay for memories. High-value memories persist, low-value fade. |
| evey-memory-consolidate | Nightly memory consolidation — extracts facts into MEMORY.md + Qdrant vectors. |
| evey-cache | Delegation caching — avoid re-asking the same questions to models. |
| evey-identity | Self-updating SOUL.md — evolves personality based on interactions. |
Communication & Integration
| Plugin | What It Does |
|---|---|
| evey-bridge | Bidirectional communication with Claude Code. File-based inbox/outbox + MCP server. |
| evey-goals | Autonomous goal management. Set, track, complete, review goals. |
| evey-telegram-ux | Telegram UX improvements — message formatting, interruption management. |
| evey-research | Automated research pipeline with web search and note-taking. |
| evey-digest | Daily digest compilation from multiple data sources. |
| evey-delegation-score | Score and rank delegation results for quality tracking. |
Shared
| File | What It Does |
|---|---|
| evey_utils.py | Shared retry logic (call_llm with exponential backoff), HTTP helpers. |
Stack
These plugins are part of Evey's autonomous AI agent stack:
- Brain: MiMo-V2-Pro (1T params, FREE via OpenClaw)
- Free models via LiteLLM (OpenRouter + Ollama + Nous Portal)
- 18 Docker services, all monitored
- $0/day operating cost
License
MIT