Hermes Agent — Community Guide

Practical guides for setting up and extending Hermes Agent.

Operator, Not Builder

Hermes is not a Builder. Hermes is an Operator.

• Builders create one-off artifacts: dashboards, static sites, prototypes, one-shot scripts.
• Operators run recurring work: briefings, alerts, scheduled reports, monitoring, memory maintenance, skill execution, escalation handling.

The pattern: a builder creates the surface, Hermes operates the workflow. Example: Claude builds a dashboard — Hermes runs daily analysis against it and sends the brief. Another agent writes a script — Hermes schedules it, monitors it, and escalates on failure.

If the task is "make me a thing," route to a builder. If the task is "keep this running," "watch for this," or "tell me when," that's Hermes.

Start Here

• LLM Start Here — Machine-oriented bootstrap contract for interpreting this repository: reading order, authority hierarchy, evidence states, discovery rules, safety boundaries, and the standard architecture-handoff shape.
• New Hermes Instance Operator Guide — The canonical setup sequence for a fresh Hermes instance or profile: install, model/provider config, toolsets, skills/proficiencies, memory, profile setup, gateway, cron, Kanban, verification, and privacy rules.
• SOUL.md Example — Public-safe templates for profile-level agent behavior: role, tone, tool rules, approval gates, quality bar, and the boundary between SOUL.md, AGENTS.md, memory, and skills.
• Copy-Ready Public SOUL.md — Compact, sanitized operator profile that can be copied into a Hermes installation and adapted without exposing private runtime details.
• Multi-Agent Profiles — How to run multiple specialized Hermes agents with isolated profiles, dedicated SOUL.md files, shared AGENTS.md, and separate invocation via hermes -p NAME or wrapper scripts.
• GBrain Memory Plugin — GBrain pull-memory provider for Hermes, including profile-specific durable context pages, active-source discovery, the legacy standalone plugin notes, setup commands, privacy notes, and verification guidance.
• Hermes Dreaming Plugin — Background memory consolidation for Hermes, including hermes plugins install alejandroiglesias/hermes-dreaming, review/run commands, audit files, and safety notes for memory-mutating workflows.
• Installing Skills from a URL — How to install a Hermes skill directly from any .md URL, including Gists and raw hosted skill files, then refresh it later with hermes skills update.
• Recommended Skills to Install — Profile-specific list of external/community skills worth adding, including last30days and its target profiles.
• Skill Bundles — How to group repeatable skill clusters into one slash command with hermes bundles create.

Operations Guides

• Legacy Workspace Auto-Start & Auth — Reference for the previous Hermes Workspace setup. The selected browser interface for this operator setup is now nesquena/hermes-webui, not Hermes Workspace.
• Message Queuing, Telegram Batching & Rich Messages — How display.busy_input_mode (interrupt vs queue) controls what happens when the agent is busy, plus Telegram-specific text/media batching and native rich-message rendering.
• Cloudflare Tunnel — How to expose your local Hermes gateway to the internet using Cloudflare Tunnel.
• 1Password CLI — Using 1Password with Hermes Agent for secure credential management.
• Python Environment with uv + venv — How to run Hermes from a uv-managed virtualenv, fix hermes doctor warnings about global Python, and make shell hermes resolve to the venv.
• Telegram Mini App Setup — Full walkthrough for deploying the Hermes Telegram Mini App with standalone proxy architecture, Cloudflare Tunnel, auth, and the gateway API.
• Mini App Standalone Proxy Plan — Detailed implementation plan for a standalone mini-app proxy that decouples mini-app support from Hermes core.
• Extending the Gateway API — How to add new endpoints, middleware, and features to the Hermes API server.

Daily Operator Surfaces

These are worth checking on any serious Hermes setup because they change daily operation, not just setup polish:

• Context-compression indicator — watch compression count as a session-length signal; compression keeps work moving but repeated compression means you may need to split or persist context.
• /browser connect / browser automation — use it when logged-in browser state or real UI interaction matters; prefer deterministic readers/scripts for repeatable extraction when available.
• Hermes WebUI (selected) — the browser and phone interface used by this operator setup for chat, sessions, profiles, workspaces, tasks, skills, memory, and other Hermes controls. It runs against the existing Hermes installation and models.
• Official Web Dashboard — an alternative administration surface documented by Hermes for config, sessions, logs, credentials, cron jobs, skills, gateway, webhooks, and memory.
• Desktop App — use the native app when a visual/operator workspace is better than pure CLI or chat, especially for streaming tool output, files, profiles, cron, skills, settings, and voice.

Official docs: Browser Automation, Web Dashboard, Desktop App, Context Compression.

Routing, Models, and Coordination

• Model Tier Rankings — Best LLM models for Hermes Agent by role (orchestrator, executor, auxiliary), based on community testing and fleet experience.
• Tiered Model Routing — How to configure delegate-first routing with high-reasoning, budget, specialist, local, and fallback lanes.

Research, Memory, and Context Extraction

• Daily Personal-Context Questions — A daily Hermes cron pattern that asks one thoughtful personal question per day, files the answer into durable memory, and accumulates rich user context over weeks.
• Q100 Taste Interview — A 100-question structured extraction protocol for capturing writing voice, taste, aesthetic boundaries, and structural preferences.
• PageIndex Evaluation — Technical evaluation of VectifyAI/PageIndex, a reasoning-based RAG system with architecture notes and integration tradeoffs.
• xurl — X API CLI for Hermes — How to install, authenticate, configure apps/users/redirects/auth modes, safely verify, and use the official X Developer Platform CLI from Hermes.
• Lessons Learned — Hard-won Hermes operating rules: PRD-first workflows, circuit breakers, self-scoring, quality gates, explicit handoffs, and bounded second-order improvement loops.
• Agent Honesty and Cost Controls — Practical profile/skill/cron guardrails for uncertainty, verify-before-done behavior, hook-style checks, independent verification, push-memory hygiene, and token-cost controls.
• Agent-Native Toolbelt Implementation Plan — Plan for converting repeated agent workflows into deterministic CLIs with local mirrors, budget guardrails, Hermes skills, and compound commands.
• Signal-to-Skill Report — Weekly loop for scanning Hermes tool-use patterns, ranking Top 5 reusable workflow candidates, applying them as skills, and verifying the result.

Operational Monitoring

Hermes runs automated weekly jobs that measure and maintain operational health. Data is collected from:

• ~/.hermes/skill-health.log — skill health check results
• ~/.hermes/cost-log.csv — per-task cost tracking (schema defined, auto-population blocked — see Lessons Learned)
• ~/.hermes/daily-logs/failures.md — documented failures
• ~/.hermes/scorecards/ — historical scorecard results

Weekly Cron Schedule

Time	Job	Script
Monday 10:00	Hermes Scorecard (10 dimensions, 0–5 scoring, max 50)	~/.hermes/skills/hermes-scorecard/scripts/score.py
Monday 11:00	Signal-to-Skill Pipeline	~/.hermes/scripts/signal-to-skill.py
Monday 11:30	Project Burial Scanner	~/.hermes/scripts/project-burial-scan.py

See Telegram Mini App Setup for full scorecard documentation. See Lessons Learned for the process discipline rules that underpin these systems.

Contributing

These guides document real setup sessions — errors included. If you hit something not covered here, open an issue or PR.