Hermes Agent is an open-source AI agent framework by Nous Research. It connects any AI model to your existing tools, orchestrates specialist agents, executes multi-step workflows, and learns from every interaction.
Open-Source AI Agent Framework by Nous Research
Automate. Delegate. Scale.
Overview
Hermes connects any AI model to your existing tools — Slack, Gmail, GitHub, Notion, Jira — and executes multi-step workflows autonomously. From drafting emails to shipping code, it handles the full pipeline without human intervention.
Agents run around the clock on Telegram, Discord, Slack, or API. They never sleep, never forget context, and improve over time through a persistent skills system that learns from every interaction.
Fully self-hosted on your infrastructure. No data leaves your environment. Supports local models, private API keys, and credential pooling across multiple providers for maximum security and cost control.
Industry benchmarks show 171% average ROI from agentic AI deployments. Hermes delivers real cost savings from day one — automating repetitive tasks that consume 10-15 hours per employee per week.
Value
What it actually does for your business — no jargon, no hype
A fully configured Hermes deployment costs a fraction of a single employee. No salary, no benefits, no office space. For the price of a few contractor hours, you get a 24/7 workforce.
Hermes never sleeps, never takes vacation, never calls in sick. Nights, weekends, holidays — your workflows keep running without interruption.
One person with Hermes produces the output of a small team. Research, reporting, outreach, monitoring — in parallel, across multiple specialists. You scale without hiring.
Email, Slack, Google Drive, Notion, Jira. Hermes connects to the tools you already use. No rip-and-replace, no vendor lock-in, no custom integration projects.
No onboarding, no management overhead, no PIPs, no turnover. You don't need to find, interview, train, or retain talent. The agent comes pre-trained.
Tasks that normally take a full day — research reports, competitive analysis, email sequences, data aggregation — are completed in hours. Set it up once, collect results on autopilot.
Real Use Cases
Actual users, actual numbers — pulled from the community
Achieved an end-to-end content operations and client workflow pipeline that runs 24/7. Built Hermes agents for pre-call client research (saving 20-30 min per call), automated podcast digestion converting 10+ hours of weekly listening into a 2-hour workflow, and auto-drafted meeting notes into client follow-ups. Added lead scraping from YC, Twitter, and Reddit, plus daily news briefings delivered automatically to Telegram and Discord.
Delivered over $100,000 in client work value entirely through autonomous Hermes agents — the longest-running documented business automation on the platform with a 297-day continuous streak. Generated 5 billion+ tokens and 900,000+ seconds of compute time running 24/7, replacing an entire client service delivery pipeline without human intervention.
Built a live inventory tracking system for logistics ops using Hermes's persistent memory and subagent parallelization. The system maintains context over time, processes real-time inventory data feeds, and runs multiple parallel agents to monitor stock levels, orders, and supply chain events simultaneously — ideal for e-commerce and warehouse operations.
Moved from guessing Yes/No prices to data-driven prediction market trading. Runs 4 analysis layers in parallel — order book depth, on-chain wallet movements, lag between news events and price action, and position changes. All coordinated through Hermes's Polymarket module and News Skill running simultaneously. The agent is self-improving: it writes strategy notes after every session and adjusts its approach.
Source: hermes-agent.nousresearch.com/docs/user-stories
Pain Points
Employees spend 10-15 hours per week on tasks that could be automated: email triage, report generation, data entry, scheduling, and status updates.
Data scattered across 5+ tools with no single source of truth. Teams waste hours copying information between Slack, email, project trackers, and docs.
Automated workflows run without visibility. No logging, no accountability, no way to trace what happened or why a decision was made.
Companies invest in AI tools but can't measure impact. Generic chatbots add noise, not value. There's no structured way to connect AI to real business processes.
Platform
Hermes Agent is an open-source AI agent framework by Nous Research. It connects any AI model to your existing tools, orchestrates specialist agents, executes multi-step workflows, and learns from every interaction.
01
How It All Works Together
Architecture
Your choice of any AI model. OpenRouter routes requests to the best model for each task. Supports Claude, GPT, Gemini, DeepSeek, Mistral, and local models. Credential pooling rotates across multiple API keys automatically.
The central brain that routes tasks, manages agents, and orchestrates multi-step workflows. Runs on CLI, Telegram, Discord, Slack, WhatsApp, or API. Includes context compression, session management, and security enforcement.
Reusable skill documents that encode procedures, workflows, and domain knowledge. Skills accumulate over time — the agent learns from corrections, discovers new workflows, and persists them for future use.
SQLite-backed task board with queue-aware delivery. Tasks are structured as stateful records (pending → sent → claimed → completed) with full event audit trail. 10 persistent specialist agents work in parallel — no race conditions, no lost tasks, no silent failures.
15 automated workflows run on schedule 24/7 — n8n handles deterministic execution while Hermes provides AI reasoning. Five integration interfaces: REST API (workflow management), MCP Tools (Hermes-triggered execution), Proxy (n8n calls Hermes for AI), Generator (Python SDK), and Install (npm-based). Every workflow follows: Schedule → Data Collection → AI Processing (Hermes via proxy) → Delivery (Telegram).
REST API: Programmatic workflow management
MCP Tools: Hermes triggers workflows as first-class tools
Proxy :8765: n8n calls Hermes /mark_ii for AI reasoning, /telegram for delivery, /codice for code generation
Generator: Python SDK builds workflow JSONs programmatically
Install: npm-based deployment with env var configuration
Architecture
MARK_II orchestrates 10 specialist agents across 3 teams
Orchestrator — routes tasks, manages agents, coordinates workflows
🏗️ Engineering
5 agents
📊 Marketing
2 agents
💰 Revenue
3 agents
| Agent | Role | What They Do |
|---|---|---|
| 👑 MARK_II | CEO Orchestrator | Routes tasks, selects specialists, coordinates workflows |
| 🏗️ Ingegno | Lead Architect | System architecture, technical design, multi-tenant systems |
| 💻 Codice | Senior Engineer | Deep coding, code review, patterns, performance optimization |
| 🔧 Arco | Infrastructure Engineer | DevOps, deployment, monitoring, security hardening |
| ✅ Guardia | QA Manager | Testing strategy, quality standards, acceptance, code review |
| 📝 Memento | Memory Steward | Wiki management, knowledge organization, long-term memory |
| 📋 Stella | Product Strategist | Requirements analysis, UX design, feature opportunities |
| 🚀 Idea | Growth Hacker | User acquisition, retention, marketing, viral loops |
| 💰 Faro | Revenue Architect | Business models, monetization, financial planning, pricing |
| 🔬 Oracolo | Tech Innovator | Emerging tech research, prototyping, feasibility analysis |
| ⚠️ Avaro | Devil's Advocate | Risk analysis, critical thinking, edge cases, threat modeling |
Use Cases
Auto-generate product descriptions from specs, compile inventory reports from multiple sources, draft customer support emails, and monitor competitor pricing — all running on a nightly cron schedule.
AI screens incoming CVs against job requirements, ranks candidates, schedules interviews via calendar integration, and sends personalized status updates to applicants automatically.
Ship MVPs in hours: generate landing pages, write cold outreach sequences, set up automated onboarding flows, and create investor update decks from raw metrics.
Real-time dashboards from Jira/Linear data, automated roadmap updates, weekly stakeholder digests, and feature request triage — all synced to your project management tools.
Automated PR reviews with inline comments, CI/CD monitoring with failure alerts, bug report summaries from GitHub issues, and codebase documentation that stays current.
Weekly newsletters drafted from RSS feeds and trending topics, social media posts scheduled across platforms, SEO briefs generated from keyword research, and content calendars maintained automatically.
Workflow Engine
n8n workflows handle the heavy lifting — scheduled, autonomous, zero human intervention
Hermes agents handle reasoning, coding, and decisions. n8n handles deterministic execution — APIs fire, data flows, schedules run. Together they form a complete autonomous operations layer.
5 integration interfaces connect them:
REST API (management) · MCP Tools (Hermes → n8n) · Proxy :8765 (n8n → Hermes) · Generator (Python SDK) · Install (npm)
Every workflow follows the same pipeline:
Schedule → Data Collection (SerpAPI) → AI Processing (Hermes via proxy) → Automated QA Gate → Delivery (your preferred channel)
Content is independently reviewed by quality assurance and risk analysis specialists — checking for accuracy, formatting integrity, bias, and brand risk — before any delivery occurs. Both must approve. Zero human intervention in the QA loop.
| Workflow | Tier |
|---|---|
| All Tiers | |
| 🩺 System Status Monitor | All |
| 🔧 Workflow Updater | All |
| 🔬 Weekly Model Scanner | All |
| Professional & Enterprise | |
| 📰 Smart AI Newsletter | Pro + Ent |
| 🌐 Website Uptime Monitor | Pro + Ent |
| 🛡️ VPS Security Monitor | Pro + Ent |
| 🔍 Competitor Analyzer | Pro + Ent |
| Enterprise (includes all above) | |
| 📞 Pre-Call Research Agent | Enterprise |
| 📧 Smart Email Manager | Enterprise |
| 💼 LinkedIn Auto-Post | Enterprise |
| 🔎 SEO Automation | Enterprise |
| 📹 UGC Ad Pipeline | Enterprise |
| 📦 13 On-Demand Tools | Enterprise |
Workflow Engine
16-node pipeline: SerpAPI search → Hermes editorial AI → your preferred channel — fully autonomous
Cron-based scheduling kicks off each workflow at its configured time. The Smart Newsletter fires daily at 9AM — fully autonomous editorial pipeline. System Status runs every 30 min, Website Monitor every hour. Zero human intervention.
Workflows call SerpAPI (DuckDuckGo News) directly to gather current data — AI news, competitor intel, SEO keywords. n8n credentials handle API key rotation. Tavily MCP serves as backup for Hermes-triggered searches from the CLI side.
Raw data is sent to Hermes via the n8n proxy (:8765/mark_ii). Hermes makes editorial decisions (is this news relevant?), then formats the selected articles. Two separate AI calls: one for filtering, one for formatting — both using your chosen model. The proxy supports per-workflow model selection for cost optimization.
Before delivery, every AI-generated piece passes through a dual quality gate: one specialist checks for accuracy and factual errors, another attacks it from an adversarial perspective. If either finds issues, the content is automatically regenerated with the feedback incorporated — up to 3 attempts. Only verified output reaches your channels. Zero human review needed.
Formatted content is routed to your preferred channel — Telegram, Slack, Discord, or any supported platform. Newsletters, ops reports, competitor analysis, and pre-call briefings each arrive in the right place — with clean formatting, no broken tags. Delivery is instant and native to the platform.
The n8n-First Architecture
n8n is the DEFAULT execution engine. Every new automation starts with "Can n8n do this?" n8n handles pipelines; Hermes provides AI reasoning at decision points. Together: tiered workloads, 24/7 uptime, zero human intervention.
03
Persistent, structured, agent-maintained
Memory
Your agent never forgets — wiki-primary architecture with multi-sync
The wiki lives in a local git repository with full version history, rollback capability, and commit discipline. Every change is tracked — who changed what, when, and why. A post-commit hook mirrors to iCloud for mobile read access, but git is the source of truth.
Seven native MCP tools give agents structured access to the wiki: search across all pages, retrieve full page content with backlinks, navigate the knowledge graph via wikilinks, list pages by type and confidence, and get stats on orphans and broken links. No filesystem reads — agents query the wiki like a database.
Short-term context lives in Hermes session memory and session_search. Reusable workflows are saved as Skills (SKILL.md documents) that the agent loads automatically when relevant tasks appear.
A daily cron job (02:00 UTC) runs Memento — the Memory Steward — to perform gap analysis between Hermes memory and the wiki. New discoveries are auto-documented, stale entries flagged, and the knowledge base stays current.
Every codebase is mapped into a queryable knowledge graph. Agents query "how does X connect to Y?" in ~300 tokens instead of reading thousands of lines of source code — 50-150x token savings. Graphs are rebuilt nightly, with AST-powered incremental updates and community detection to reveal hidden code structure.
The Wiki
100+ pages, 9 directories, self-maintained by the agent squad
The wiki lives in a local git repository — full version history, commit tracking, and rollback capability. A post-commit hook mirrors to iCloud for Obsidian Mobile read access. Every file is plain Markdown — no lock-in, no proprietary format, fully portable. The wiki spans 100+ pages across 9 directories: entities, concepts, operations, projects, plans, queries, evaluations, raw research, and skills — each with structured frontmatter, confidence scores, sources, and last-evaluated dates.
Agents access the wiki through seven native MCP tools — wiki_search, wiki_get_page, wiki_get_related, wiki_get_backlinks, wiki_list_pages, wiki_get_index, and wiki_get_stats — treating the knowledge base like a queryable database rather than flat files. Memento (Memory Steward) maintains it, adding cross-references, pruning stale entries, and validating facts. A pre-commit lint hook catches broken wikilinks, orphan pages, and missing frontmatter before anything is committed. The wiki is the sole long-term memory — MemPalace was fully removed.
A live walkthrough of the Obsidian wiki vault — 100+ interconnected Markdown pages, graph view, MCP-powered search, and automated maintenance by Memento.
02
Your data is protected at every layer
Reliability
Your system is protected — here is how
Automated backup every 3AM — full ~/.hermes/ state, skills, profiles, wrappers, sessions stored as compressed archives as compressed git bundles.
Rolling 7-day backup window ensures you can recover from any incident within the past week — even if corruption went unnoticed.
One-command restoration from git backup with restore.sh: verify integrity, extract archives, replace ~/.hermes/. Full operational recovery in under 20 minutes.
The Obsidian wiki as compressed git bundles is completely outside the ~/.hermes/ backup scope — survives even total agent loss. Knowledge is never lost.
3-layer defense: daily encrypted backups + WSL VM export + wiki as git repo. Integrity checks every Sunday.
Extensibility
Your agent grows smarter with every task
Hermes saves reusable procedures as skills — build a growing library of expertise that loads automatically into every session. Skills accumulate over time, making the agent better at your specific stack.
Pin critical skills so they're always active — no need to load them manually. The curator manages lifecycle: create, update, delete, and organize skills into categories.
When Hermes solves a complex problem or gets corrected, it persists that knowledge as a skill. The system continuously refines itself through real usage — not just static documentation.
Browse, search, and install from a growing public registry. Community-contributed skills for everything from DevOps to data science — or publish your own with a single command.
Infrastructure
No cloud dependency, no shared infrastructure, no data leakage
Hermes installs on your local machine (macOS, Linux, or Windows/WSL). No cloud servers, no shared tenancies. Your data never leaves your hardware. You control the environment, the model, and the API keys.
The wiki, configuration, and deployment files live in version-controlled git repositories. Every change is tracked with commit history, authorship, and rollback capability. No mysterious config drift — git diff shows exactly what changed.
API keys stay in your .env file — never transmitted, never stored by us. You choose your LLM provider (OpenRouter, Anthropic, OpenAI, DeepSeek, or local models). We configure the agent; you bring the credentials.
Hermes Gateway connects to Telegram, Discord, Slack, WhatsApp, and 10+ platforms via their official APIs. Messages are encrypted in transit by each platform's native security. No man-in-the-middle, no proxy that sees your conversations.
03
Choose the right plan for your needs
Pricing
| Feature | 🚀 Starter | ⭐ Professional | 👑 Enterprise |
|---|---|---|---|
| Core Platform — Identical Across All Tiers | |||
| AI Agent Setup (Hermes / Claude Code / OpenClaw) | ✅ | ✅ | ✅ |
| 10-Specialist Architecture (Phase 3 delegation) | ✅ | ✅ | ✅ |
| Complete Skill Library (600+ skills hub) | ✅ | ✅ | ✅ |
| Skill Factory (auto-learns from your usage) | ✅ | ✅ | ✅ |
| LLM Wiki System (self-building knowledge base) | ✅ | ✅ | ✅ |
| n8n Execution Engine (workflow automation) | ✅ | ✅ | ✅ |
| Git-Backed Everything (version control, rollback) | ✅ | ✅ | ✅ |
| BYOS Storage (Google Drive, OneDrive, iCloud) | ✅ | ✅ | ✅ |
| Integrations | |||
| Communication Channels | 1 (CLI or Telegram) | Up to 3 platforms | Unlimited |
| Workflow Library | |||
| System Status Monitor | ✅ | ✅ | ✅ |
| Workflow Updater (self-improving) | ✅ | ✅ | ✅ |
| Model Scanner (weekly discovery) | ✅ | ✅ | ✅ |
| Smart Newsletter (AI-gated, 2x daily) | — | ✅ | ✅ |
| Website Uptime Monitor (Telegram + Email) | — | ✅ | ✅ |
| VPS Security Monitor (SSH threat detection) | — | ✅ | ✅ |
| Competitor Analyzer (32-node pipeline) | — | ✅ | ✅ |
| LinkedIn Auto-Post (AI content → social) | — | — | ✅ |
| Smart Email Manager (AI classify + auto-draft) | — | — | ✅ |
| Pre-Call Research Agent (auto-briefs) | — | — | ✅ |
| SEO Automation | — | — | ✅ |
| UGC Ad Pipeline (product → AI video ads) | — | — | ✅ |
| 13 On-Demand Tool Workflows | — | — | ✅ |
| Support & Documentation | |||
| Priority Support | 1 week | 14 days + weekly check-ins | 30 days + quarterly check-ins |
| Training | — | 1-hour live session + guide | 1-hour live session + guide |
| Documentation | Quick-start README | README + prompting guide | README + guide + ops playbook + runbook |
| Backups & Security | |||
| Daily Backups | 30-day retention | 1-year retention | 7-year retention (encrypted) |
| Disaster Recovery | ✅ | ✅ | ✅ |
| Infrastructure & Memory | |||
| Redis Context Engine (auto-generated MCP tools, real-time enterprise DB sync, dual-layer agent memory) | — | — | ✅ |
| One-Time Setup | €1,200 | €2,200 | €3,500 |
Contact
1. Discovery Call We discuss your workflows, pain points, and automation goals. You choose the right package. 2. Setup & Configure We install Hermes Agent, connect your tools, configure specialist agents, and train on your data. 3. Go Live Your AI workforce starts delivering results immediately. We monitor, tune, and support you through the first weeks.
Pick how you'd like to start — most leads prefer async (no calendar, 2-min form, reply within 24h). If you're ready to talk now, book a call instead.
Async: 2-min form, no calendar, reply within 24h. Call: live 30-min discovery session.