--- title: "Deploy Hermes — Self-Hosted AI Agent with Persistent Memory [Sep'26]" description: "Self-host an AI agent with persistent memory. No per-call fees." category: "Automation" url: https://railway.com/deploy/ai-agent-persistent-memory --- # Deploy Hermes — Self-Hosted AI Agent with Persistent Memory [Sep'26] Self-host an AI agent with persistent memory. No per-call fees. **[Deploy Hermes — Self-Hosted AI Agent with Persistent Memory [Sep'26] on Railway](https://railway.com/template/ai-agent-persistent-memory)** Machine-readable deploy manifest (JSON, validated by TemplateCI): https://railway.com/deploy/ai-agent-persistent-memory/manifest.json - **Creator:** SilverBanana - **Category:** Automation - **Total deploys:** 9 ## Template content ### Hermes Agent https://res.cloudinary.com/dh2nt6hgh/image/upload/v1777723693/image_9_umztwp.png - **Source:** Shinyduo/hermes-agent - **Public domain:** Yes ## Documentation # Deploy and Host Hermes AI Agent on Railway ![Hermes AI Agent dashboard](https://res.cloudinary.com/asset-cloudinary/image/upload/v1777189955/c6e0f435-b614-4c09-b71e-ebdf42f172c3.png) Hermes is an open-source AI agent framework with persistent memory, multi-step tool use, and a fully LLM-agnostic design — connect OpenAI, Anthropic, Groq, Ollama, or any OpenAI-compatible endpoint without changing a line of agent logic. This template deploys Hermes as a production-ready agent service on Railway in under 5 minutes, with PostgreSQL-backed memory and Redis session handling pre-configured out of the box. Self-host for ~$5–10/month flat — bring your own LLM key and pay only for the tokens you use, with every conversation and memory stored on infrastructure you own. --- ## What This Template Deploys | Service | Purpose | |---|---| | **Hermes Agent** | Core AI agent runtime — LLM routing, tool execution, session management, and API serving on port `3000` | | **PostgreSQL** | Persistent memory — conversation history, session state, and agent context survive restarts and redeploys | | **Redis** | Session cache — concurrent conversation handling and high-throughput request management | All services connect over Railway's private network. Credentials are injected automatically via reference variables — no manual connection string configuration required. --- ## About Hosting Hermes Running a production AI agent requires a persistent runtime with database-backed memory, secure credential management for LLM API keys, and a public HTTPS endpoint for integrations and webhooks. Without a managed host, you're configuring Docker, inter-service networking, SSL, and environment variables manually. Railway provisions all of it automatically. Hermes runs as an always-on agent service — persistent memory, secret management, and HTTPS handled out of the box. The agent's memory and session state survive every redeploy. Typical cost: **~$5–10/month** on Railway's Hobby plan for the full three-service stack. The OpenAI Assistants API charges per tool call and per session-second on top of token costs. Hermes gives you unlimited agent sessions at flat compute pricing — you pay only your own LLM token usage. --- ## Deploy in Under 5 Minutes 1. Click **Deploy on Railway** and wait for the build to complete (~3–5 minutes) 2. Set `LLM_API_KEY` and `LLM_PROVIDER` in the Variables tab 3. `DATABASE_URL` and `REDIS_URL` are auto-injected — Hermes initializes its memory schema on first boot automatically 4. Open your Railway-assigned URL — Hermes is live and ready to handle requests No SSH. No Docker configuration. No reverse proxy setup. --- ## Common Use Cases - **Self-hosted alternative to OpenAI Assistants API** — bring your own LLM provider, pay flat Railway compute instead of per-tool-call fees, and keep conversation history in your own PostgreSQL instance - **Self-hosted alternative to AutoGPT** — run a persistent, memory-backed AI agent without AutoGPT's local Python setup or Docker dependency management - **Multi-channel AI agent via Telegram and WhatsApp** — connect Hermes to Telegram bots or WhatsApp via Evolution API; the agent maintains per-user conversation memory across sessions - **AI agent with persistent memory across sessions** — unlike stateless LLM API calls, Hermes stores full conversation context in PostgreSQL; users continue where they left off after hours or days without re-explaining context - **LLM-agnostic agent infrastructure** — swap between OpenAI, Anthropic, Groq, or local Ollama models without rewriting agent logic; useful for cost optimization or provider fallback - **Private AI assistant with zero data leakage** — all conversations, tool outputs, and memory stay on your Railway PostgreSQL instance; nothing sent to third-party analytics --- ## Configuration | Variable | Required | Description | |---|---|---| | `LLM_API_KEY` | ✅ Required | API key for your chosen LLM provider | | `LLM_PROVIDER` | ✅ Required | Provider — e.g. `openai`, `anthropic`, `groq` | | `AGENT_NAME` | Optional | Agent display name across sessions — defaults to `Hermes` | | `DATABASE_URL` | ✅ Auto-injected | PostgreSQL connection string via Railway reference variable | | `REDIS_URL` | ✅ Auto-injected | Redis URI via Railway reference variable | | `LLM_BASE_URL` | Optional | OpenAI-compatible endpoint — Ollama, vLLM, or OpenRouter | | `PORT` | Pre-set | `3000` — auto-configured by Railway | | `NODE_ENV` | Pre-set | `production` | --- ## Hermes vs. Alternatives | | Hermes (self-hosted) | AutoGPT | CrewAI | LangGraph | |---|---|---|---|---| | **Persistent memory** | ✅ PostgreSQL + Redis | ⚠️ File-based | ❌ Session only | ⚠️ Manual setup | | **One-click cloud deploy** | ✅ Railway | ❌ Local only | ❌ Local only | ❌ Local only | | **LLM agnostic** | ✅ Any provider | ✅ Yes | ✅ Yes | ✅ Yes | | **Self-hosted and private** | ✅ Yes | ✅ Yes | ✅ Yes | ✅ Yes | | **Production HTTPS** | ✅ Auto via Railway | ❌ Manual | ❌ Manual | ❌ Manual | | **Multi-step tool chaining** | ✅ Yes | ✅ Yes | ✅ Yes | ✅ Yes | | **Setup complexity** | ✅ Low — browser config | ❌ High — local Docker | ❌ Medium — Python env | ❌ High — custom code | --- ## Dependencies for Hermes Hosting - API key from at least one LLM provider (Anthropic Claude or OpenAI recommended) - Railway account — Hobby plan (~$5–10/month) covers all three services - Optional: Ollama or vLLM endpoint for zero-API-cost local model inference ### Deployment Dependencies - [Hermes GitHub Repository](https://github.com/sahilrupani/hermes-agent) — source and config - [Railway Documentation](https://docs.railway.com) — scaling and secrets guides - [OpenAI API Documentation](https://platform.openai.com/docs) — LLM backend integration - [Railway Volumes Documentation](https://docs.railway.com/volumes) — optional persistent storage ### Implementation Details ![Hermes AI agent interface](https://res.cloudinary.com/asset-cloudinary/image/upload/v1774992146/hermes-home-page_hnm12e.png) This template deploys the Hermes agent runtime alongside Railway-managed PostgreSQL and Redis over private internal networking. Database and cache ports are never exposed publicly. On first boot, Hermes auto-detects `DATABASE_URL` and `REDIS_URL` and initializes its memory schema. All session state and conversation history persist across restarts. The agent API is served at your Railway public HTTPS domain on port `3000`. --- ## Frequently Asked Questions **How much does Hermes cost compared to OpenAI Assistants?** Hermes on Railway runs at ~$5–10/month flat for the full three-service stack. The OpenAI Assistants API charges per tool call and per session-second on top of token costs — at moderate usage that compounds quickly. Hermes gives you unlimited agent sessions at flat compute pricing plus your own LLM token usage. **What makes Hermes different from AutoGPT or CrewAI?** AutoGPT and CrewAI are primarily local frameworks — they run on your machine and require Python environment management, Docker setup, and manual configuration to expose publicly. Hermes is built for cloud deployment: persistent PostgreSQL memory, Redis session handling, and HTTPS are pre-configured. You get a production agent service, not a local script. **Does Hermes retain memory between conversations?** Yes. Session state and conversation history are stored in PostgreSQL. Hermes retrieves prior context on every new conversation from a returning user — no re-introduction needed, no context lost between sessions, restarts, or redeploys. **Can I switch LLM providers without changing agent logic?** Yes. Update `LLM_PROVIDER` and `LLM_API_KEY` in Railway Variables and redeploy. Agent logic, tool definitions, and memory are fully decoupled from the provider layer. **Can I run Hermes with a local model instead of a cloud API?** Yes. Set `LLM_BASE_URL` to any OpenAI-compatible endpoint — Ollama, vLLM, or LM Studio. This enables fully private deployments where no conversation data leaves your Railway environment. **Is my data private?** Yes. Hermes is self-hosted on Railway. Conversations, tool outputs, and agent memory stay in your PostgreSQL instance. API keys are stored as encrypted Railway environment variables and are never logged or sent to third-party services. --- ## Why Deploy and Host Hermes on Railway? Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it. By deploying Hermes on Railway, you get a production-ready AI agent with persistent memory, LLM-provider flexibility, automatic HTTPS, and zero server administration — at ~$5–10/month flat with full ownership of your conversations and agent memory. ## Similar templates - [N8N Main + Worker](https://railway.com/deploy/n8n-main-worker) — Deploy and Host N8N with Inactive worker. - [Evolution API with n8n](https://railway.com/deploy/evolution-api-with-n8n) — Automate WhatsApp workflows with Evolution API, n8n, and Postgres. - [Postgres Backup](https://railway.com/deploy/postgres-s3-backups) — Cron-based PostgreSQL backup to bucket storage Open this page in a browser: https://railway.com/deploy/ai-agent-persistent-memory