How to use an AI agent that builds skills from experience in messengers and terminals
Hermes Agent is an AI agent by Nous Research that embeds a learning loop to extract and improve skills from user experience. It allows conversation not only in local terminals but also in messengers like Telegram, Discord, and Slack, and it remembers user characteristics even after sessions end. It runs in various environments, from $5 VPS to GPU clusters and serverless infrastructure.
A learning loop that turns experience into skills
The agent automatically generates skills after performing complex tasks and improves them during use. It prompts knowledge storage through periodic notifications and uses FTS5-based search and LLM summarization to find past conversation records, maintaining context across sessions. It understands user characteristics through Honcho-style user modeling, a process compatible with the agentskills.io open standard.
Interfaces covering terminals and messengers
Users can communicate with the agent via terminal UI or messenger apps. The terminal supports multi-line editing, slash command autocomplete, conversation history, task interruption and reassignment, and streaming tool output. All platforms, including Telegram, Discord, Slack, WhatsApp, Signal, and CLI, are handled by a single gateway process, which also supports voice memo transcription and cross-platform conversation continuity. You can start the terminal with the hermes command or run hermes gateway to chat in messengers.
Infrastructure support for running anywhere
Daytona and Modal provide serverless persistence, putting environments into a dormant state when idle and waking them when needed. This results in minimal cost between sessions, allowing execution on both $5 VPS and GPU clusters. You are not tied to a laptop and can issue commands via Telegram while working on a cloud VM.
Model selection freedom and automation
You can freely choose your desired model from Nous Portal, OpenRouter, OpenAI, or self-hosted endpoints. The built-in cron scheduler allows setting regular tasks like daily reports, nightly backups, and weekly audits in natural language. It creates isolated sub-agents to handle parallel tasks and writes Python scripts that call tools via RPC to efficiently process multi-step pipelines.
Pre-installation checks and license
Installation is available via bash scripts on Linux, macOS, and WSL2 environments, and via PowerShell scripts on Windows. An APT repository for Android Termux is also provided. Windows Defender or antivirus software may misidentify uv.exe as malware; in such cases, verify the file's authenticity using GitHub CLI or add the folder to the exception list. It is recommended to check the cost structure for serverless backends and API key requirements for specific messenger integrations in advance.
