Hermes Agent is a self-improving AI agent framework built by Nous Research, written in Python and designed to operate across multiple platforms and deployment environments. The project is positioned as an agent that grows with users through a built-in learning loop, distinguishing itself from static AI assistants by creating skills from experience, improving them during use, and building persistent knowledge across sessions.
The framework supports flexible model selection across multiple providers including Nous Portal, OpenRouter, OpenAI, and custom endpoints, with the ability to switch models via simple commands without code changes or vendor lock-in. Users can deploy Hermes on various infrastructure options ranging from a $5 VPS to GPU clusters or serverless platforms like Modal and Daytona, where the agent's environment hibernates when idle and activates on demand, minimizing operational costs.
Hermes provides multiple interface options for interaction. The primary interface is a full-featured terminal UI with multiline editing, slash-command autocomplete, conversation history, interrupt-and-redirect capabilities, and streaming tool output. Beyond the CLI, the agent operates as a gateway supporting Telegram, Discord, Slack, WhatsApp, Signal, and email, enabling users to interact with a single agent instance across multiple platforms while maintaining conversation continuity. The framework includes voice memo transcription capabilities for cross-platform communication.
The core learning mechanism centers on a closed learning loop that includes agent-curated memory with periodic nudges, autonomous skill creation following complex tasks, and self-improving skills during use. The system implements FTS5 session search with LLM summarization for cross-session recall and uses Honcho dialectic user modeling to develop deepening models of individual users across sessions. Skills created by the agent are compatible with the agentskills.io open standard.
The framework includes built-in scheduling capabilities with a cron scheduler that delivers automations to any platform, enabling unattended execution of tasks like daily reports, nightly backups, and weekly audits specified in natural language. Hermes supports delegation and parallelization through isolated subagents for parallel workstreams and Python scripts that call tools via RPC, collapsing multi-step pipelines into single turns.
Deployment flexibility extends to six terminal backends including local, Docker, SSH, Singularity, Modal, and Daytona, with serverless persistence options available through Daytona and Modal. The installation process is streamlined across platforms, with native Windows support through PowerShell installation that bundles Python 3.11, Node.js, ripgrep, ffmpeg, and a portable Git Bash, alongside Linux, macOS, WSL2, and Termux support.
According to GitGenius tracking data, the repository has demonstrated steady growth with 11 additional stargazers and 6 new forks since July 2026, reaching 209,181 stars and 38,165 forks. The project maintains active issue and PR management with a median response latency of 0.2 hours across nearly 10,000 tracked items. The most active issue categories are type/bug with 5,019 items, followed by P3 and P2 priority classifications. Primary contributors include alt-glitch with 44,895 tracked events, teknium1 with 6,861 events, and konsisumer with 737 events. The repository shares overlapping contributors with anthropics/claude-code and openclaw/openclaw projects.
