Nous Research Launches Hermes Agent v0.5.0: A Self-Learning Open-Source AI Agent

Nous Research has announced the release of Hermes Agent v0.5.0, a fully open-source autonomous AI agent licensed under MIT that learns and improves with every task it executes. The project has garnered over 15,700 stars on GitHub since its initial launch in February 2026.

What Makes Hermes Agent Different?

Unlike traditional coding assistants or chatbot wrappers, Hermes Agent operates as a fully autonomous agent that lives on your own server. It remembers what it learns, builds new skills from experience, and becomes more capable the longer it runs.

Key Features:

• Self-learning loop: The agent creates reusable skills from every task it performs and automatically refines them over time
• Persistent memory: Retains conversation and project context across sessions, building a deepening model of each user
• Support for 400+ models: Via Nous Portal, OpenRouter, Ollama, and other providers
• Multi-platform gateway: Works across Telegram, Discord, Slack, WhatsApp, Signal, and email
• 40+ built-in skills: Covering MLOps, GitHub workflows, research, and more

What's New in v0.5.0?

Nous Research describes this as "the hardening release," focusing on security, stability, and expansion:

• Hugging Face integration: Full support for HF Inference API with a curated agentic model picker
• Telegram private chat topics: Project-based conversations with per-topic skill binding for isolated workflows
• Plugin lifecycle hooks: Including pre_llm_call, post_llm_call, on_session_start, and on_session_end
• OpenAI reliability improvements: Added GPT_TOOL_USE_GUIDANCE to prevent GPT models from describing actions instead of executing them
• Anthropic output limits: Replaced the hardcoded 16K max_tokens with per-model native limits, up to 128K for Opus 4.6
• Nix Flake support: Full uv2nix build with NixOS module and persistent container mode

Critical Security Updates

The release placed special emphasis on supply chain security. The compromised litellm dependency was removed, all dependency versions were pinned with regenerated uv.lock hashes, and a new CI workflow was added to scan for supply chain attack patterns.

Why This Matters

Hermes Agent represents a shift in AI agents from simple assistants to autonomous digital workers capable of learning and evolving. It can run on a VPS costing as little as $5 per month, making intelligent agent technology accessible to independent developers and small teams.

Being fully open-source also means data never leaves the user's servers, a significant advantage for organizations concerned about data privacy and compliance.

What's Next?

As competition heats up in the AI agent market, Hermes Agent positions itself as a strong open-source alternative focused on continuous learning and privacy. The project supports an active community of contributors and provides the agentskills.io platform for sharing skills between users.

Source: Nous Research - GitHub