Hermes Agent adds self-learning memory, MCP server mode

Hermes Agent is shifting the "local LLM" narrative from stateless chat to persistent autonomous operations through a clever mix of Markdown-based active memory and SQLite archives.

• –Self-learning via "Skills" (structured Markdown docs) allows the agent to codify successful workflows and reuse them, effectively building its own procedural memory.
• –Dual-mode MCP support is a game-changer: it can use external tools (Client) or expose its own memory and skills to other editors (Server).
• –The tiered memory architecture (Active MEMORY.md vs. Archive SQLite) solves the context window bloat problem while maintaining long-term recall.
• –Integration with Honcho for user modeling ensures the agent adapts to individual developer styles over time rather than resetting every session.
• –Its model-agnostic nature prevents vendor lock-in, supporting everything from local Ollama instances to high-end Anthropic/OpenAI APIs.