OpenSquilla is a Python 3.12+ microkernel runtime for AI agents that emphasizes token efficiency through smart routing across multiple model tiers, adaptive prompts, and on-demand skills, claiming 60–80% token cost savings.
It provides a four-tier memory system (working, episodic, semantic, raw) with local embeddings, hybrid search, temporal decay, and daily consolidation, plus a three-tier security sandbox with OS-level isolation and multiple channel integrations (terminal, web, Slack, Discord, Telegram, Teams, and others). The system is modular and plugin-based, with built-in gateway, tools, scheduler, and multi-LLM support, installable via Git, Git LFS, and a bundled installer script.