agentic-os-org/ANOLISA

What it solves

ANOLISA is a server-side operating layer designed to optimize and secure AI Agent workloads. It addresses three primary constraints: the lack of an AI-native terminal interface, high token costs associated with tool responses, and the need for secure, isolated execution environments for agents.

How it works

ANOLISA operates as a layer between the agent framework and the system, providing several modular components:

• Agent Entry: Provides cosh-ng, an AI-native Linux terminal that integrates natural language and shell commands into a single interface, along with system and DevOps expertise via OS Skills.
• Context Efficiency: Uses Token-less to compress tool schemas and responses by removing redundancy before they reach the LLM, and AgentSight to provide visibility into token spend using eBPF on Linux.
• Runtime & Security: Implements Agent Sec Core for isolating risky operations and ws-ckpt for workspace checkpointing and rollback, ensuring agents have a boundary and recovery points.

Who it’s for

Developers and DevOps engineers who run AI agents in terminal-based environments and want to reduce LLM operational costs, improve observability, and secure their system execution.

Highlights

• Token Reduction: The Token-less component can significantly reduce token usage in tool responses and schemas (e.g., saving up to 62.9% in some pipelines).
• AI-Native Terminal: cosh-ng allows agents to drive the system directly from the terminal with approval-based risky work.
• Kernel-Level Observability: AgentSight uses eBPF to trace agent execution and token use without requiring code changes to the agent.
• Crossover Compatibility: Works with existing agent frameworks and sandboxes without requiring framework code changes.