lmcli v0.7.0 – TUI harness with smooth performance up to 1M context (codeberg.org)

The recent release of lmcli v0.7.0 introduces significant enhancements to this versatile large language model (LLM) harness, notably boasting seamless performance with support for up to 1 million tokens using sliding-window rendering. This update enhances user experience by enabling smoother interaction in coding, chatting, and custom workflow creation. Key features include support for composable nesting agents, custom tools, and sandboxed execution, alongside compatibility with OpenAI and Anthropic APIs. The tool's lightweight design, requiring less than 50MB of RAM per instance, makes it accessible for broader use in various applications. For the AI/ML community, lmcli’s modular architecture unlocks new possibilities for constructing complex conversational agents that can delegate tasks effectively through a structured agent system. Users can now create sophisticated tools that interact with external APIs, manage branching conversations with SQLite, and seamlessly export interactions to different formats. Future updates promise to expand capabilities further, including enhancements in customization and tool functionality, thereby solidifying lmcli's role as a significant facilitator of LLM integration into diverse workflows.