Insights on local agentic AI, model training, and open-source development
A source code analysis of the three-layer defense system hidden in Claude Code — fake tool injection, connector text summarization, and streamlined output transforms — designed to prevent model distillation from API recordings.
A source code analysis of Claude Code's tool system — how tools are defined with Zod schemas, sent via the Anthropic Messages API, streamed back as incremental JSON, and executed locally with permission control.
Introducing LocoTrainer-4B, a 4B-parameter specialized code analysis agent distilled from Qwen3-Coder-Next. Learn how we combined multi-turn tool-calling with deep MS-SWIFT framework knowledge to create a local agent that generates comprehensive codebase analysis reports.
Announcing LocoOperator, a specialized framework for training and deploying local agentic AI models. Discover how we built an agent loop with tool calling, planning, and execution capabilities that runs entirely on your hardware.