Small Local LLM Recommendations for Tool Calling on Low-End Hardware

A user in the LocalLLaMA community requested recommendations for lightweight local language models capable of tool calling and agentic tasks on low-end hardware, specifically a GTX 750 Ti GPU with 4GB VRAM and 16GB RAM. Running agentic workloads locally on entry-level hardware makes AI development accessible to hobbyists without relying on expensive GPUs or cloud APIs. However, smaller models often struggle with complex reasoning and strict format adherence required for reliable tool execution. Hardware constraints like 4GB VRAM limit selection to small, quantized models (such as 1B-3B parameter models or heavily quantized 7B/8B models partially offloaded to system RAM). A major trade-off of small models is reduced instruction-following precision and a higher rate of tool calling syntax errors compared to larger models.