As modern networks continue to scale and become more complex, manual configuration processes become increasingly inefficient and error-prone. Although intent-driven self-configuration using large language models has demonstrated considerable potential, these models remain computationally intensive and resource-hungry, and they often raise privacy concerns due to their reliance on external cloud infrastructure. This work presents SLM_netconfig, a fine-tuned, small-language-model–centered framework that adopts an agent-based architecture and employs parameter-efficient adaptation techniques to translate configuration intents—expressed as natural-language requirements or questions—into syntactically and semantically correct network configurations. The system is trained using a domain-specific dataset generated through a dedicated pipeline grounded in vendor documentation, ensuring strong alignment with real-world configuration practices. Extensive evaluation demonstrates that SLM_netconfig, when using its question configuration model, achieves higher syntactic and goal accuracies than LLM-NetCFG, GPT4o Few Shot, GPT4o Zero Shot, Llama 3.1 Few Shot, and Llama 3.1 Zero Shot, while significantly reducing translation latency and producing concise, easily interpretable configurations. These results indicate that fine-tuned small language models, as embodied in SLM_netconfig, can provide efficient, accurate, vendor-agnostic, and privacy-preserving automated configuration generation entirely on-premise, establishing them as a practical and scalable solution for modern autonomous network configuration.