SiGit Code: local-first coding agent (github.com)

SiGit Code has been announced as a local-first coding agent for smbCloud, designed to enhance coding efficiency by running entirely on local machines without reliance on API keys or cloud services. The agent operates in two modes: ACP mode, which integrates with ACP-compatible editors like Zed, and an interactive terminal chat mode for direct command line use. Upon first startup, siGit downloads a GGUF model (approximately 1-2 GB) from HuggingFace, enabling quick subsequent launches from local storage. This development is significant for the AI/ML community as it addresses concerns about cloud dependency, data privacy, and latency associated with traditional coding tools. By leveraging local processing, developers can maintain greater control over their data and streamline their workflows. Presently, siGit supports macOS and Linux for both modes, while Windows users can utilize ACP mode only. The tool’s integration with existing setups, including sharing model caches between its desktop app and CLI, further demonstrates its utility in optimizing coding experiences while ensuring performance and accessibility.