Most ways of talking to a language model are either locked to one vendor's SDK or trapped in a chat window that forgets everything the moment you close the tab. This tool makes the opposite bet: every model — GPT, Claude, Gemini, or a Llama running on your laptop — becomes an interchangeable Unix command, and every prompt and response lands in a SQLite database you can later query, pipe, and audit.
What Sets It Apart
- Provider-agnostic by design. Swapping
gpt-4ofor a hosted Claude model or a local one is a single flag, so prompts and scripts outlive any one vendor's API. - Everything is logged to SQLite. Each interaction is searchable and replayable, turning throwaway prompting into an auditable record you can mine weeks later.
- A plugin ecosystem, not a monolith. Community plugins add new model providers, embedding models, tools, and fragment loaders, keeping the core small while coverage keeps expanding.
- Structured output and tool calling are first-class. JSON-schema extraction and function calling let it act as a data-extraction or agent primitive inside shell pipelines, not just an interactive chat.
Who Should Use It
Great fit if you live in the terminal, want reproducible and auditable LLM calls, or need to wire models into shell scripts, cron jobs, and embedding workflows. Look elsewhere if you want a polished graphical chat interface or a managed agent platform — this is deliberately a composable building block, and getting the most from it means embracing the command line and SQLite over a point-and-click UI.
