context-artist

Location: Philadelphia, USA Remote: Yes Willing to relocate: Yes Résumé/CV: https://louisfoster.com LinkedIn: https://linkedin.com/in/louis-foster-software Email: contact@louisfoster.com Over ten years as an engineer and team lead at startups. Product-minded, love to code (with and without LLMs), and willing to lead. Things I'm excited about: custom virtual assistants as UI, home automation, mesh networking. Ollama context engine with streaming API: https://github.com/ohmstone/context-artist More here: https://github.com/ohmstone

I needed a better way to create and compare prompts when using local LLMs (e.g. via Ollama) in a workflow. You either get huge harnesses that do everything, stuff that's mostly designed for cloud LLMs, conversational chat UIs, or text inlined with code (i.e. what you see with most LLM docs). My approach to this is to instead define an Objective (what you want to achieve in your task) which contain a Draft and committed Versions of context (system prompt, and optional example user/assistant turns), alongside test Runs (reusable test input). All managed via a clean REST API, and with a granular subscription model over WebSockets for streaming run output. I'm building a local-first natural language UI system for some various technical, social, and creative projects. This tool seems to fit nicely between a UI agent and various service APIs.