EdgeChains
A full-stack generative AI library built on Hono and Jsonnet, treating prompt and chain management as a configuration-management problem rather than a UI or library design problem.
EdgeChains writes prompts and chains declaratively in Jsonnet, the configuration language Google built for managing its Borg infrastructure, rather than hiding them inside classes or a fragile custom DSL, which makes prompts versionable and diffable like any other configuration. It is built on Honojs and Jsonnet rather than a bespoke runtime, so a production application is one script file plus one Jsonnet file.
It automatically parallelizes LLM chains and chain-of-thought tasks across CPUs, GPUs and TPUs using a WebAssembly runtime, retries and backs off when some requests in a chain fail, and is designed to scale across large numbers of APIs, prompt lengths and vector datasets. The project covers prompt-engineering patterns such as ReAct-style and HyDE prompt chains.