EdgeChains
EdgeChains is a new Language & Grammar for production-friendly Generative AI. Based on Jsonnet & works everywhere (java, python, js,etc). Prompts live declaratively & "outside code". Easy to Reason/Test/Deploy.
Free8,358 impressions#173 of its week
EdgeChains is a declarative config framework for production-ready generative AI applications using Jsonnet.
- for
- Developers building and deploying LLM chains and prompt workflows.
- pricing
- open source
- license
- AGPL-3.0
Key features
- Single-file deployment — Run a production-ready GenAI app with just one script and one Jsonnet file.
- Versionable prompts — Prompts are written in Jsonnet, making them easy to diff and version control.
- Automatic parallelism — Parallelizes LLM chains across CPUs, GPUs and TPUs via WebAssembly.
- Fault tolerance — Retries and backs off on failed requests to keep the workflow running.
- Scalability — Handles large numbers of APIs, long prompts and vector datasets.
- Token cost tracking — Built-in measurement of token usage for each prompt or chain.
Use cases
- Deploy a chat-with-PDF assistant that queries a Supabase vector store.
- Run REACT-style or HyDE prompt chains across multiple LLM providers.
- Version and test prompt changes without modifying application code.
- Scale chain-of-thought reasoning tasks on GPU or TPU clusters.
EdgeChains FAQ
Is EdgeChains production ready?+
Yes, it builds on Cloudflare's honojs and Google's Jsonnet, both proven in production.
How are prompts managed?+
Prompts are written in Jsonnet files, keeping them declarative, versionable and testable outside the code.
Does EdgeChains handle parallel execution?+
It automatically parallelizes LLM chains across CPUs, GPUs and TPUs using a WebAssembly runtime.
What licensing options are available?+
EdgeChains is released under AGPL-3.0 and also offers a commercial license on request.
How can I try the library?+
Clone the GitHub repo, install dependencies, configure secrets, and run the provided examples.
Summarized by DevHunt from github.com · Oct 4, 2026. Details may change; check the official site.
