griptape

Why we included this project

LLM-backed apps tend to start as glue code: loops that shuffle prompts to a model, stitch in retrieval results, and trigger tool calls whenever the model asks for one. Griptape gives that work a structure without forcing an architecture on you. You define tasks, chain them through a pipeline or fan them out across a parallel workflow, and attach tools and memory where a step needs them. Memory is the quietly valuable part, since conversation memory keeps context across turns while task memory parks large or sensitive outputs off the prompt so they do not eat your token budget. Providers sit behind drivers, so moving from OpenAI to a local model is a config change rather than a rewrite, and built-in engines for RAG, extraction, summarization, and evaluation cover the patterns most teams need anyway.