SynapseKit
A minimal, async-first Python framework for building LLM applications with only two hard dependencies, offering unified access to dozens of model providers, vector stores, and agent tools.
SynapseKit is a Python framework for retrieval-augmented generation, agents, and graph-based workflows, built with only two required dependencies, numpy and rank-bm25, so every other integration is optional. It exposes a single interface across 46 language model providers and 32 vector stores, along with data loaders, agent tools, and an embeddings and reranking layer, with every abstraction implemented as plain, readable Python.
Every API is async by default, with synchronous wrappers for scripts and notebooks, and token-level streaming is the default response mode across providers. The framework includes a built-in, zero-dependency live dashboard that streams each language model call, tool use, retrieval, database write, and cost in real time, bound to localhost and inactive unless explicitly enabled.