Build RAG pipelines, agents, and graph workflows in Python.
Async-native · Streaming-first · 2 core dependencies.
Why SynapseKit
Built for production Python
Designed for engineers who want full control without writing everything from scratch.
Every API is async/await first. No sync-first retrofit. Sync wrappers included for scripts and notebooks.
Token-level streaming is the default, not an afterthought. Works identically across all 13 LLM providers.
numpy and rank-bm25 only. Every other capability is behind an optional extra. Install what you need.
13 LLM providers and 5 vector stores behind the same API. Swap providers without rewriting a single line.
RAG pipelines, agents, and graph nodes are interchangeable. Wrap anything as anything.
No hidden chains, no callbacks, no global state. Every step is plain Python you can read and override.
Explore the docs
Everything you need
From a 3-line quickstart to production graph workflows.
Retrieval-augmented generation with streaming, BM25 reranking, conversation memory, and token tracing.
Read docs →ReAct loop for any LLM. Native function calling for OpenAI, Anthropic, Gemini, and Mistral. 16 built-in tools, fully extensible.
Read docs →DAG-based async pipelines. Parallel execution, conditional routing, typed state, fan-out/fan-in, SSE streaming, event callbacks, human-in-the-loop.
Read docs →OpenAI, Anthropic, Ollama, Gemini, Cohere, Mistral, Bedrock, Azure, Groq, DeepSeek, OpenRouter, Together, Fireworks — all behind one interface.
Read docs →InMemory, ChromaDB, FAISS, Qdrant, Pinecone. One interface for all backends. Swap without rewriting.
Read docs →Complete reference for every public class and method in SynapseKit.
Read docs →