Lazarus, a coding agent for long-horizon tasks
Persistent Python runtime keeps state alive across tool calls, unlike Claude Code's stateless tools.
LLM framework for long running agents
Yet another LLM orchestration layer over LiteLLM + Pydantic when DSPy and LangChain dominate.
Python developers building multi-step AI agents
LangChain · LangGraph · DSPy
This is Andrei from askmanu and I'm super happy to share a new framework I've been working on: acorn.
It takes all the best parts of DSPy, langchaing, instructor, etc and wraps it in a beautiful and easy to use API. Very easy to define model I/O, branches, define callbacks for every step, etc
See the getting started docs here: https://github.com/askmanu/acorn/blob/main/docs/getting-star...
Try out the different demos here: https://huggingface.co/spaces/askmanu/acorn
Persistent Python runtime keeps state alive across tool calls, unlike Claude Code's stateless tools.
Engine-managed memory beats model self-summarization on OOLONG benchmarks.
Layer 2 execution testing without LLMs when eval frameworks only test intelligence.
Lifecycle-aware security pipeline, not point tools—shared context from ingress through output.
Python FFI bridge via PyO3 is clever, but tool-calling frameworks are crowded.
Empirical data from a shipped app, but it's a paper not a tool.