DuraLang

<div align="center">

<h1>duralang</h1>

<p><strong>Agents that cannot fail. One decorator. Durability for Stochastic AI Systems.</strong></p> <p> <code>duralang</code> is the missing durability layer for LangChain.<br> Write the same LangChain code you already know — add <code>@dura</code> — and every LLM call,<br> tool call, MCP call, and agent-to-agent call becomes individually recoverable,<br> automatically retried, and fully observable through Temporal. </p>

<p><strong>No new framework. No graph DSL. No code rewrite. Just durability.</strong></p>

[](https://pypi.org/project/duralang/) [](https://python.org) [](LICENSE) [](https://temporal.io)

</div>

https://github.com/user-attachments/assets/e129971c-6bc6-437b-9a7c-646e753c93e6

---

Most AI agent failures are infrastructure failures, not intelligence failures.

The model picked the right tool. The reasoning was correct. But a network timeout at minute 47 of a 60-minute run killed the entire pipeline — and you lost every completed step along with it.

Modern AI agents are stochastic programs.

They are not workflows. They are not pipelines. They are runtime-generated execution graphs driven by an LLM.

But every existing durability system is built for deterministic programs.

They assume:

• a known execution graph
• fixed control flow
• predefined steps

This assumption is fundamentally incompatible with LLM-driven agents.

The result: There is no durability model for stochastic programs.

Not in LangChain. Not in LangGraph. Not even in Temporal without rewriting everything.

This is the missing layer.

---

This is the reality of production agent systems today:

• LangChain gives you the best composability layer for L