NOOA solves the fragmentation that slows down AI agent development today. Teams building agents for developer tooling, cybersecurity, cloud automation, data analytics, financial services or customer support currently juggle prompt templates, tool schemas, callback code and workflow graphs scattered across files. This makes it hard to test behavior, trace failures, refactor logic or version‑control agents like regular software. The result is longer cycles, hidden bugs and difficulty moving from prototype to production.
NOOA collapses all of those pieces into a single Python class. Each method becomes an action the model can take; class fields hold the agent’s state; method docstrings serve as prompts; type annotations act as contracts that the runtime enforces. When a method body is just an ellipsis (…) the framework hands it to an LLM‑driven loop—either a single typed call with retry or an iterative Python REPL where the model can execute code until it returns a validated result. Methods with normal bodies stay deterministic Python, so the model can call them as reliable tools.
A key innovation is pass‑by‑reference: large data structures stay live in the execution environment while the model sees only a bounded preview (type, length, head/tail sample). This prevents the context window from becoming a bottleneck, cutting token usage roughly in half compared with traditional harnesses. Benchmarks show NOOA achieving 82.2% on SWE‑bench Verified and 86.8% on CyberGym L1 with far fewer tokens and model calls than alternatives.
The framework is model‑agnostic, works with any LiteLLM‑compatible API (hosted services, Ollama, vLLM) and installs via pip install nooa (Python 3.12‑3.13). Because it can execute LLM‑generated code, it must run inside OS‑level isolation such as a container, VM or NVIDIA OpenShell. The current release is Apache 2.0‑licensed and marked as alpha, suitable for research pilots and internal platforms; regulated workloads should wait for a stable tag.
By giving developers one clear interface—methods, fields, docstrings and type annotations—NOOA makes agents testable, traceable, refactorable and version‑controllable like any other code, reducing integration risk and accelerating delivery of reliable AI‑powered automation.
#AI #Product #Automation #LLM #DevOps #CyberSecurity