-
Notifications
You must be signed in to change notification settings - Fork 0
Home
Welcome to the samf-framework wiki!
The SAWANT Agentic MoSCoW Framework (SAMF) is a lightweight, deterministic verification layer designed to enforce strict structural and semantic boundaries on Large Language Model (LLM) outputs within complex, stateful multi-agent orchestration loops.
In multi-agent architectures (such as LangGraph), a single unvalidated, out-of-bounds LLM response can propagate down the graph, corrupting application states, causing infinite loops, and inflating token costs. SAMF solves this by adapting the classic project management MoSCoW matrix (Must-have, Should-have, Won't-have) into concrete, machine-readable software execution contracts.
In non-deterministic AI pipelines, prompt engineering alone cannot guarantee zero runtime failures. When an agent in a multi-step workflow fails to output a critical parameter or hallucinates an excluded string, the system typically breaks down-stream.
🏠 Home │ 🛡️ Core Concepts │ 📈 Usage & Benchmarks
Disclaimer: This wiki documents an independent open-source hobby project developed entirely on personal time and hardware. It is not affiliated with, sponsored by, or endorsed by my employer.