Autonomous Intelligence for Deep Technical Audits & Refactoring
Beyan has evolved from a static prompt library into a fully autonomous agentic framework. It no longer just provides "what" to ask; it manages the "how" of analysis and execution.
- Autonomous Discovery: Automatically fingerprints 20+ technologies in your project.
- Smart Compilation: Dynamically builds a context-dense master prompt by selecting only relevant expert modules.
- Agentic Loop (Mode 3): A semi-autonomous cycle that follows a Analyze → Plan → Code → Test → Commit workflow with human-in-the-loop safety gates.
- Token Budgeting: Intelligent module pruning to fit massive codebase context into LLM limits.
Get your first deep analysis report in seconds:
# 1. Setup
git clone https://github.com/XINMurat/beyan.git
cd beyan/v2
pip install -r requirements.txt
# 2. Run Analysis (Mode 1)
python cli/analyzer.py --target /path/to/your/project --mode 1 --lang en --api openai
# 3. Interactive Fix (Mode 3)
python cli/analyzer.py --target /path/to/your/project --mode 3 --lang en --api anthropicUnlike the manual process of v1.0, Beyan v2.0 automates the entire analytical pipeline:
graph TD
A[Target Project] --> B{Discovery Engine}
B -- Identifies Tags --> C[Go, Node, K8s, Security, etc.]
C --> D{Compiler}
D -- Loads Modules --> E[Custom Compiled Prompt]
E --> F{Agentic Loop}
F -- Mode 1 --> G[Final Analysis Report]
F -- Mode 2 --> H[Implementation Plan]
F -- Mode 3 --> I[Interactive Code Refactoring]
Beyan v2.0 is powered by the battle-tested v1.0 prompt family. You can still use these prompts manually for specialized needs:
- Application Analysis
- OS / System Software
- Research / AI-ML
- Data & Analytics
- Infrastructure / DevOps
- Legacy / Migration
- Blockchain
- NOT DETECTED Contract: If information isn't found, the agent flags it. Never halluncinates.
- Two-Layer Analysis: Descriptive layer (facts) first, then Evaluative layer (judgment).
- Safety Gating: Mode 3 creates a safety branch and asks for confirmation before every single file change.
- Reconstructibility: "A new developer should be able to reconstruct the entire system solely from the analysis output."
v2/: The core Agentic Framework (CLI, Orchestrator, Discovery).tr/&en/: The multi-language expert prompt modules..github/: CI/CD workflows for automated testing.
MIT — free to use, modify, and distribute.
Beyan was developed through an iterative self-referential process: designed, then audited using its own Meta Audit prompt, improved based on findings, and health-scored before release.