autogen requirements agent.stub

---

id: autogen-requirements-agent name: AutogenRequirementsAgent repo_path: ~/Documents/sw30labs/repos/AutogenRequirementsAgent remote_url: https://github.com/sw30labs/AutogenRequirementsAgent.git primary_language: Python framework: pyautogen + Ollama (Nemotron/Hermes) ingested: 2026-04-13 last_commit_date: 2024-11-29 category: agentic-frameworks stacks: [agentic]

Purpose

Experiment in strict JSON messaging between AI agents to generate and refine project requirements. Uses nested GroupChat with QA review simulation, powered by locally-served open-source LLMs via Ollama instead of proprietary APIs.

Architecture

• Four-agent system: Initializer → Planning Agent → Analyst → QA Manager (nested in loop per step)
• Strict JSON messaging: All inter-agent communication follows strict JSON schema for state transitions
• Nested GroupChat structure: Inner monologue-like review process per planning step
• State transition function: Validates JSON responses, extracts specific keys, applies DoCase logic
• Semantic error trapping: Python calculation verification for inference results
• QA approval threshold: ≥ 4.5/5 ratings required for step completion
• Concurrent processing: ThreadPoolExecutor with Semaphore (max 3 simultaneous)
• .env configuration: Flexible model, base URL, temperature, and timeout settings

Dependencies

• pyautogen - agent framework
• PyYAML - configuration parsing
• ollama (or Ollama service via API) - local LLM serving
• Python stdlib (threading, json, yaml)

Key Features

• Automated requirements generation workflow
• Nested GroupChat with feedback loops
• Functional + Non-functional + Integration requirement coverage
• QA Manager simulation for quality gates
• Local OSS LLM support (Nemotron 70B, Hermes 70B)
• JSON + TXT output formats
• Token budget awareness
• Random seed generation for varied outputs
• Parallel step processing

Tags

• python
• llm
• local-inference
• json
• developer-tools

Entities and Topics

• agentic-frameworks
• developer-tools