CAF Multi-AI-Agent Framework
The CAF Multi-AI-Agent Framework is an AI orchestration platform that coordinates multiple specialist agents to answer complex queries.
Think of it as an intelligent dispatcher — it receives a user's question, figures out which agents are best equipped to handle it, runs them in the right order, and combines their answers into one coherent response.
There are two primary ways to interact with the platform
- As a user — send queries through the UI or REST API and receive AI-generated answers
- As an integration team — register your own agent and its APIs so the Supervisor can route tasks to it
Examples of what the framework enables
- Routing a sales query to a Sales Order agent and a pricing agent simultaneously
- Chaining an order-lookup agent's output into a fulfilment-status agent
- Answering hybrid questions that require inputs from multiple enterprise systems
- Plugging in a new SAP or non-SAP backend agent without changing the core platform
Key Concepts
| Concept | What it Means |
|---|---|
| Supervisor | The central orchestrator that decides which agents to invoke and in what order |
| Agent | A registered specialist service that handles a specific domain or set of APIs |
| Planner | Breaks a user query into a plan of sub-tasks for the Supervisor to execute |
| Guardrails | Pre-processing checks applied to every query before it reaches the AI pipeline |
| Prompt Refiner | Cleans and sharpens the user's query for better agent routing and responses |
| Aggregator | Combines outputs from multiple agents into a single final answer |
| Decision Trail | A log of routing decisions made during query processing |
| GPT-4o (SAP AI Core) | The LLM backend powering all reasoning, planning, and response generation |