prompt
FreeMulti-agent RAG orchestrator prompt for grounded, traceable answers
FreeFree tier
Inputs: textOutputs: text
About prompt
A system prompt designed for orchestrating multiple AI agents in a retrieval-augmented generation (RAG) workflow. It defines four specialized agent roles—Retriever, Synthesizer, Critic/Verifier, and Coordinator—and provides rules for iterative evidence gathering, synthesis, and critique. The prompt emphasizes grounded, traceable answers, minimal role proliferation, and efficient stopping criteria. It includes a structured output format with sections like User Question, Agent Roles Used, Retrieval Plan, Evidence Table, Conflicts/Gaps, Final Synthesized Answer, Confidence Level, and a decision on whether another retrieval round is needed.
Key Features
Defines four agent roles: Retriever, Synthesizer, Critic/Verifier, Coordinator
Iterative retrieval and critique loop to improve evidence quality
Structured output format with evidence table, conflicts, and confidence level
Stopping criterion: stop when marginal evidence no longer changes the answer materially
Every claim must map to explicit sources; surface conflicts instead of smoothing over
Pros & Cons
Pros
- Emphasizes source grounding and traceability of claims
- Avoids unsupported synthesis and role proliferation without reason
- Uses critique to iteratively improve evidence quality, not just add volume
- Provides a clear, structured output format for decision-relevant answers
Cons
- Specific to RAG tasks and may be overkill for simple queries
- Requires careful design of agent roles and coordination logic
- No built-in implementation; it is a prompt template, not a runnable system
Best For
Research question answering with multi-step evidence gatheringComplex retrieval-augmented generation tasks requiring source verificationMulti-agent RAG experiments and prototyping
FAQ
What is the Multi-Agent RAG Orchestrator prompt?
It is a system prompt that orchestrates multiple AI agents (Retriever, Synthesizer, Critic/Verifier, Coordinator) for retrieval-augmented generation. It defines roles, orchestration rules, and a structured output format to produce grounded, traceable answers.
What are the agent roles defined in this prompt?
The four roles are: Retriever (gather evidence, diversify sources), Synthesizer (combine evidence, produce current best answer), Critic/Verifier (challenge weak claims, detect conflicts), and Coordinator (allocate work, decide when to stop retrieval).
What is the output format of the prompt?
The output must include: User Question, Agent Roles Used, Retrieval Plan, Evidence Table, Conflicts/Gaps, Final Synthesized Answer, Confidence Level, and a decision if another retrieval round is needed.
Does the prompt allow using fewer agents if sufficient?
Yes. The prompt states: 'If one strong retriever is enough, say multi-agent is unnecessary.' It discourages role proliferation without reason.