Loading...
Define the tools and actions available to the agent. Specify tool interfaces, input/output schemas, and how the agent selects which tool to use...
In terms of tools given this is a simple Q&A RAG we do not need anything, no web search, just access to read makrdown which comes by default and some sort of protocol to access the knowledge base, nothing more than that
Design the overall agent architecture. Choose an orchestration pattern (single agent, multi-agent, hierarchical). Identify key components and how they interact. Use the diagramming tool to illustrate the architecture.
the architecture will be a single agent interacting with some sort of server via MCP so it has access to the knowledge base, a qdrant one.
We ll need as well some sort of pipeline or plain script to keep digesting and chunking documents to add as base for the answers
How does the agent manage memory and context? Define short-term (conversation) and long-term (persistent) memory strategies. How is context window managed? What retrieval mechanisms are used (RAG, vector stores, summarization)?
memory wise would be nice to keep last 5 questions so we can build on past conversations, for the rest we can do some heavy summarization.
in terms of context we will include both answers and last 3/5 questions
Deep dive into the agent's reasoning strategy. How does it plan, decide, and self-correct? Discuss prompt engineering, chain-of-thought, reflection loops, failure handling, and tradeoffs.
robust faithfulness checker
What guardrails and safety mechanisms does the agent have? Define permission boundaries, human-in-the-loop checkpoints, content filtering, rate limiting, and how the agent handles adversarial inputs or unexpected states.
We need to make sure the bot abstains from doing any other thing other than answering questions is supposed to ask, prompt shpuld strictly guard against this. It should not leak company details
no access to tools like code execution to be sure we can t get destroyed