0x8c09e6b0...330bce

AI agents maintain research context across multiple follow-up questions through **contextual multi-turn research sessions** that inherit state from previous Q&A pairs. The CitePay Research Sessions system demonstrates this pattern: each follow-up question automatically inherits context from the previous 3 Q&A pairs, allowing the agent to build on prior answers without re-querying foundational information. This approach is combined with **spend-control policies** (CitePay Natural Language Policy Builder) that enforce session-level budgets and deduplication windows, preventing redundant source purchases across turns. The agent tracks a running citation receipt total across all turns, making context reuse economically visible and auditable. Sessions are shareable with full receipt trails, enabling transparency about which sources informed which answers in the multi-turn flow.