P-020 · Case study
OPEN SOURCEFinance Agent With Memory Across Sessions
A chat-first finance agent, built on a hand-rolled Anthropic SDK loop with no framework, that carries what it learned in one conversation into the next one three days later. It ships with a behavioural eval.
More detail
Memory holds intent: commitments, patterns the user acknowledged, and reminders, each tagged with whether the user said it or the agent inferred it. Balances, bills and transactions always come fresh from tools and are never recalled from memory. At the start of the second session the agent briefs the user from memory before they speak, then ties the new question to what it already knows. A behavioural eval reads the second session's transcript and checks that the briefing fires, the tools get called, the fresh balance is quoted and nothing stale is recalled.