Short return windows
Retailers set tight return windows, often just 15 to 30 days from delivery — and every store is different.
Autonomous return-deadline agent
Return Window Tracker
Never lose money to a missed return deadline again — an autonomous agent tracks every return window, decides what actually needs you, and drafts the message.
The problem
Returns should be simple, but the clock is always ticking — and it's easy to lose real money by forgetting.
Forgotten deadlines cost money
People forget the cutoff, miss the window, and end up stuck with items they meant to send back.
No unified tracker
Purchases live scattered across inboxes and stores, with no single place to see what's still returnable.
How it works
A Strands agent combines deterministic tools with LLM reasoning — so the math is exact and only the real decisions reach you.
Deterministic deadline tool
A pure check_return_deadlines tool computes days remaining for every purchase — reliable, repeatable, and never hallucinated.
Autonomy decision
Each item is classified as AUTO_HANDLED or NEEDS_YOU, so the agent only surfaces real decisions instead of noise.
only surfaces real decisionsLLM-drafted return messages
When action is needed, the LLM drafts a clear, polite return request you can send in seconds — no blank page.
LLM draftingBackground memory across runs
The agent remembers what it already handled between runs, so it never nags you twice about the same purchase.
persistent memoryArchitecture flow
Data flows through a single Strands agent that pairs deterministic tools with an LLM to produce structured, actionable output.
Purchase data
Synthetic orders with dates & return policies
Strands Agent
Reasoning + tool use
check_return_deadlines
decide_autonomy
LLM drafting
Writes return messages for NEEDS_YOU items
Structured output
Clear list of actions and auto-handled items
Built with
A lean, modern stack for autonomous agents.
Strands Agents SDK
Amazon Bedrock
Python
Streamlit