Quiet Admin

• 

  Quiet Admin architecture

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for Quiet Admin

Inspiration

Life admin is noisy: autopay receipts and spam sit next to rate increases and school forms. We wanted an agent whose default is silence—an Everyday Agent that protects attention instead of competing for it.

What it does

Quiet Admin watches bills, renewals, appointments, and forms, classifies each as needs_decision / auto_handle / ignore, and notifies you only when a real choice is required.

• Ingests a sample inbox of bills, renewals, appointments, forms, and spam
• Classifies every item and drafts replies only for decisions
• Auto-handles confirmations and small autopay with quiet reminders
• Ignores spam and persists a full audit trail in a JSON state store

How we built it

• AWS Strands Agents SDK (strands-agents) with custom @tool functions
• Online path: Agent(system_prompt=..., tools=[...]) using the default Bedrock model provider
• Offline/demo path: QUIET_ADMIN_OFFLINE=1 rule-based classifier calling the same tools (judges can demo with zero AWS credentials)
• Optional Bedrock AgentCore entrypoint stub; Python CLI: python -m quiet_admin demo

Challenges & accomplishments

We balanced a real Strands path with a bulletproof offline demo and designed boundaries so “quiet” never hides costly mistakes (rate increases always surface). The MVP includes 8 fixtures, shared tools, state store, tests, architecture docs, and demo script. Apache-2.0 licensed.

What's next

Real Gmail/IMAP connectors, user-tunable money thresholds, approval UX for drafts, and full AgentCore deployment with observability.

Built with AWS Strands Agents SDK, Python 3.10+, Amazon Bedrock (online), JSON state store, and pytest.

Built With

• amazon-bedrock
• aws-strands-agents-sdk
• json-state-store
• python-3.10+