Match with confidence
Check a clean invoice and bank match against deterministic reconciliation controls.
Put Cherry to work
Choose a scenario. Get a decision, its evidence, and the next step.
Know when to stop
Review a high-value match and see why a named human must approve the next step.
Resolve the uncertainty
Investigate conflicting evidence and find out what a reviewer needs to supply.
Demo capacity: one new run per two-minute window across all visitors. Results can take a few minutes.
Cherry’s findings
Ready when you are
A clear answer starts with the evidence.
Run a scenario above. Cherry will return the decision, the checks behind it, and any action needed from you.
Your approval stays yours.Cherry provides reconciliation and decision support. It cannot initiate payments, change beneficiaries, or grant human approval.
Deployed AWS architecture
Agent intelligence.
Deterministic financial authority.
Strands specialists investigate and explain. Cherry code governs reconciliation and policy. Humans retain approval authority.
Synthetic demo inputs
Invoice fixtures
Bank transaction candidates
Selected scenario
Amazon Bedrock AgentCore Runtime
Strands orchestrator
Amazon Bedrock · Claude Sonnet 4.6
Workflow specialistControl specialistEvidence specialist
Bounded tools → Cherry deterministic engineMatching · thresholds · workflow state · audit evidence
Controlled outcomes
Routine auto-reconciliation
Human approval required
Evidence exception
Amazon CloudFrontHTTPS website and signed origin requests
Lambda web gatewayValidates the scenario and starts a job
Lambda workerAsynchronously invokes the named runtime
Bedrock AgentCoreRuns the Strands specialist hierarchy
Results & observabilityThe worker stores its report in private S3. The browser polls the gateway for the result. Reports expire after one day. CloudWatch collects runtime logs.
Access & financial controlsIAM scopes the worker to this AgentCore runtime. The public demo accepts three predefined scenarios. Payment and human-approval actions are absent from the agent’s tools.
AgentCore, Lambda and S3 are deployed in London (eu-west-2); the Bedrock model uses an EU inference profile. CloudFront is global. Demo workflow state is in memory and is not a durable accounting ledger. The existing Google ADK/Gemini implementation is reused project history; it is not the runtime serving this AWS demo.