WHAT IT DEMONSTRATES
THE FINANCE AGENT SUITE
Find your finance focus.
Pick a domain. See the story. Try a sample.
Browse by domain
Working prototypes by domainREPORTING & FP&A / DOMAIN SPOTLIGHT
Check the metric. Shape the story.
Compare KPI definitions and compose supported monthly narratives with two visual stories.TREASURY & CASH / DOMAIN SPOTLIGHT
See the cash. Inspect the assumptions.
Inspect current cash and forecast deviations with two visual stories.RECEIVABLES & COLLECTIONS / DOMAIN SPOTLIGHT
Explain the balance. Focus the follow-up.
Trace short payments and review collection priorities with two visual stories.PAYABLES & SPEND / DOMAIN SPOTLIGHT
Trace the obligation. Explain the balance.
Compare supplier items and approved order capacity with two visual stories.CLOSE & RECONCILIATION / DOMAIN SPOTLIGHT
Connect the balances. Inspect the support.
Trace reciprocal balances and journal evidence with two visual stories.FRAUD & RISK / DOMAIN SPOTLIGHT
Spot the signal. Trace the evidence.
Inspect payment signals and payee verification with two visual stories.PAYMENTS / DOMAIN SPOTLIGHT
Follow the payment to its outcome.
Start with a timeout. Explore five working Payments agents.No agents match these filters.
Try another finance topic or clear the filters.
The agent build roadmap
BEHIND THE STORIES
Follow the evidence further.
Explore the source files, context graph, saved reviews, and measured checks behind the lab.
Each prototype exposes its assumptions, calculations, and review boundary.
Download the agent build listInvestigation workbench
Document intake → Coordinated checks → Cited brief → Human review
Evaluation lab
Saved results · Explicit expectations · Measured runtime
What this agent demonstrates
Explore all three sample scenarios
Choose an investigation
Start with a result, then change the assumptions.
Calculations and template explanations run in your browser. No live model, account connection, or financial action is involved.
BEHIND THE DEMONSTRATIONS
How the Finance Agent Lab works
Explore the capabilities, architecture and review boundaries behind the visual stories.
03 / AGENT CAPABILITY MAP
Reusable capabilities.
Purposeful combinations.
A generic finance interpretation of the agent catalog. Each workflow combines capabilities around a specific reviewer decision.
Read & structure
Extraction · Document intelligence · Document retrieval
Import defined synthetic text documents, inspect extracted fields, and open line-level citations. Original source files are preserved with the case.
Investigate & reconcile
Transaction analysis · Reconciliation · Duplicate detection · Case investigation
Trace transaction and invoice relationships, compare receipt allocations and settlement totals, and leave unsupported conclusions unresolved.
Validate & explain
Validation · Evidence validation · Summarization · Communication
Run rule checks and prepare cited packets. The connected workbench includes a model explanation endpoint awaiting its provider connection. Document authenticity is not verified.
Investigate payment exceptions
Refund & duplicate charge · Chargeback insight · Dispute resolution · FX investigation
Inspect refund/dispute overlap, merchant payouts, and supplied-rate currency conversion. An illustrative evidence checklist now assembles linked dispute exhibits. Applicable network rules and external submission remain outside the prototype.
Explain financial performance
Budget variance · Driver analysis · Finance commentary
Separate expense changes into volume, unit cost, fixed costs, and unexplained ledger residuals using comparable, evidenced inputs.
Risk synthesis, KYC, adverse-media screening, and voice servicing are adjacent directions for future exploration.
04 / SYSTEM DESIGN
Reliable arithmetic.
Grounded explanations.
Separate facts, calculations, and judgment so a reviewer can follow every conclusion.
Source records
Versioned synthetic records with identifiers and explicit relationships.
ImplementedDeterministic tools
Integer-cent calculations, validation, matching rules, and uncertainty checks.
ImplementedEvidence packet
Finding, source references, checks, policy, and proposed reviewer action.
Template-based todayHuman review
Record a reviewer decision against an exact case version. Evidence changes invalidate earlier approvals.
Saved review historyThe boundary is part of the design.
Changing an input invalidates the previous packet. A missing policy or unverified link blocks a conclusion. Pending disputes stay separate from confirmed postings.
The prototype demonstrates workflow design and finance controls. It does not claim autonomous agents, production integration, or measured business savings.
Available in this version
Twenty-seven interactive workflows · Eighty-one standalone scenarios · Saved investigations · Source document intake · Reviewer history · Twelve measured evaluation fixtures
Next implementation milestone
Connect the prepared model explanation endpoint, then evaluate model grounding separately from the deterministic fixture suite.
Agent not found
Choose a workflow from the suite.
Open agent gallery