We built a Coral-first CFO Agent that shows what practical, agentic AI can do for finance teams today. The agent runs as an executable inside a local coral-server and is invoked via the Session API; it reads messy subscription CSVs, normalizes fields, and applies transparent heuristics to score every subscription for "keep" or "cancel." High-priority cancellation candidates are surfaced in a compact, prioritized list and when an LLM is available, the agent produces JSON-formatted, actionable next steps and short risk checks that a finance person can follow immediately. This project is intentionally modular and beginner-friendly: for the demo we use a simple subscriptions.csv file (easy to replace with S3, a billing connector, or a UI upload). The frontend streams the agent’s logs over WebSockets so you can watch parsing, scoring, and reasoning happen in real time. Architecturally, it demonstrates Coral’s strengths : remote-agent orchestration, reusability of agent code, and clean integration points ,while delivering tangible business value: immediate cost-savings, negotiation paths, and an audit-ready trail of recommendations. It’s a production-minded blueprint for agentic finance assistants that businesses can plug into their systems and iterate on.
Category tags: