đŸ’ŧ Agentic Finance Crew

Triage expense requests with a crew of AI agents — the model advises, a deterministic policy guardrail decides.

Intake → Analyst → Approverpolicy guardrail risk scoringhuman-in-the-loopruns 100% in your browser
Try an example:
Submit a request to see the verdict.

Run the bundled sample of 8 requests through the crew at once.

ℹī¸ How it works, safety & data

Intake validates & normalises · Policy Analyst checks limits, receipts, duplicates & scores risk 0–100 · Approving Manager recommends a decision within a hard guardrail. An LLM can never authorise spend on its own.

This page runs the exact deterministic engine from the source project — ported to JavaScript so it runs entirely in your browser (no server, no key, nothing sent anywhere). The full project also ships CrewAI & LangGraph engines that run the same workflow with real LLM agents. All figures and the spend policy are fictional sample data.

Demo policy: auto-approve ≤ $200 · always human > $2,000 · receipt required > $50. 🔗 Source & full architecture