Only your task text leaves the browser. A Cloudflare relay holds the AgenticOrg access key server-side — no credentials ever reach this page.
A virtual employee on Pine Labs' AgenticOrg platform. It reads the task, reasons with the platform's language model, and decides which flight tools to call — nothing here is scripted.
The agent has its own registered identity (DID) and is only authorised for four flight tools. Every run is scored for confidence; low-confidence runs are flagged to a human review queue. It runs in shadow mode while it earns trust.
Tool calls travel through the registered connector mcp_easemytrip_flights_planet to mcp.agentcheckout.app/emtflight — an API-key-protected service exposing exactly four actions: search, reprice, book, booking status.
Real FlightSearch, Reprice and AirBook calls on EaseMyTrip's UAT system. Fares are live airline fares; bookings issue real test PNRs you can query any time. The fare quoted after reprice is exactly the fare booked.
planet.agentcheckout.app shows the product experience: the PLANET app calls the flight APIs directly, so results are instant and every screen is deterministic — search cards, fare re-check, UPI, PNR.
This page shows the agentic architecture: the same flight capability is published as a governed connector on Pine Labs' AgenticOrg marketplace, and an autonomous agent uses it the way an employee would — reading a plain-language task, choosing tools, and acting. A run typically takes 1–2 minutes because a real reasoning loop is executing, not a script.
hitl_triggered on results here — that is governance working, not an error.https://mcp.agentcheckout.app/emtflight (streamable HTTP, API-key auth). The quoted reprice fare is contractually the booked fare — booking reuses the exact repriced segment.mcp_easemytrip_flights_planet; the tool catalog was discovered automatically from the MCP endpoint.PLANET Flight Buyer (type planet_flight_buyer, travel domain) is authorised for those four tools only and carries the search-then-reprice-then-book discipline in its prompt./agenticorg/run) that triggers the agent via AgenticOrg's run API and returns the output, confidence, status and reasoning trace shown here.