Carrier Tracking Narrator

What it does

Carrier Tracking Narrator pulls shipment status from a carrier tracking API (FedEx/UPS/DHL), normalizes carrier-specific status codes into a common vocabulary, derives an ETA and delay flag, and classifies exceptions (customs hold, delivery failed, damaged, address issue, weather delay, lost) — all deterministically in code. An LLM is used only to phrase a human-readable status/alert message; it never invents or changes a status, ETA, or exception. Gated: a live run needs a carrier sandbox API and approval; this build is connector-ready and validated against deterministic mock fixtures. Guardrails: read-only GET only, prompt-injection defang on carrier free text, PII masking, no fabrication. Built for logistics and customer-support teams who want automated shipment status narration instead of manually checking carrier portals.

Example prompts

• Track these shipments and tell me their status and ETA
• Check if any of my shipments have exceptions
• Classify the exception for this DHL shipment
• Derive the ETA for this UPS shipment

Tools (9)

Tools the agent exposes — your AI client calls them automatically when it needs them.

• run_full — Run the full pipeline: collect tracking data, normalize status, derive ETA, classify exceptions, synthesize a status/alert message.
• fetch_tracking — Fetch raw normalized tracking data (carrier API read-only GET; mock fixtures when no live credentials).
• normalize_status — Deterministically map a carrier-native status code to the common status vocabulary.
• derive_eta — Deterministically derive ETA and a delay flag from ship date / carrier ETA / last scan and the promised date.
• classify_exception — Deterministically classify a shipment exception from carrier status and scan notes.
• synthesize_message — Produce a human-readable status/alert message for a shipment. LLM for phrasing with a deterministic template fallback.
• list_capabilities — List tools, supported carriers, normalized status vocabulary, and credential slots.
• discover_intent — Understand your goal and co-design the exact input via clarifying questions before running.
• plan_inputs — Plan/brainstorm the inputs for a tool: returns questions, schema and a ready-to-edit example.