ForeSail

ForeSail is a external-risk monitoring agent for cross-border trade shipments. Cross-border merchants have already automated their internal paperwork (contracts, POs, and letters of credit) generation but they have almost no systematic read on the external world: weather, geopolitics, port congestion, canal and strait closures. Those are exactly the black-swan events (a typhoon over the loading window, a Strait of Hormuz crisis on the route, a Red Sea war-risk diversion) that break a fixed LC-presentation or latest-shipment deadline and turn into six-figure losses — demurrage, LC discrepancies, rejected documents, missed cover. ForeSail tracks each shipment as a single trade Case. For every Case it: 1. Extracts and confirms the trade facts from the underlying documents (Contract/PO, Booking Confirmation, Letter of Credit), with evidence text and confidence per field, plus human confirmation and conflict resolution. 2. Maps the Incoterm responsibility split (CIF/FOB — who bears risk and cost at each leg) and builds the obligation & deadline calendar. 3. Continuously watches the outside world against that Case — live GDELT (global news/geopolitics), Open-Meteo (weather), and RSS connectors, plus a predictive layer that aligns the voyage schedule with weather forecasts, typhoon tracks, and a geopolitical corridor state machine. 4. Scores and classifies each event's relevance to this shipment, transitions the Case through a risk state machine (DRAFT → ACTIVE → WATCHING → AT_RISK → ACTION_REQUIRED → MONITORING), and generates evidence-backed, ready-to-send action drafts. The differentiator: every risk decision is produced by a deterministic, auditable engine. The LLM assists only with document extraction and drafting wording; it never invents a risk score. That is what makes the output safe to put in front of a money decision. Who is it for: Trade operations, risk, and compliance teams at cross-border trading SMB to mid-market players, like exporters and importers who run active shipments under LC and Incoterm obligations (where a single missed deadline is materially painful and there is no dedicated risk desk). Current product surface Frontend (React/TypeScript): Create Case, Case Library, and a per-Case workspace with nine tabs: Overview, Documents & Evidence, Conflicts, Agent Runs, External Events, Risks & Obligations, Actions, Treatment Plans, Audit — a dark Leaflet route map with voyage timeline and threat markers, a company-profile Settings surface. Backend (FastAPI, Python): a deterministic decision core plus orchestration agent, with dedicated engines for relevance, Incoterm rules, corridor risk, hazards, and voyage schedule; event connectors for GDELT, Open-Meteo, RSS/real-search, typhoon tracks, a risk calendar, and a policy registry; document extraction for TXT/DOCX/text-based PDF; and a port registry spanning 1,610 ports. Runs in REAL mode by default, with an offline demo mode; optional OpenAI LLM enhancement behind explicit flags.