AIRLINES

AI for the moments airlines actually need it.

Disruption rebooking that does the work the call centre can't keep up with. MRO technical search across 20,000 manual pages. Crew scheduling Q&A inside FAR / EU OPS rules. EU 261 compensation calculations that survive an audit.

Customer-hosted on Kubernetes, in your cloud or on premises. For enterprise teams managing AI across business units.

Book a demo

Or email sales@kosmoy.com.

Airlines are 24/7 operations with the most regulated customer-rights environment of any consumer industry. The IROPS (irregular operations) day — weather, ATC, mechanical, strike — is where everything compounds: passenger volume on the contact channels spikes, crew schedules unravel, MRO constraints cascade, EU 261 exposure mounts. AI is well-suited to the disruption surface; the discipline is in not letting AI compound the chaos.

Regulators have already shown they will treat AI-driven decisions on rebooking, compensation and customer rights as part of the airline's obligations. EU 261/2004 forces specific compensation outcomes; an AI that miscalculates eligibility creates direct liability. EASA Part-145 governs maintenance with documented procedures. GDPR governs passenger data. The EU AI Act puts customer-facing chatbots under transparency obligations.

Kosmoy is the operating layer for airline AI from passenger ops to flight ops to MRO. Single-tenant Kubernetes deployment fits airline IT (typically private cloud, often co-located with the OCC and the maintenance systems). Every AI decision that affects passenger rights is logged, attributable and reviewable.

What this industry runs into.

IROPS volume vs human capacity

A major weather event can generate 20–50x normal contact volume in a six-hour window. Human staffing can't flex that fast; AI rebooking is the only way to keep customer commitments.

EU 261 compensation accuracy

EU 261 eligibility depends on cause (extraordinary circumstances vs operational), delay length, distance and offered alternatives. AI has to get it right; an error rate that's tolerable in retail is unacceptable here.

MRO documentation depth

Maintenance manuals span thousands of pages per aircraft type. Engineers can't memorise; AI search has to be fast, accurate and citation-grounded.

Multi-channel passenger experience

Web, mobile, voice, kiosk, lounge agent, gate agent — passengers expect consistent answers across channels. Each channel needs the same agent intelligence.

Regulatory landscape.

The regulations that shape AI in airlines — and where each one bites on AI deployment.

EU 261/2004Regulation establishing common rules on compensation and assistance to passengers· EU

Specific compensation rules based on delay length, distance and cause. AI-driven eligibility determinations create direct legal exposure on errors.

EASA Part-145Approved Maintenance Organisation requirements· EU

Maintenance procedures must follow approved data. AI-assisted technical searches must surface the approved procedure with citation, not summarise.

FAA Part 121Operating Requirements: Domestic, Flag, and Supplemental Operations· US

US carriers operate under Part 121 with specific maintenance, dispatch and crew rules. AI in operational systems is in scope of FAA inspections.

GDPRGeneral Data Protection Regulation· EU

Passenger PNR data is highly sensitive. Profiling under Art. 22 covers AI-driven offer personalisation. Cross-border data transfers constrained.

IATA TIMATICTravel Information Manual· Global

AI travel-document advice must reference TIMATIC; outdated advice creates passenger denial-of-boarding risk.

EU AI ActRegulation (EU) 2024/1689· EU

Customer-facing chatbots fall under transparency obligations. Crew scheduling AI may fall under workforce management high-risk classification.

Use cases, and what to measure.

IROPS passenger rebooking

Storm cancels 40 flights at LIN over 6 hours. The agent reads the affected PNRs, available seats on the carrier's network and partner carriers (interline), the passenger's status and EU 261 entitlements, and proposes rebooking options ranked by passenger preference and operational feasibility. Passenger accepts via app, web or voice; the agent books, reissues and notifies. Human ops review exceptions.

What to measure

Minutes from cancellation to confirmed rebooking per passenger, NPS during disruption, and whether every EU 261 decision is traceable case by case.

Powered byAction CapsuleAI GatewayGuardrailsMCP Gateway

MRO technical Q&A

Engineer working an A320-family check: 'what's the inspection interval for this brake unit P/N at this cycle count?'. The agent retrieves from the AMM, the operator's MEL/CDL, recent service bulletins and the carrier's engineering practices, returns the answer with the AMM page citation. Engineer verifies and signs.

What to measure

Time to answer common technical look-ups, and any work performed against superseded data.

Powered byRAG-in-a-BoxAction CapsuleGuardrails

Crew scheduling Q&A

Crew member: 'can I swap this leg with crew member X next Tuesday?'. The agent reads the duty schedule, the carrier's CBA, FAR 117 (US) or EU OPS flight time limitations, and the operational impact, and returns: yes/no/conditional with the rule that bound the answer. Crew planner reviews exceptions; the agent never overrides the planner.

What to measure

Admin time per crew swap, and flight time limitation breaches caught at request time rather than at audit.

Powered byAgent BuilderGuardrailsInsights Dashboard

EU 261 compensation calculation

Delay of 4 hours on a Bologna-Madrid flight, mechanical cause. The agent reads the flight record, the cause classification, the distance band and the passenger's claim, and computes EU 261 entitlement: €400 per Article 7 + Article 8 alternatives. The agent prepares the response with the specific clauses that apply; customer service reviews; the agent never auto-pays.

What to measure

Time from claim to compensation decision, the share of decisions that cite the applicable articles, and how many decisions are overturned on appeal.

Powered byAgent BuilderGuardrailsAI Gateway

Flight ops dispatch synthesis

Pre-departure synthesis: weather, NOTAMs, ATC restrictions, fuel planning context, alternates. The agent reads the relevant data, summarises into a dispatch briefing, flags items requiring captain attention. The dispatcher and captain review and sign.

What to measure

Dispatcher time per briefing, and items the captain or dispatcher had to add that the briefing missed.

Powered byAgent BuilderRAG-in-a-Box

Agent governance

Where airlines agents need extra discipline.

Airline agents fall on a steep risk gradient. A loyalty programme Q&A agent is low-risk; an IROPS rebooking agent that issues new tickets is operationally consequential and creates legal commitments. Kosmoy's Action Capsule is mandatory for any agent that takes ticketing, payment or compensation actions. Pre-flight authorisation enforces allowed-action scope: the agent can rebook within the carrier's interline agreements, for the passenger's class of service, but cannot offer a rebooking that breaks operational rules.

EU 261 and crew scheduling decisions are auditable for years. The Agent Registry captures every agent's version, model and configuration. The Insights Dashboard captures decisions and their evidence chain. Regulatory disputes survive on the platform's audit trail rather than on email reconstruction.

Chatbot use cases

Chatbots, by surface and risk class.

Airline chatbots run across the most channels of any industry: web, app, kiosk, lounge, gate, voice, in-flight (premium), corporate booking tools. Each channel carries the same brand promise but has different data scope and different latency expectations.

Web/app passenger chatbot

Booking, check-in, baggage, status, schedule changes, compensation. Action Capsule for any agent that issues tickets or initiates refunds.

Lounge / gate agent copilot

Agent-facing — VIP recognition, irrops handling, special service requests. Reads the PNR ecosystem; suggests next actions.

MRO technical chatbot

Engineer-facing — manual Q&A, troubleshooting, parts lookup. Citation-grounded; never substitutes for approved maintenance data.

Corporate / TMC portal

Travel manager queries — fare class availability, policy compliance, compensation history. Read-only retrieval; sales agents bind contracts.

How Kosmoy fits.

Airlines benefit from Kosmoy's combination of scale economics and audit discipline. Volume on customer-facing chatbots demands the LLM Router and Cost Tracking. Audit on rebooking, compensation and crew decisions demands the Agent Registry and Insights Dashboard. The Action Capsule contains the agents that take actions with legal and operational consequences.

Deployment fits the airline IT pattern: single-tenant in the carrier's cloud or data centre, often co-located with OCC and MRO systems. Open-weight models on carrier infrastructure cover high-volume tiers; commercial models cover complexity. PNR, crew rosters and engineering data never leave the carrier's perimeter.

Module questions, answered straight.

Does the rebooking agent commit the airline legally?

Only within the scope the carrier configures. The Action Capsule enforces allowed actions: which fare classes, which interline partners, which compensation thresholds. Out-of-scope attempts fail at the boundary; human ops review the exceptions.

How do we keep crew agreement / labour-rule compliance?

Carrier-specific CBA rules, FAR 117 / EU OPS FTLs and any local labour rules are encoded in the Guardrails and the agent's policy scope. The agent never offers a swap that breaks a rule; rule-breaking attempts surface to the planner.

Can MRO chatbots run offline at remote stations?

Common AMM/SRM lookups can run offline with cached manuals; live-data lookups (recent service bulletins, MEL updates) need connectivity. Engineers see clearly what's current vs cached.

How does Kosmoy support EU 261 audit?

Every compensation decision is logged with the rule chain that applied — cause classification, delay band, distance, alternatives offered. Regulators receive a structured audit pull rather than email archaeology.

Enterprise proof

Built for environments where AI cannot become another unmanaged SaaS silo.

Kosmoy works with regulated and operationally critical organizations, including Banca d’Italia and Leonardo.

Banca d'Italia
Leonardo
  • Runs in your Kubernetes
  • Azure · AWS · GCP · on-prem
  • LLM · MCP · A2A
  • No mandatory vendor-hosted control plane

Run airline AI inside the regulator's audit envelope.

See how the AI Gateway, Action Capsule and Agent Registry handle IROPS rebooking, MRO Q&A, crew scheduling and EU 261 calculations.

Book a demo

Or email sales@kosmoy.com.