Become agentic-native
Move from disconnected digital touchpoints to AI agents that can understand intent, coordinate workflows and act across the journey.
The agentic layerfor airlines
flyjeeves helps airlines become agentic-native across the full passenger lifecycle: flight search, booking creation, reservation management, notifications and post-sale service.
- Search to post-sale
- Create and manage bookings
- Notify across every channel
- Learns gaps. Ships fixes.
Live agentic journeyActive
JourneyBCN · NTE
IntentBook and manage trip
Flight intentSearch understood
The agent captures route, dates, party needs and preferences before a booking exists.
Reservation flowBooking created
Selected flights, passenger details, ancillaries and payment steps move through controlled workflows.
Journey updatesNotifications sent
Relevant trip events can trigger outbound messages across the right passenger channel.
Post-sale actionTrip managed
Changes, check-in readiness, baggage and service requests pass through policy and execution controls.
Solution
From flight search to post-sale, one agentic airline layer
flyjeeves connects passenger intent, booking context, airline workflows and outbound communications into one AI operating layer. Instead of fragmented search, booking and support journeys, airlines can offer one always-available agent that helps passengers move from question to transaction to trip completion.
Convert intent into bookings
Assist passengers from flight discovery to reservation creation, payments and trip configuration in the channels they already use.
Operate after the sale
Manage changes, ancillaries, check-in readiness and proactive notifications through the same agentic service layer.
Capabilities
Built for real airline commercial and service flows
flyjeeves is not positioned as a generic chatbot. It is an agentic layer for airline operations: search-aware, booking-aware, action-oriented and designed to move a passenger from intent to outcome.
Agentic discovery before the booking exists
Capture passenger intent, understand route and schedule constraints, and guide users from flight search into bookable options.
- Flight search and preference capture
- Passenger-friendly comparison of options
- Channel-native assistance before checkout
Create and manage reservations through controlled actions
Move from intent to reservation by coordinating passenger data, ancillaries, quotes, payment steps and booking state.
- Reservation creation and lookup
- Baggage, seats, priority and payment flows
- Offer-confirm-execute patterns for sensitive actions
Service, notify and adapt after purchase
Keep assisting after checkout with changes, check-in readiness, disruption messaging and proactive trip notifications.
- Reservation changes and post-sale servicing
- Outbound notifications across channels
- Operational traceability with cost and token signals
Operational controls
Built with the controls airlines expect around agentic execution
flyjeeves is not just about answering well. It is designed to act with approvals, policy controls, privacy-aware handling and quality review loops around every runtime action.
AUTH / OTP
Verify sensitive passenger actions with explicit authentication and OTP gates before execution.
Files and knowledge
Ground shopping, booking and servicing flows with policy files, FAQs and curated airline knowledge sources.
GDPR-aware handling
Passenger interactions and booking data can be operated with privacy-sensitive handling, controlled memory and traceability requirements in mind.
Guardrails
Apply policy-aware controls around what the agent can say, trigger and execute across commercial and service flows.
VoiceTwilio voice
Web chatPassenger chat
WhatsAppWhatsApp
EmailPassenger email
OrchestratorJourney orchestrator
PlannerDefault planner
ComposerDefault composer
MemoryHybrid memory
GuardrailDefault guardrail
AssistantSearch assistant
AssistantBooking assistant
AssistantNotify assistant
AssistantService assistant
ToolSearch flights
ToolCreate booking
ToolSend message
ToolManage order
Human-in-the-loop
Escalate sensitive or ambiguous cases to human teams with explicit HITL routing instead of forcing full automation.
Memory
Use controlled memory and booking context so the concierge stays consistent across the passenger journey.
Quality assurance
Semantic QA and runtime evaluation help teams review quality, observe issues and tighten operational performance over time.
Tools
Expose airline actions like flight search, booking creation, baggage, payment or notification operations through controlled tool execution paths.
Continuous improvement
The system can learn where journeys break down and help ship the fix
flyjeeves is not limited to handling interactions. Through Intel and Dev Squads, it can analyze live journey patterns, surface improvement opportunities and translate those findings into delivery work that expands the product itself.
01 · Intel
Detect operational findings from real journeys
Intel can review searches, bookings, notifications and service conversations to detect repeated friction, missing capabilities, unresolved intents, handoff patterns and ancillary opportunities that the current runtime is not fully capturing.
02 · Dev Squads
Convert findings into concrete implementation work
Those findings can flow into Dev Squads, where the system structures the work needed to add or improve runtime behavior instead of leaving insights trapped in dashboards.
03 · Runtime
Deploy new tasks, nodes and workflows back into service
The outcome is an agentic airline layer that can evolve from production evidence: new tasks, new nodes, new orchestration paths and better coverage tied directly to observed passenger demand.
Intel reads the journey layer
Searches, bookings, notifications, service traces and unresolved intents can be analyzed to identify where the agentic journey is underperforming or leaving revenue on the table.
Dev Squads turn findings into delivery work
Detected opportunities can become scoped implementation work, so insight moves from reporting into product improvement.
New nodes and tasks close the gap
The platform can evolve by adding new tasks, workflows and nodes that address real passenger issues discovered across live journey behavior.
The product is not static after launch. It can inspect real journey behavior, identify what is missing and help create the code work required to expand coverage where it matters most.
Model training
Own the learning loop, not just the prompts
A key part of the flyjeeves operating model is that findings from real journeys can feed training and evaluation work. That creates airline-specific model behavior over time, with custom datasets, versioned model weights and promotion gates before runtime use.
Own training corpus
Production findings can become curated training data for airline-specific intents, policies, edge cases and execution patterns.
Custom model weights
The platform can train and version its own model weights instead of relying only on generic off-the-shelf model behavior.
Evaluate before promotion
Model candidates can be tested against quality gates before they are promoted into passenger-facing runtime paths.
Technology
Built on a stack that can support production-grade airline agents
The product story is agent-led, but the operating model matters. flyjeeves runs on proven infrastructure, runtime and data components that support scale, control and extensibility.
aws
AWS
Cloud infrastructure, network boundaries, storage and runtime services provide the operational base for production-grade airline agents.
Deepgram
Real-time speech recognition turns passenger voice into structured conversation input for low-latency agentic flows.
ElevenLabs
Text-to-speech capabilities support natural outbound voice responses across Twilio-powered passenger journey flows.
Google Gemma
Open model routing can support low-latency semantic tasks, focused classification and evaluation paths where quality thresholds are met.
Hugging Face
Training, evaluation and artifact management support the creation and promotion of airline-specific model variants.
Neo4j
Graph projections connect profiles, aliases, memory and interaction context so retrieval can follow passenger relationships instead of isolated records.
Node.js / Next.js / TypeScript
The development stack combines typed backend services, fast app routing and production-ready web surfaces across Studio and landing experiences.
OpenAI
High-quality language intelligence remains available for passenger-facing reasoning, booking assistance and tasks that need stronger model guarantees.
Postgres
Structured tenant data, booking records, memory items, traces and operational state sit on a robust relational foundation.
Redis
Low-latency session state, cache coordination and fast runtime lookups support responsive conversations where speed matters.
NestJSPrismaBullMQAPI GatewayCloudFrontRoute53SESLambdaECRSSM / Secrets
AWS architecture
How the platform is shaped in AWS
Based on the Terraform topology in the repo: Route53 and CloudFront in front, ALB and API Gateway for ingress, ECS for the app runtime, Aurora/Postgres and Redis for state, Neo4j on EC2 for graph memory, and SES plus Lambda for inbound email flows.
Edge and delivery
Route53
CloudFront
S3 landing
Ingress and APIs
ALB
API Gateway REST
API Gateway WS
Runtime on ECS
Studio
Agents
Intel
Squads
State and memory
Aurora / Postgres
Redis
Neo4j on EC2
Semantic ECS / EC2
Support services
SES inbound
Lambda
ECR
SSM / Secrets
Contact
Talk to us about your agentic airline roadmap
Share the airline journey you want to improve. We will confirm receipt by email and follow up with the right product context.
ACK email sent automatically
Minimal data, no public account required