Six AI agents.
One travel platform.

Most "AI travel" is a chatbot bolted onto a 2010 booking engine. Ours is six autonomous agents that talk to each other through MCP — booking, reconciliation, credits, traveler safety, and policy. Robert is the one your team will know by name.

Agent-to-agent via MCP Powered by Anthropic 99.97% uptime
6 AGENTS · LIVE
MCP · A2A
CORE · MCP Travel Code
Platform
robert
rateguard
credit-recovery
ocr-reader
duty-of-care
policy-builder

A chatbot answers questions. An agent does the work.

Most corporate travel platforms launched "AI assistants" in the last 18 months. They look like chat windows — type a question, get an answer. The pattern is the same: a conversational layer sitting on top of legacy infrastructure that hasn't changed.

An agent is different. An agent has tools. It can book a hotel, void it, refund the difference, file the expense, message the traveler, and update the GL — autonomously, with audit trails. It runs on a schedule, watches for events, escalates when uncertain.

Travel Code agents work together via the Model Context Protocol — RateGuard finds savings, Robert notifies your team, Credit Recovery files the difference. Same framework, agent-to-agent, no human in the middle for routine work.

Chatbot
Agent
What it does
Answers questions
Takes actions
When it runs
When you ask
Continuously, on triggers
State
Stateless or session-based
Persistent memory + audit trail
Talks to others
No
Yes — A2A via MCP
Failure mode
Wrong answer
Escalates to human, logs the why

Robert. Your travel and finance ops, on call.

Robert is the agent your travelers actually talk to. He books, rebooks, answers policy questions, generates reports, and respects every user's role and access level. He lives on every page of the platform.

User
Rebook me on the next flight if mine is delayed.
Robert
Watching BA178. Will auto-rebook on next available in-policy fare if delay > 60 min. Notification will go to your manager when triggered.
User
What's our spend on Marriott this quarter?
Robert
$47,820 across 23 stays. 18 in policy, 5 OOP with manager approval. Want the breakdown by traveler or by city?
User
Show me everyone in MENA right now.
Robert"
5 travelers · 3 cities — S. Kim (Dubai · Marriott), J. Roy (Riyadh · Four Seasons), A. Patel (in-flight QR-12 → DOH), +2 more. Crisis mode is armed.

What Robert can do

  • Book flights, hotels, rail, and cars within your policy
  • Rebook automatically on disruptions
  • Answer questions about spend, budget, travelers, policy, suppliers
  • Generate reports — XLSX, PDF, charts — in seconds
  • Take actions: cancel, refund, escalate, message
  • Respect role-based permissions: traveler, manager, director — each sees only their scope

What Robert won't do

  • Make bookings outside policy without escalation
  • Disclose other travelers' data to non-authorized users
  • Take irreversible actions without confirmation when amount > threshold
  • Pretend to know answers it can't verify (it links to the source)

Robert lives in chat, in-platform, in mobile apps, and via Slack/Telegram for teams that prefer messaging.

RateGuard. Watching every hotel booking. Forever.

You book at the best rate you can find. Then prices drop. RateGuard watches every hotel reservation after issuance. When the rate goes down, we void and rebook within the property's rules. You keep 20% (Premium) or 50% (Pro) of the savings.

Flights coming soon
Property · Stay
Original
Rebooked
Saved
Your share
Marriott Marquis · NYC · 3 nt
$1,284
$1,140
−$144
Park Hyatt · Tokyo · 2 nt
$940
$812
−$128
Four Seasons · Riyadh · 4 nt
$2,180
$1,980
−$200
Soho House · Berlin · 2 nt
$760
$688
−$72
Mandarin Oriental · London · 3 nt
$1,540
$1,388
−$152
Hyatt Regency · Chicago · 2 nt
$890
$806
−$84

Most travel platforms claim "best rate guaranteed" — and most don't follow through. RateGuard makes the claim auditable. Every hotel reservation has an entry. Every refund has a timestamp, an amount, and a destination. Your finance team can export the full log to CSV.

In practice, RateGuard captures meaningful savings on a sizable share of monitored hotel reservations — every rebook is logged with the original rate, the new rate, and the audit trail attached to the booking record.

Credit Recovery.
Never lose another airline credit.

Industry data: companies your size leak $80K–$200K per year on expired airline credits. They sit in airline portals, attached to former employees, and quietly disappear at the 12-month mark. Credit Recovery hunts them down, applies them, and reassigns them when employees leave.

01

Auto-track

Every credit from every airline in every traveler's account, in one dashboard. Live balances, expiration countdowns, transferability flags.

02

Auto-apply

At search time on the same airline, the credit is offered first. If the booking happens, the credit is applied automatically. No agent intervention.

03

Auto-reassign

When an employee leaves, transferable credits move to the company pool — eligible for any other traveler within the airline's rules. For non-transferable credits, the agent opens a case with the airline and negotiates an exception where possible — outcome isn't guaranteed, but we try.

Airline
Traveler
Amount
Expires
Status
BA
S. Kim
$840
Aug 2026
Available
UA
J. Roy
$420
Applied · trip TC-8842
LH
(former) M. Cole
$1,180
Mar 2027
Reassigned · pool
AA
(former) D. Vega
$320
Feb 2026
Expired
EK
A. Patel
$2,140
Nov 2026
Available

Three more on the team.

Policy Builder · Beta

Policy as configuration, not code

Upload your existing 30-page travel policy as a PDF. Our agent reads it, asks clarifying questions on ambiguous rules, and configures the platform automatically — cabin restrictions, hotel caps, approval workflows, group-based rules. Average setup: 4–14 minutes.

Setup walkthrough →
OCR-Reader · Live

Receipts, line by line

Most expense platforms read receipt totals. We read every line. Coffee, room service, parking, tip — each item categorized, each VAT split, each policy violation flagged. 99.2% line-level accuracy.

See itemized OCR →
Duty-of-Care · Live

Where every traveler is, in real time

Real-time traveler tracking with geo-queries. Geopolitical and weather alerts. Off-platform itinerary import via email forwarding. Crisis playbooks: rebook proposals, insurance claim drafts, evacuation contacts.

Visit duty of care →

Agent-to-agent work,
via our full MCP.

Each agent runs on the Model Context Protocol — Anthropic's open standard for agent communication. They share context, hand off tasks, and complete multi-step workflows without a human in the middle.

A hotel rate drops. RateGuard sees it, rebooks within property rules, and emits an event. Credit Recovery picks it up — checks if the original booking was paid with a credit, files a refund request. Robert sees both events, generates a one-line note for the traveler and an audit entry for finance. The whole sequence takes 90 seconds. No human touched it.

This is what "AI" should mean. Not a chat box that knows your policy — a team of specialists that own their part of the workflow and finish it.

RateGuardDetects rate drop on Marriott NYC reservation emits price.dropped
RateGuardVoids original booking, rebooks at new rate emits booking.rebooked
Credit-RecoveryListens for booking.rebooked verifies refund routing, files airline credit if applicable
OCR-ReaderUpdates expense entry with new amount, re-reconciles to GL emits expense.updated
RobertNotifies traveler, logs audit trail, posts to finance Slack channel done in 90s
MCP
Model Context Protocol — Anthropic open standard
A2A
Agent-to-agent · no human in routine workflows
Sonnet 4
Claude model backbone
99.97%
Uptime · 12-month rolling

What we don't pretend.

Marketing AI is easy. Shipping it responsibly is harder. Here's what we explicitly don't do — and why.

No agent makes irreversible high-value decisions alone.

Cancellations over $X, schedule changes that affect multiple travelers, fare class upgrades — these escalate to a human. Always. The AI proposes, the human approves. Configurable thresholds per company.

No agent shares data outside its permission scope.

A traveler can't query other travelers. A manager can't see beyond their department. A guest can't see anything. Role boundaries are enforced at the agent layer, not just the UI.

No agent fakes confidence.

If Robert isn't sure about the answer, he says so and links to the source. We trained against hallucinations more than we trained for charisma.

No agent replaces your finance team.

Agents handle the volume. Humans handle the judgment. We've seen vendors pitch "AI replaces your accountant" — that's not us. Your CFO still owns the close.

Questions, answered.

Architecture, safety, sandboxing, privacy — the seven we hear most. Talk to a product specialist if yours isn't here.

Is this all just GPT or Claude wrapped in a UI? +
No. Our agents use Claude Sonnet 4 as the language model, but they run on the Model Context Protocol — Anthropic's open standard for tool use and agent-to-agent communication. The LLM is the brain. We built the body, nervous system, memory, and the bridges between agents.
What happens if an AI agent makes a mistake? +
Three layers of protection: (1) high-stakes actions over a configurable threshold require human approval; (2) every action has a confidence score and escalates if below threshold; (3) every action is logged with a full audit trail — review, reverse, and analyze after the fact. Mistakes are contained, traceable, and learnable.
Can I see Robert before signing up? +
Yes. Book a demo and we'll give you a sandbox with your travel policy loaded. You talk to Robert, ask anything, see the responses, see the audit logs. No commitment. Most prospects spend 30–60 minutes in the sandbox before deciding.
What does RateGuard actually do? +
RateGuard watches every hotel reservation after issuance. When the price drops, we void and rebook within the property's rules. On Premium plan we return 20% of the savings; on Pro plan we return 50%. Flights are coming soon.
Are agents available in Slack and Telegram? +
Yes for Robert. Other agents are exposed via in-platform UI. Roadmap includes Slack and Teams native integrations for RateGuard alerts, Credit Recovery notifications, and Duty-of-Care updates.
What's the privacy model — does the AI train on my data? +
No. Customer data is isolated per tenant. Our LLM provider (Anthropic) does not train on inference data per their enterprise terms. Agent memory is scoped per company. Data residency is available in EU, US, and MENA.

See Robert and friends in your travel program.

30-min demo. We'll load your policy, show six agents in action, and send you a sandbox. No pitch.