The next generation of the Call Stream AI platform, and the most significant advancement we've ever shipped. Everything you already rely on keeps working — it now runs on a dramatically faster and more capable foundation.
Think of 3.0 the way you'd think about moving to a new generation of phone, or a new operating system. The tools you count on stay familiar. The engine underneath gets faster, smoother, and more capable.
Call handling, reservations, guest requests, routing, and live write-back to your PMS, POS, and CRM all carry forward unchanged.
A rebuilt engine powers every conversation — quicker, more natural, and more controllable property by property.
The new platform is designed for continuous innovation, so new capabilities reach your property faster than before.
The headline difference is how the platform feels in a live conversation — but the upgrades run across the board.
| Area | Before | Now · 3.0 |
|---|---|---|
| Conversation speed | ~1.2 second response time | Around 500 milliseconds — close to real time, with natural turn-taking |
| Voices | Previous v2 generation | All-new v3 voices built for emotion, pacing, and natural flow |
| Transfers | Basic call hand-off | Live warm transfers that carry full context to your staff |
| Call handling | General-purpose flows | Conversational pathways tuned per department |
| Continuity | Channels tracked separately | Conversational syncing across Voice, Message, Email, and Agents |
| Visibility | Standard call reporting | A Business Intelligence Command Center with ROI insight |
| Model layer | Frontier models only | Frontier plus self-hosted open-source models in the Secure Cascading LLM |
| Standard integrations | Custom build required | Weather, Time, Traffic, Directions included at no additional cost |
At about 500 milliseconds end-to-end — down from roughly 1.2 seconds — conversations feel immediate instead of mechanical, with smooth, natural interruption handling.
In hospitality that gap is the whole experience. A guest who has to wait for the system to catch up knows they're talking to a machine. At 500ms, the call simply feels like talking to a person — they interrupt, they change their mind mid-sentence, they get answered.
Voice is the first thing a guest judges, and the new generation is a major step up. The v3 voices are designed for genuine expression rather than a flat phone-tree read — with better prosody and timing so conversations flow.
Warmth on a welcome, calm on a complaint, brevity on a quick confirmation.
Delivery shifts with where the conversation actually is, not a fixed script read.
Rotating voices so a repeat caller never hears the identical greeting twice.
Custom voices tuned to your property — preserved across every language you serve.
Two upgrades make a real difference here.
When a human should take over, Call Stream hands off the full conversation. Your front desk already knows why the guest called, what was offered, and what they actually want. Guests never have to repeat themselves.
Routing is more controlled and tuned to how your property actually operates. Reservations behave like reservations. Service recovery behaves like service recovery. Neither borrows the other's script.
A guest calls to book. Two days later she texts to move her arrival. The morning of check-in she opens chat to ask about parking. That night, from abroad, she messages on WhatsApp.
Before 3.0, those were four related records your team had to reconcile. Now they're one thread — with far tighter conversational syncing across every product, so context follows the guest instead of resetting at each channel boundary.
Inbound & outbound, ~500ms, 40+ languages.
Two-way messaging with shared team inboxes.
Triage, draft, and resolve — assist or automate.
Guest chat out front, Teams & Slack behind.
Reporting tells you what happened. The new client portal is built to help you understand what to do next.
Instead of call counts, basic logs, and backward-looking snapshots you have to connect yourself, the Business Intelligence Command Center delivers real operating insight.
Reservations captured, upsells converted, and service demand met — reported as revenue influenced rather than calls handled.
Performance and outcomes rolled up across every property, so a group can compare locations on the same terms.
Demand signals, sentiment shifts, and missed-revenue alerts surfaced as things to act on — not a dashboard to interpret.
3.0 adds self-hosted open-source models to our Secure Cascading LLM. Routine, high-volume conversations are handled fast by models running inside our own infrastructure, while complex requests still escalate to the right frontier model for the job.
Conversations run on dedicated infrastructure inside our trust boundary — never routed through a third-party AI proxy. Your guest data stays yours, stays inside Call Stream AI, and is never used to train outside models.
This is the architectural point, not just a feature: Call Stream AI is a platform, not a proxy. Model selection, security, and auditability are all part of execution — which is why adding a self-hosted model layer reduces exposure rather than adding another vendor dependency. The platform remains model-flexible throughout.
Guests ask about conditions constantly — and until now, answering well meant a custom integration. In 3.0 these are standard on every subscription, including basic, at no additional cost.
Live conditions and forecast — for pool days, patio seating, tee times, and packing questions.
Local time and time-zone awareness — so an overseas guest gets accurate hours, not confusion.
Real-time conditions — for airport runs, dinner reservations, and realistic arrival windows.
Turn-by-turn guidance from wherever the guest is — spoken naturally, or texted to them mid-call.
These sit alongside the integrations you already run — Oracle Opera Infor HMS Maestro Agilysys HotSOS ALICE OpenTable Resy SevenRooms Book4Time Toast Stripe Slack Microsoft Teams — plus 3,000+ more via Pipedream MCP.
Moving to 3.0 doesn't put work on your team. Our implementation team manages the transition and tuning for your property.
Your configuration and integrations are carried onto the new platform foundation — agents, knowledge base, workflows, and write-back all intact.
Property-specific refinements are re-applied and validated against real call flows — including voice selection and mapping your departments onto conversational pathways.
As the new engine tunes in, anything you flag gets addressed quickly. Questions during rollout go straight to your implementation team — no tickets lost, no guesswork.
Faster conversations, new voices, richer insight — fully supported, and built for continuous release so new capabilities keep arriving.
Call Stream AI remains a fully managed service. There's no engineering lift on your side for 3.0 — no migration plan to own, no configuration to rebuild. Our team does the work and validates it before your property switches over.
A 30-minute walkthrough on 3.0 — the new v3 voices, the sub-500ms back-and-forth, the Command Center, and how your departments map onto conversational pathways.