Concierge

A 24/7 concierge that knows your business and can act on it.

Your customers and your team need real answers, on every channel, around the clock

Concierge is a 24/7 concierge agent that actually knows your business. It grounds every answer in the knowledge bases, policies, and records you give it, speaks and types across phone, SMS, Microsoft Teams, and embedded web chat, and carries a wide toolkit so it can do more than talk: look up an order, open a service request, check a warranty, book the next available appointment, confirm an identity, and hand off cleanly to a person when something genuinely needs one. The agent is the unit; channels are flags. The same agent that answers your main line responds to an @mention in Teams, with the same knowledge, the same staff directory, the same custom tools, and the same brand voice.

How it differs

A general-purpose chat surface is a chat surface; you bring the prompts, the data, the integrations. Concierge ships the runtime: a typed Agent record with voice tuning, channel flags, knowledge-base bindings, staff-directory access, a custom-tool framework, multi-provider voice failover, and an admin-controllable prompt library. Every conversation lives in a database row with full audit history.

A platform-bound assistant, the kind baked into a single vendor’s suite, cannot leave that platform. Concierge agents are channel-agnostic by design: the same agent ships into a customer’s Teams tenant, answers the phone, and texts customers on the side.

Legacy IVR is a tree of recorded prompts and DTMF digits. Modern voice-AI products bolt a language model onto the front but the runtime is still a phone-only black box. Concierge runs on the realtime voice WebSocket protocol that powers the major realtime model APIs: sub-second turn-taking, server-side voice-activity detection, interruption handling, native function calling mid-conversation. The same agent also handles text.

Who it’s for

Executives and operations leaders who want a knowledgeable agent supporting customers and staff around the clock, on a runtime the team can actually configure.

Internal employees in Microsoft Teams who want an @mentionable assistant for the operational systems the company runs on.

Client-facing teams deploying the same agent across the main switchboard, the customer portal chat widget, and SMS after-hours support.

Support teams running tier-one triage through the agent: identification, consent confirmation, knowledge-base lookups, scheduled callbacks, clean transfers.

Sales reps in the field texting or calling the company’s agent for quick lookups without opening a laptop.

What it does

Per-agent bot identities provision into each customer’s Microsoft 365 tenant under their own name. Multiple agents in one tenant means multiple bots, with strict per-agent memory isolation enforced at runtime.

Scoped memory is the security story. Every read and every write goes through a ScopeGuard that hard-rejects any attempt to read a turn belonging to a different conversation, and writes an audit row when it happens. Message content is encrypted at rest. Rolling summaries fire every fifty turns to keep prompt size bounded. Semantic search inside a single long conversation lets an agent answer “what did we agree on last Tuesday” without leaking other customers’ conversations.

Realtime voice with multi-provider failover. Capability-based routing means Concierge can swap from one realtime voice provider to another transparently when one degrades. The audio pipeline handles format conversion and voice-activity detection for natural turn-taking.

Custom tools are user-defined webhooks the agent can call mid-conversation. Look up an order in a customer’s API. Create a service request in their ITSM. Fetch the next available appointment. Built-in tools cover the eighty percent (knowledge search, staff search, callbacks, transfers, identity confirmation); custom tools cover the rest.

How it fits the ecosystem

Concierge reads from anywhere it has been bound. OS knowledge bases ground the agent’s answers. Orbit caller memory carries the relationship across channels and calls. Dispatch service requests can be created directly from a Concierge voice call. Foundry policies update the agent’s grounding by morning. Aria voice profiles drive how the agent sounds.

Where this product is in development

Concierge is feature-complete and in final validation against customer workloads. What is live today is in active use; production deployment timing and the next capabilities are firming up with the beta cohort.

[Join the Beta Cohort]

Beta products are feature-complete and in final validation against customer workloads. Early access available; production deployment timing is on the roadmap.