Multiple business phone numbers need a routing policy that stays correct when callers share a number, businesses share infrastructure, or an agent release changes. A successful room connection alone does not prove that the right business answered.
The current LiveKit dispatch-rule guide describes reusable SIP dispatch rules and agent selection through room configuration. Treat those as infrastructure building blocks. Your application still needs an approved mapping from the called number and authorized trunk context to business configuration.
Keep number routing separate from caller identity
The called business number selects a service context; the caller's number is a customer-search signal. Mixing them can route a returning caller to the wrong client simply because their number appears in several records.
Maintain a versioned routing register with business, accepted inbound context, intended agent, configuration version, and exception owner. Review unknown numbers explicitly instead of falling back to the first configured client. Avoid sensitive identifiers in room names and ordinary logs unless required by the approved design.
Test a fictional two-business deployment
A plumbing company and a salon share an engineering platform but have different phone numbers, policies, and calendars. A caller contacts both from the same mobile. Each call must load the correct business facts and permitted actions.
Use a separate private call context for each request. The documentation warns that removing individual-room randomness can place callers sharing a number together. Do not assume that a business-specific room name is appropriate for private customer conversations.
| Dispatch fixture | Required result |
|---|---|
| Plumbing number called | Plumbing policy and action scope |
| Salon number called | Salon policy and calendar |
| Same caller on both numbers | No shared conversation or customer state |
| Unknown called number | Approved rejection or review route |
| Agent unavailable | Visible failure and business fallback |
Verify business isolation after connection
Inspect room, agent, selected configuration, remote action destination, and staff notification. A correct greeting can conceal a wrong credential mapping. Negative tests should try another business's record identifier and confirm that the backend refuses the action.
A shared agent implementation reduces repeated maintenance while increasing the importance of explicit configuration and authorization. Separate agents can simplify operational boundaries but add release coordination. Neither choice removes the need for per-call privacy and destination checks.
Keep dispatch rules as maintained configuration rather than creating unnecessary new infrastructure for every caller. Review agency isolation for the business-data boundary beyond telephony.
Questions about multi-number dispatch
Can one implementation serve several businesses?
Yes as a proposed architecture when routing, configuration, data, and actions remain explicitly scoped and verified.
Does a separate room prove customer isolation?
It helps separate media, but application storage, retrieval, credentials, and notifications still need authorization boundaries.
What should a routing change record contain?
Affected numbers, authorized context, target agent, configuration version, acceptance fixtures, fallback, and rollback owner. Verify the actual call path after the change.