A chat conversation ID identifies a conversation participant; it does not prove who signed in to your app. Start with a working iOS or Android chat screen and agree on which conversations may return after Help closes. The app developer implements that boundary; the CX owner checks the customer experience.

The supplied public-widget recipe answers public support questions. Adding a customer ID, display name or launch payload does not make it an authenticated account channel. If the agent reads protected records or performs account actions, complete the trusted-backend handoff below before enabling those operations.

Set retention and reset behavior

1

Choose where continuity should end

Use the widget persistence options to select a deliberate policy:

The examples use sessionStorage with no supplied user ID. iOS creates a nonpersistent website store for each view; Android recreates the WebView when its Help Activity is recreated. Neither sample promises return across an app restart. Choose persistent continuity only after you implement its lifecycle.

2

Close the old conversation on identity changes

Have the developer block further sends and remove the old chat screen when logout starts or the signed-in account changes. Revoke the app/backend session and discard its cached conversation mapping before opening the next screen. Late replies from the old instance must never appear in the new one.

For the iOS sample, destroy the old view and release its nonpersistent data store. For the Android sample, finish the Help Activity so it destroys its view; open a fresh Activity only after the identity transition completes. If you add cookies or persistent web storage, the developer must clear/isolate that data and await completion before loading another account. Avoid clearing unrelated WebView data globally as an incidental logout action.

Do not use one hardcoded userID for every installation or account. A custom ID must follow the intended customer/conversation mapping, but it still does not authorize data access. Widget destroy() only unmounts the UI; it is not a server logout or transcript-deletion operation.

3

Gate protected data on a trusted backend

If account operations are required, ask the backend developer to deliver an app-authenticated conversation service with these properties before adding protected tools:

  • Verify the app’s existing login on every request and derive account/tenant identity from that verified session, not the chat payload.
  • Keep Voiceflow project credentials and runtime session keys on the server. Map the signed-in principal to the permitted conversation and Environment.
  • Allow only the intended customer actions. Reject client-supplied protected variables/state/record IDs, and reconstruct authorized context on every turn and resume.
  • Authorize every data read/write again at the business-system boundary. Stop stale sessions on logout/account changes and prevent late responses crossing accounts.

The app then uses your service and its own conversation UI; this is a different architecture from the public widget example. The backend developer uses Authentication, Sessions, Sending actions and Handling responses for the exact contracts. Creating a runtime session on every foreground transition is not resume: a new session for the same runtime user can supersede an existing one.

Have the developer provide the service endpoint, app authentication integration, logout behavior and passing two-account tests. A launch-only identity variable, private page URL or instruction telling the agent to respect an account ID does not replace these deliverables. This guide defines the handoff; it does not supply a ready-made authentication service.

Test conversation separation

  1. In test account A, open Help and send a distinctive public message. Close/reopen Help and check the selected retention policy.
  2. Sign out, then open Help anonymously or as test account B. A’s visible messages must be absent. Repeat while a reply to A is still in flight.
  3. Restart the app and repeat with the same account. Confirm the documented continuity or fresh-start behavior on both platforms.
  4. If you added protected operations, ask for B’s record while signed in as A. Also alter an account/record ID in a later request and a resume request. The backend must deny access or replace the input with A’s authorized mapping.
  5. After logout, retry an old app/backend request. It must fail authorization. Do not call this boundary complete from a changed UI alone.

Another account's messages appear

Check retained views, storage and reused identifiers.

Stop the rollout. Trace whether the old view survived, storage was reused, cleanup finished after the new load, or all accounts shared a custom user ID. Hide/remove the old view immediately and finish the identity transition before creating the next one. Clearing only displayed text leaves the underlying conversation boundary unresolved.

The backend trusts a launch value

Move authorization to the verified app session and operation boundary.

A client can change context and later runtime inputs. Verify the server-side identity mapping on every request, including later turns and resume. Keep runtime credentials behind an allowlisting proxy when the client must not alter protected state, and authorize each business operation. Re-run the later-turn override test.

Logout is expected to delete transcripts

Treat local cleanup, authorization revocation and retention separately.

Destroying a WebView, clearing local storage or revoking an app session does not establish deletion from Voiceflow or a connected business system. Apply your separate transcript/data-retention process; do not label a UI reset as deletion.

Use the up and down arrow keys to select a result, Enter to open it, and Escape to close the search.