Once your assistant is live, this chapter is where you'll spend your time: every call and chat with its transcript and summary, recordings, an AI quality reviewer, a persistent memory of who's called before, a log of every third-party call your integrations made, and a full history of every configuration change with one-call restore.

Each completed call or chat is stored with its full transcript, caller metadata, sentiment scores and analysis.

Available to any signed-in member of the account that owns the app.

Query params: limit (default 50), offset (pagination), mode. Pagination is offset-based.

Response (200):

Available to any signed-in member of the account that owns the app. Returns the full transcript + metadata:

Available to any signed-in member of the account that owns the app. Returns a secure, time-limited download link (15-minute expiry), or 404 if no recording exists.

Response (200): { "url": "https://...", "expiresIn": 900, "contentLength": 123456, "format": "mp3" } (plus filteredUrl when a noise-filtered version exists)

Used for web-chat continue-session flows. Rare in API integrations.

Imports historical conversations — useful when migrating from another system so your history lives in one place.

Body: { "chatId": "...", "userIdentifier": "...", "messages": [ { "role": "user"|"ai", "content": "..." } ], "source": "..." } — userIdentifier and a non-empty messages[] are required.

Available to any signed-in member of the account that owns the app. A per-app aggregate snapshot — daily call volume, conversion rate, dropped-call rate, abandoned/mid-drop breakdown, recent-issues count. One call instead of N queries for rendering an app overview.

The platform can run an AI reviewer over each completed call to assess assistant quality — flagging issues with severity tags (info / warn / critical) and suggesting prompt improvements. Review cadence and model are configured per app; the review data itself is what you see in the dashboard's quality surfaces.

Requires enable_reviews.

Body:

• reviewCadence — off (default), per-call, daily, weekly. Default off means no reviews until explicitly enabled per app.
• reviewModel — sonnet (default), opus, haiku

Your assistant can remember the people who contact it. When enabled, the platform builds a persistent caller profile per phone number per app: identity (name, organisation), a rolling AI-written summary refreshed each call, structured traits with confidence scores, and open issues — thread-style records that persist across calls until resolved. A returning caller gets an assistant that already knows the context.

Scope: profiles are per (appId, phoneNumber). Two callers with the same phone but different apps are distinct records. Do not assume cross-app continuity.

Requires view_caller_profiles. Query params: ?q=<substring>&limit=200&offset=0&includeArchived=false (q is a substring match across the profile; there is no dedicated phone/since filter). Returns the list of contacts (phone, name, last-called-at, callCount).

Requires view_caller_profiles. Returns the full profile — identity, rolling summary, traits with confidence, open issues, complete call history references, operator notes.

Requires edit_caller_profiles. Human-side updates — correct an extracted name, adjust a trait, add a note. Body shape mirrors the GET response with optional fields. AI-extracted fields can be overridden; future profile updates respect high-confidence human edits.

Requires delete_caller_profile (owner default). Supports GDPR Article 17: removes the profile, the caller's issue records and analytics keyed to them. Transcripts of past calls remain in your conversation records (needed for service-delivery and accounting) but their linkage to the deleted profile is severed. An audit entry records who, when, why. Irreversible.

Requires view_caller_profiles. Bulk export across all profiles — supports GDPR Article 20 (data portability). Returns standard CSV with one row per contact + their analytics columns. Issue detail comes via the issues CSV (below). (There is no single-profile .csv endpoint — use the per-contact JSON detail route for one caller.)

Requires view_caller_profiles. Call volume, common topics, conversion rates, average call length, etc. for one caller.

Requires view_caller_profiles. Returns only contacts matching the target's phone within the same {appId} — duplicates are strictly per-app.

Requires edit_caller_profiles. Strictly per-app: any sources[] entry whose contact belongs to a different app than the URL {appId} is refused with a per-source error in the response errors[] array (in-scope sources in the same payload still process normally). The target's own app is validated up front; a target-app mismatch returns 400 for the whole request. Cross-app merging is not performed regardless of caller role.

Issues are persistent records tied to a caller; they have a status (open / monitoring / completed), kind, summary, severity, lastObservedAt, and notes. The assistant sees open issues on the next call from that number (when profile-apply is on), so "the engineer never called back" doesn't get forgotten between calls.

Requires view_caller_profiles. The app-wide /issues form returns the consolidated issues view — typically used for daily triage. Query params: ?status=open&severity=critical&since=....

Requires edit_caller_profiles. Update an issue's status and notes — close it out, mark it monitoring, or add context for the next person.

Every outbound HTTP call your app makes to a third party during a conversation is recorded as one row in the call log — a flight recorder for the integrations wired into an app, so you can see what the app called, when, and whether it succeeded.

Requires view_call_logs.

Query params:

• limit (default 50, max 200)
• offset (default 0)
• outcome — optional exact-match filter, one of ok, http_error, network_error, timeout

Rows are returned newest-first (descending timestamp).

Response (200):

Row fields:

What gets logged. The source field is prefixed by the kind of caller:

Reliability & retention:

• Fire-and-forget. Log writes never block or fail a conversation. If a write is slow or errors, the call path proceeds unaffected — a missing log row never means the underlying call failed.
• Query-stripped URLs. Only the path portion of each URL is stored (truncated at 500 chars); anything after ? or # is discarded before write, so credentials passed as query params are not retained.
• 30-day retention. Rows are permanently deleted 30 days after their timestamp. There is no grace window — past 30 days the row is gone.

Every configuration change to an app is recorded, so you can always answer "what changed, and when?" — and undo it.

Available to any signed-in member of the account that owns the app.

List response (200):

The single-version form returns the full attribute set as it was at that point.

Requires edit_settings. Restores the app's configuration to that snapshot. Like any configuration change, the restore is a draft until you take the assistant live.