Conversation intelligence

Sellable product

This capability is granted by an API key scoped to the `` product (product face). See the product reference below.

Conversation intelligence as a shared capability. POST /v1/analyze scores one transcript synchronously (sentiment/intent/summary/topics/entities/outcome via an LLM over provider-gateway, with a deterministic lexicon/keyword fallback for sentiment+intent). /v1/analyze/batch analyses many transcripts asynchronously (poll a job). /v1/analyses retrieves stored analyses (org-scoped). The store is OPTIONAL — with the DSN unset /v1/analyze still serves fully; the persisted-retrieval + batch paths return 503.

• Group: Voice & AI
• Contract: contracts/conversation-intelligence/v1/openapi.yaml
• Public base: https://api.vagarylabs.com (the consolidated API gateway — one host, per-brand sibling api.<zone>)
• Auth: a product API key (vgk_…) issued from the console — Authorization: Bearer vgk_…
• Product face (customer-keyed):
  • https://api.vagarylabs.com/product/v1/conversation-intelligence/analyze

Endpoints

Method	Path	Summary
POST	/v1/analyze	Analyse one transcript (sentiment/intent/summary/topics/entities/outcome), synchronous
GET	/v1/analyses	Retrieve stored analyses (org-scoped; optional session_ref filter)
POST	/v1/analyses/erase-subject	GDPR Art-17 erasure of a data subject's stored analyses (first-party admin only)
POST	/v1/analyze/batch	Start an asynchronous batch analysis of many transcripts
GET	/v1/jobs/{job_id}	Fetch a batch job's status + per-item results (org-scoped)
GET	/health	liveness + store/analyzer readiness
GET	/metrics	Prometheus exposition (fleet observability SDK — fail-open)

Schemas

EraseSubjectResult

matched is reported separately from deleted so a zero-match is visible to the compliance fan-out rather than reading as a successful erasure. jobs_swept is always false — see the endpoint's KNOWN RESIDUE note.

Field	Type	Description
success	boolean
capability	string
organization_id	string
disposition	string
selector	string
matched	integer	stored analyses belonging to those session refs
deleted	integer
session_refs	array	the refs actually applied (blank entries dropped)
dry_run	boolean
jobs_swept	boolean	always false — conversation_jobs has no session_ref column
jobs_swept_reason	string

Turn

one transcript turn.

Field	Type	Description
role	string	speaker role (user
content	string

AnalyzeRequest

Field	Type	Description
organization_id	string	product-tier end-customer tenant (I6); optional in v1 (persistence needs it; the analysis itself does not).
transcript	array	the conversation under analysis (≥1 turn).
analyses	array	which analyses to run (default = all six). Unknown values are ignored.
session_ref	string	the product's own call/session id — lets GET /v1/analyses fetch every analysis for one conversation.

Sentiment

5-band sentiment + signed score + confidence + an optional dominant emotion. source marks whether it came from the LLM or the deterministic fallback.

Field	Type	Description
label	string
score	number	signed sentiment in [-1, 1]
confidence	number
emotion	object
source	string

Intent

Field	Type	Description
label	string
confidence	number
source	string

Entity

Field	Type	Description
text	string
type	string	person

Outcome

Field	Type	Description
label	string
reason	string
source	string

Unavailable

an LLM-only analysis that could not run (LLM unreachable) — surfaced explicitly, never fabricated.

Field	Type	Description
available	boolean
reason	string

---

Generated by scripts/gen-capability-docs.py from contracts/conversation-intelligence/v1/openapi.yaml — the contract IS the source of truth; edit the contract, not this page.