Data Models

Core entities and how they relate to each other.

Entity relationships

Slice (analysis template)
  └── Run (one execution of a slice)
        ├── Result (structured output conforming to slice's outputSchema)
        └── Evidence (quotes from transcripts and CRM records)

Call Type (category of call, e.g. "Discovery Call")
  └── Post (AI-generated output from a processed call)
        ├── Summary (high-level call recap)
        └── Scorecard (element-by-element performance evaluation)

Slices

A slice is a pre-built AI analysis template. Each slice has a business objective (pipeline, deals, expand, or product) and evaluates conversations against a set of elements (e.g., “budget confirmation”, “competitive mention”).

Field	Description
key	Unique identifier (e.g., high_intent_deals)
title	Human-readable name
objective	Business category: pipeline, deals, expand, product
elementKeys	Elements this slice evaluates
outputSchema	JSON schema defining the structured output
preferredModel	LLM model used for analysis
evidenceGrouping	How evidence is organized: ByElement or ByAccount

Runs

A run is one execution of a slice. Runs are asynchronous — you start one and poll for completion.

Field	Description
id	Unique run identifier
status	pending, running, completed, failed
phase	Execution phase: queued → gathering_evidence → analyzing_calls → synthesizing → completed
matchedCallCount	Number of calls matching the filters
callsAnalyzed	Number of calls analyzed so far
result	Structured output (null until completed)
suggestedPollAfterSeconds	Recommended polling interval

Evidence

Evidence items are the supporting data behind a run’s results. Each item is a quote from a transcript or CRM record, grouped by element.

Field	Description
quote	Extracted text from the source
source.type	Source type (e.g., transcript, crm_deal)
source.id	Source identifier (e.g., file:98123)
source.label	Human-readable label (e.g., “Acme Corp — Discovery Call”)
confidence	Confidence score (0.0 to 1.0)

Evidence is capped at 20 items per element per run.

Posts

Posts are AI-generated outputs from processed calls. Two types:

Type	Description
Summary	High-level call recap with key topics, decisions, and next steps
Scorecard	Element-by-element performance evaluation with scores and coaching

Field	Description
postId	Post identifier
fileId	Underlying call identifier (use to deduplicate summary + scorecard for same call)
templateType	summary or scorecard
title	Call title
result	Structured post payload
callOwnerEmail	Who ran the call
companyDomain	External participant’s company

Call Types

Call types categorize your conversations (e.g., “Discovery Call”, “Demo”, “QBR”).

Field	Description
key	Unique identifier (e.g., discovery_call)
label	Human-readable name
description	What this call type represents
numCalls	Number of calls with this type

Elements

Elements are the specific things a slice evaluates (e.g., “budget confirmation”, “competitive mention”).

Field	Description
key	Unique identifier (e.g., budget_confirmation)
label	Human-readable name
description	What this element detects

Scoping

Entity	Scope	Meaning
Slices, Runs, Posts, Call Types, Elements	Team	Visible to the API key’s team only
CRM Deals, Companies, Stages, Fields	Org	Shared across teams (CRM is org-level)