Skip to main content
Parable turns ambient conversations into speaker-attributed transcripts. The egress API gives external systems programmatic access to every finished transcript in a workspace — for ingestion into a knowledge base, search index, CRM, or anything else that wants to know what was said. Every capture (one recorded conversation) that finishes processing with a non-empty transcript enters an append-only stream. You consume it one of two ways, or both:

Pull API

Cursor-paginated REST endpoints. Cron-poll “what’s new” with a single stored cursor — the first un-cursored pull returns full history.

Webhook push

Parable POSTs each finished transcript to your HTTPS endpoint within about a minute, signed with HMAC, with retries.

​
Basics

  • Base URL: https://api.parable.so
  • Auth: Authorization: Bearer <api key> — see Authentication
  • Latency: new captures appear in the stream within a few minutes of the conversation ending (processing time plus a short sweep interval)
  • Format: JSON. Timestamps are ISO-8601 UTC instants.

​
Stream semantics

  • Append-only. A capture enters the stream exactly once, after processing completes. There are no updates or deletions in the stream; late metadata fixes (for example a speaker rename) are visible by re-fetching GET /api/v1/captures/:id.
  • Non-empty only. Captures that processed to zero transcript segments never appear.
  • Ordering is by processing completion, not conversation start time — a capture that needed retries can appear later in the stream than a newer conversation. Use started_at / ended_at for display ordering.