Skip to main content
For AI agents: a documentation index is available at https://docs.parallel.ai/llms.txt. The full text of all docs is at https://docs.parallel.ai/llms-full.txt. You may also fetch any page as Markdown by appending .md to its URL or sending Accept: text/markdown.
This guide enumerates the contract differences between the Alpha Monitor API (/v1alpha/monitors) and the GA version (/v1/monitors), and outlines the steps required to migrate.
All ongoing development targets V1. The Alpha endpoints remain reachable but receive no new features:
  • Capabilities introduced after Alpha — snapshot monitors, structured output with basis, advanced_settings.location, and processor selection — are V1-only.
  • The Python and TypeScript SDKs expose typed bindings (client.monitor.*) only for V1. Alpha is reachable solely via the low-level HTTP client (client.post("/v1alpha/monitors", ...)).
  • The Parallel CLI targets V1 endpoints exclusively.

Highlights

  • Required type discriminant"event_stream" (default Alpha behavior) or new "snapshot"; determines the settings shape. See Snapshot Quickstart.
  • Nested settings / advanced_settingsquery, output_schema, include_backfill move under settings; source_policy and the new ISO 3166-1 location move under settings.advanced_settings. See Advanced Settings.
  • processor selection — Top-level "lite" (default) or "base". base increases recall and breadth for harder queries, at higher cost. See Monitor Quickstart.
  • Endpoint renames — Update → POST /{id}/update; Cancel → POST /{id}/cancel. New POST /{id}/trigger enqueues off-schedule one-off runs.
  • Unified events endpointGET /events supersedes both Alpha endpoints with cursor pagination and an optional event_group_id filter. See Events.
  • Restructured event payload — Stable event_id, event_type discriminator, and typed output with basis (citations, reasoning, confidence) replace the deprecated output string and the result object. See Events and Research Basis.
  • V1-only SDKs and CLI — Typed client.monitor.* bindings and the Parallel CLI target V1 exclusively.

Overview of Changes

Endpoints

Create Request

Response

Events

V1 unifies the Alpha MonitorEventDetail shape into a single typed event with a stable event_id, a structured output object, and a basis array carrying per-field citations, reasoning, and confidence. Webhook event types (monitor.event.detected, monitor.execution.completed, monitor.execution.failed) are unchanged. The webhook payload still wraps an event_group_id to be resolved against the events endpoint.

SDK and CLI Surface

V1 exposes typed bindings in both the Python and TypeScript SDKs and is the only version supported by the Parallel CLI. Alpha has no typed or CLI surface — it is reachable only via the low-level HTTP client.

Migration Example

Before (Alpha)

After (V1)

Migration Checklist

Required changes

  • Update the base path from /v1alpha/monitors to /v1/monitors.
  • Add the type discriminant ("event_stream" or "snapshot") to every CreateMonitorRequest.
  • Move query, output_schema, and include_backfill from top-level into settings.
  • Move source_policy from top-level into settings.advanced_settings.source_policy.
  • Migrate Update calls from POST /{id} to POST /{id}/update.
  • Migrate Cancel calls from DELETE /{id} to POST /{id}/cancel.
  • Replace GET /{id}/event_groups/{event_group_id} with GET /{id}/events?event_group_id=....
  • Replace the lookback_period query parameter with cursor-based pagination (cursor, limit).
  • Update the status enum check from "canceled" to "cancelled" (double l).
  • Replace reads of result.content and source_urls with output.content and output.basis[].citations[].url.
  • Drop the deprecated top-level string output field on event records.

Optional enhancements

  • Set processor: "base" for harder queries that need higher recall and breadth.
  • Set settings.advanced_settings.location to scope retrieval to a single country.
  • Replace low-level client.post(...) calls with client.monitor.* SDK bindings.
  • Use event_id for idempotent client-side dedup across pagination and webhook retries.
  • Consume output.basis for per-field citations, reasoning, and confidence.
  • Issue POST /{id}/trigger for off-schedule one-off executions.
  • Pass include_completions=true to enumerate executions that produced no events (useful for audit traces).
  • Adopt type=snapshot for field-level diffing of structured Task Run outputs (see Snapshot Quickstart).

Additional Resources

Questions? Contact support@parallel.ai.