Docs
Early Access hello@polariapi.com
Resources

Changelog

API updates, bug fixes, and infrastructure improvements. Breaking changes always ship with a new API version and 90 days notice.

Versioning policy. The current stable API version is v1. Non-breaking additions (new fields, new endpoints) ship continuously under v1. Breaking changes introduce a new version; the prior version remains supported for a minimum of 90 days with a migration guide provided.

Dates are approximate during the early access period. Semantic versioning is used for internal builds; external changelog entries track meaningful milestones.

November 2025
v1.0 · Early Access

Initial Early Access Release

The Polari API is now available to private beta customers. All four pipeline layers are deployed and operational.

Added
  • Layer 0 — Token Intelligence: quality scoring, 384-dim embeddings, semantic deduplication
  • Layer 1 — Semantic Analysis: named entity recognition (PERSON, ORG, GPE, LOC), sentence-level embeddings, sentiment scoring
  • Layer 2 — Story Clustering: two-stage cross-source clustering with confidence scoring
  • Layer 3 — Intelligence Graph: entity relationship networks, trend detection, constellation visualization
  • Core endpoints: /v1/articles, /v1/entities, /v1/clusters, /v1/relationships, /v1/graph, /v1/trends
  • Bearer token authentication via Authorization header
  • Per-key rate limiting with X-RateLimit-* response headers
Infrastructure
  • Containerized microservices per layer, independently deployable
  • PostgreSQL for article and entity persistence, ChromaDB for vector storage, Redis for coordination
  • Nginx reverse proxy with TLS termination
  • Processing cadence: Layer 0 real-time, Layer 1 15-min batches, Layer 2 30-min batches, Layer 3 hourly
November 2025
v1.0

Layer 2 — Clustering reliability improvements

Fixed
  • Singleton cluster problem: articles ingested within the same batch window could not discover each other as candidates, producing single-article clusters instead of merged stories
  • Added a re-clustering pass at the end of each batch cycle so intra-batch articles are evaluated against one another before results are committed
  • Backfill script deployed to retroactively re-cluster historical articles that were affected
Improved
  • Cluster confidence scores are now more conservative for small clusters (fewer than 3 sources), reducing false-positive groupings
November 2025
v1.0

Layer 0 — published_date field patch

Fixed
  • published_date was not being persisted correctly for a subset of ingested articles; affected articles returned null on the published_at field
  • Root cause: date parsing edge cases in RSS feed ingestion when the feed omitted or malformed the pubDate field
  • Historical articles with null published dates are being reprocessed incrementally
Changed
  • Ingestion pipeline now falls back to updated and dc:date feed fields before leaving published_date as null
  • Articles with unresolvable publish dates are still ingested but flagged with a date_confidence: "low" metadata field (visible in Layer 0 response)
November 2025
v1.0

Layer 1 — Entity extraction deployed

Named entity recognition is now live across all ingested articles. Entities are extracted per article and stored as typed, searchable records linked to their source articles.

Added
  • Entity types supported: PERSON, ORG, GPE, LOC, EVENT
  • Per-entity sentiment scoring alongside article-level sentiment
  • /v1/entities endpoint with filtering by type, mention count, and time range
  • /v1/entities/{id}/timeline and /v1/entities/{id}/sentiment sub-endpoints
Known Limitation
  • Entity variant normalization is in progress — surface forms such as "Strait of Hormuz" and "Hormuz" may appear as separate entities in the current release. A normalization pass is planned for an upcoming update.
Coming up

On the roadmap

The following are actively in development or planned for the near term. These are not commitments — dates and scope may change.

Planned
  • Entity normalization — collapsing surface-form variants into canonical entity records
  • Python and JavaScript SDKs
  • Webhook subscriptions (cluster.created, trend.detected, entity.spike)
  • Customer dashboard for API key management and usage monitoring
  • Historical data API — query articles and clusters beyond the rolling window
  • Extended language support — Spanish and French (targeted Q3 2026)