> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zeroset.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Changelog

> Public release notes for the Nebula platform. We publish notable product, security, and reliability changes here as they ship.

The canonical record of changes to the Nebula platform.

<Update label="July 4, 2026">
  ## Changed

  * **Search memory classification field.** Search results now return `memory_class` for semantic and procedural memory classification.
</Update>

<Update label="May 23, 2026">
  ## Security & Compliance

  * **Sanctions-compliant access controls.** Access from OFAC-sanctioned jurisdictions is blocked at the network edge, with audit-grade match logging.

  ## New

  * **Self-serve enterprise downloads.** Eligible customers retrieve signed Nebula bundles, signatures, and SBOMs directly, replacing email delivery.
</Update>

<Update label="May 20, 2026">
  ## New

  * **Microsoft 365 connectors.** First-party ingest for OneDrive, SharePoint, Outlook, and Teams. Joins Google Workspace, Slack, and Notion as supported sources.

  ## Improved

  * **Cursor-stable pagination.** List endpoints move to keyset pagination — consistent under concurrent writes, faster at scale.
</Update>

<Update label="May 19, 2026">
  ## New

  * **In-house SDK generator.** Python and Node.js SDKs now generate from our own pipeline, replacing Stainless. Ships with a canonical error envelope, and new endpoints land in the SDKs within minutes of release.

  ## Changed

  * **`zeroset.com` is now our primary domain.** `trynebula.ai` continues to work as a transparent alias, so existing integrations, allowlists, and bookmarks keep working without changes.
</Update>

<Update label="May 16, 2026">
  ## Security & Compliance

  * **Signed releases.** Every container image ships with a signed SBOM and provenance attestation, gated on a CVE scan. Available on request for security review and procurement.

  ## New

  * **Per-customer release notifications.** Enterprise customers subscribe specific recipients with per-customer delivery preferences.
</Update>

<Update label="May 15, 2026">
  ## New

  * **On-premise.** Nebula ships as a `docker-compose` bundle for evaluation and a production Helm chart for self-hosted Kubernetes.

  ## Improved

  * **Two-stage cold-collection admission.** When a cold collection takes its first query after a quiet period, foundation load and cache-fill now run as separate stages — queries land far sooner.
</Update>

<Update label="May 11, 2026">
  ## Security & Compliance

  * **Leaked-password protection.** Registration checks new passwords against the Have-I-Been-Pwned database and rejects known-compromised credentials.
  * **Hardened network perimeter.** Customer-facing services are reachable only through our managed edge; direct-to-origin paths are closed.
</Update>

<Update label="April 27, 2026">
  ## Security & Compliance

  * **RS256 access tokens.** All API access tokens move to RS256 signatures end-to-end.
  * **Centralized audit logging.** Account and infrastructure audit logs consolidate into a single retained store for SOC 2 and customer audit windows.
  * **High-availability database tier.** Every Nebula deployment runs with a warm cross-AZ replica for automatic failover.

  ## Improved

  * **Zero-downtime deploys.** Rolling restarts drain gracefully; in-flight requests are no longer dropped.
</Update>

<Update label="April 26, 2026">
  ## Security & Compliance

  * **Encryption at rest.** All workload and customer-facing storage is encrypted at rest with managed keys.

  ## New

  * **Live document sync.** Slack, Google Drive, Gmail, and Notion sources update through change webhooks instead of interval polling. New content reaches memories within seconds of creation.
</Update>

<Update label="April 23, 2026">
  ## New

  * **Contract-first public API.** Every public endpoint derives from a published OpenAPI contract — the single source of truth for the API surface.

  ## Improved

  * **Batched entity merges.** Per-group merges and deduplication runs collapse into a single relationship scan, eliminating the N+1 path that previously dominated dedup latency on collections with thousands of related entities.
</Update>

<Update label="April 20, 2026">
  ## Improved

  * **Retrieval, rewritten.** Ground-up rewrite of the query engine with single-pass evaluation and filter pushdown. Latency and quality gains across the board.
  * **Stronger recall on time-anchored questions.** Questions about specific events, moments, or "when" something happened return more relevant results.
</Update>

<Update label="April 17, 2026">
  ## New

  * **Selective OCR for PDF ingestion.** PDFs route page-by-page through OCR only when needed, on a unified streaming pipeline. Faster ingest and higher fidelity on mixed text/image documents.

  ## Improved

  * **Persistent warm-search state.** Search state for active collections stays warm continuously instead of cold-starting on demand, eliminating most cold-start penalties before they happen.
</Update>

<Update label="April 16, 2026">
  ## Improved

  * **Resilient embeddings.** The embedding pipeline absorbs upstream rate limits transparently and keeps ingestion infrastructure warm. Transient failures on large uploads largely disappear.
</Update>