In April, semantic search grew up, storage got simpler under the hood, and we started a top-to-bottom rethink of the administration experience.

Embeddings, your way

• External embedding providers. Generate sentence embeddings from providers like OpenAI, Anthropic and Cohere. When you create an embeddings index, pick the external model type and supply the provider details — Curiosity uses it for similarity and semantic search.
• Smart caching. Already-encoded text is cached so unchanged content is never re-sent, keeping cost and latency down.

A fluent similarity engine

A new "find items like this" engine builds similarity queries directly against the graph — combining multiple scored signals and filters, with rule explanations that show why each result matched.

Storage, consolidated

• One database. Workspaces move to unified on-disk storage, automatically upgrading to a single consolidated database. Existing deployments migrate on first startup — no manual steps.
• On-demand storage diagnostics at startup, and periodic cleanup of expired cached sign-in data.

A redesigned admin experience

April kicked off a redesign that would continue through the spring:

• Administration screens rebuilt around a consistent segmented-tab layout.
• Users and Teams combined into a single Access Management view.
• Recent and pinned admin settings stay at the top of the sidebar.
• Serve different file sets per domain, with per-folder access rules.
• New graph operations to create users and teams, manage membership, and restrict access programmatically.

The little things

• The schema view now defaults to a visual graph tab for a clearer overview.
• PDF search highlighting is configurable between precise and extended modes.
• The SDK gained a method (and a public endpoint) to retrieve a node type's schema — its name, key field, and typed fields.

---

This overview covers builds v26.4.65832 → v26.4.66204. For the full, build-by-build detail see the changelog.