Curiosity

Release overview — January 2026

The Curiosity Team
•

We kicked off 2026 by making Curiosity more flexible about where AI runs and more capable with the messy real-world content inside your workspace. Here are the January highlights.

Bring your own model

The big theme this month: you choose the model, and where it runs.

  • Ollama support. Connect a local or self-hosted model server and stream responses just like any cloud provider — keep everything on your own hardware.
  • Provider & model picker in the chat box. Switch models on the fly, mid-conversation, without leaving the chat.
  • Assistant template, separate from provider. Pick the behaviour (the template) independently from the model that powers it.
"Why it matters"

Different questions deserve different models — a fast local model for quick lookups, a larger one for hard reasoning. January makes that a one-click choice instead of a configuration change.

Media and documents that just work

  • In-browser media streaming. Audio and video now stream and seek directly in the browser, with scrubbing support. New files are stored in a seek-friendly form, and S3-backed storage serves byte-range requests.
  • Better document rendering. Improved Word rendering and fidelity, custom font installation for document rendering, and PostScript font embedding.

Search results as tables

Search results can now render as a table layout with columns, so structured data reads like the records it actually is — not a wall of cards.

For administrators

  • WebDAV access to workspace definitions: browse and edit them over a standard /dav/ mount, toggled on or off from workspace settings.
  • Edit list, dictionary and table fields directly in the node editor.
  • Import code from another endpoint via a searchable endpoint list.
  • Log search with a text filter plus a severity-level selector, and a richer index list with a tag column for custom-code indexes.
  • Endpoints can read incoming HTTP request headers, and edge timestamps are settable when writing to the graph.

The little things

  • Paste images straight into AI chat from the clipboard — multiple pastes stay separate.
  • Tool names show up as readable, spaced words while a chat is working.
  • One-archive export/import for AI tools and entity post-processing.
  • Bulk entity toggling now saves correctly.

This overview covers builds v26.1.63412 → v26.1.63776. For the full, build-by-build detail see the changelog.