Skip to content
← All Posts
Build notes

Prism 1.10 answers from whole passages

Prism 1.10 reads the full passage behind each source when it answers a question, instead of a short excerpt. Ask has room for about four times as much of your documents per answer, and follow-up questions keep their subject. The current version is 1.10.2.

Answers with more to read

Before 1.10, Ask worked from short excerpts of each matching document. An excerpt could miss the sentence that held the answer, and Prism would sometimes decline a question your documents did answer. Ask now reads whole passages within a budget sized to the model. In testing it found and cited the right document more often and stopped declining questions the documents answer.

The summary above search results now uses the same engine and is written from the full passages of the results on screen. It is still included on every plan.

Follow-ups that stay on topic

A follow-up like “what about the costs?” is rewritten into a complete question before Prism searches, so it finds documents about the subject of the conversation rather than costs in general. If the model is slow to rewrite it, Prism falls back to the earlier topic.

Search also sends your whole question to semantic search instead of a list of keywords, so “how do I set up backups” matches on meaning. Semantic results show the part of the document that matched.

Safer rebuilds

1.10.1 keeps the current index searchable while its replacement builds. An interrupted rebuild recovers, and BookStack syncs retry when the server rate-limits them.

Ask and summaries now budget the whole prompt and say when retrieval was incomplete, instead of treating a failed source as one with no results. Ask also limits how long each step can run and cancels work that a newer question has replaced.

Interface and settings

The palette and Settings have a frosted look in dark or light, with rounded corners and an animated background that stays still when your system asks for reduced motion. Sources separate the services you have configured from the integrations available to add. AI settings expose the Ask and Summary limits.

In 1.10.2, recent searches start collapsed, and links in About, citations and Markdown open in your system browser.

Before you update

  • The first sync after updating rebuilds each semantic index (Paperless, BookStack and Obsidian) once. Settings shows “Rebuilding index…” while it runs.
  • Saved prompt templates are gone, along with their selected-text capture. Data you saved is kept.
  • Local models can still make claims the sources do not support, or produce citation markers that do not render as links. Check an answer against its sources before relying on it.

Prism runs on macOS, Windows and Linux. It uses local AI by default, and cloud providers are optional with your own key.

More at prismapp.xyz.