Skip to content
Dusori

v0.2.0 release notes

Released July 21, 2026. Version 0.2.0 adds a deliberate web-research loop and makes larger knowledge graphs easier to read without changing Dusori’s local-first file contract.

Open Dusori · View the GitHub release

Research from an objective

Section titled “Research from an objective”
  • Start with the next incomplete roadmap objective or choose another objective.
  • Search the keyless Microsoft Learn catalog or English Wikipedia only after explicit, per-provider consent.
  • Review a sanitized result and the exact Markdown capture before adding it.
  • Accept a result through the normal URL-source path so it deduplicates by URL, records provenance, appends the topic update log, and appears in the portable graph.
  • Dismiss irrelevant results persistently without sending dismissal data anywhere.

Microsoft Learn captures are catalog references; Dusori does not fetch the linked page. Wikipedia captures use the public article extract API and cap stored content at 2 MiB.

A clearer constellation

Section titled “A clearer constellation”
  • Constellation geometry now scales deterministically with workspace size.
  • Wikilink affinity influences topic ordering without writing coordinates into the workspace.
  • Marigold rings identify connection hubs.
  • Pointer and keyboard focus dim unrelated artifacts and can open the selected source directly.

The public product surface also adopts the supplied ensō, rangoli, and katana geometry with a finite, reduced-motion-aware identity animation.

Safety and portability

Section titled “Safety and portability”

Provider consent stays in browser storage. Research starts only after a deliberate search action, all network access uses the browser’s injected fetch boundary, and the offline service worker ignores cross-origin API responses. Topic research.json files are schema-validated and use hash-guarded writes so concurrent dismissals are merged rather than lost.

No account, telemetry, hosted backend, paid API key, AI model, or graph database is required.

Known limitations

Section titled “Known limitations”
  • Research is limited to the Microsoft Learn catalog and English Wikipedia.
  • Dusori does not crawl arbitrary pages, extract PDFs, run key-based search, generate schedules, transform content with AI, or perform unattended work.
  • Direct folder access still requires a supported Chromium browser; other browsers use ZIP portability.
  • The optional companion remains source-only and is not published to npm in this release.