Vertana

Appearance

Vertana changelog ​

Version 0.2.0 ​

Released on May 5, 2026.

@vertana/core ​

  • Added keep() helper function for creating glossary entries that preserve terms without translation. [#3, #5]

  • Added properNoun() as an alias for keep(), for semantic clarity when preserving brand names and product names. [#3, #5]

@vertana/context-web ​

  • Added per-page and total character caps to fetchLinkedPages() and configurable fetchWebPage() sources, plus optional fetched-page summarization with an explicit model. Use these options to keep web reference material from overwhelming the source text in translation prompts. [#8, #11]

@vertana/context-memory ​

  • Added the new @vertana/context-memory package with a pluggable TranslationMemoryStore interface, deterministic InMemoryTranslationMemoryStore, and passive lookupMemory() context source. Use it to provide previous source/target segment pairs as compact translation examples for consistency across recurring text. [#2, #12]

Version 0.1.2 ​

Released on May 5, 2026.

@vertana/core ​

  • Hardened the system prompt against context pollution by wrapping required context inside an explicit <reference_material> block and prepending a rule instructing the model not to translate, quote, or echo any of it. This significantly reduces the chance that large fetched reference pages are emitted as the translation. [#7, #10]

Version 0.1.1 ​

Released on March 3, 2026.

@vertana/core ​

  • Fixed evaluate() structured output schema compatibility by removing unsupported numeric bounds from the score JSON schema and clamping scores to the [0, 1] range after parsing, which resolves failures with Anthropic models in the refinement phase. [#6]

Version 0.1.0 ​

Released on December 30, 2025. Initial release.