Changelog
The complete history of every version — Keep a Changelog format, semantic versioning — lives in CHANGELOG.md in the main repository. This page summarizes the most recent milestones.
Recent highlights
v1.0.57 (2026-09-28) — Protection against instructions hidden in indexed documents (indirect prompt injection): malicious content in a video, PDF, or text file is no longer followed as a command by the assistant; when an answer still looks influenced this way, a warning appears in the chat. Dependencies updated, including React and axios (with a security fix).
v1.0.56 (2026-09-02) — Fixed the automated accessibility audit in CI, which had been failing since v1.0.55.
v1.0.55 (2026-08-30) — OpenRouter as a new free chat provider; search index storage migrated to LanceDB (ranking is still BM25, no perceptible change in answer quality); Especialista profile can hide the Study Mode card on the home screen; tables and lists in assistant replies no longer render glued together on screen; contrast and keyboard focus improvements.
v1.0.54 (2026-08-22) — Thumbs-down in chat now has a real effect: it accumulates a per-project counter that widens the search instead of just discarding the response.
v1.0.53 (2026-08-14) — Keyboard shortcuts with ⌘ (Cmd) on macOS; a saved flashcard modal now opens the full review experience (flip card, rate); fixed a chat search bug that ignored the actual question term in bases with a lot of formal text; per-file indexing progress with working cancellation; fixes to the expanded chat, a checkbox, and i18n.
v1.0.52 (2026-08-14) — Intent routing in chat: questions about the base itself ("how many videos does this base have?", "when was it indexed?") answer with real data from disk, no LLM call; pure calculation questions ("what's 15+27?") are solved by a safe arithmetic evaluator, no LLM or search; automatic retry in Broad Search before showing "not found". macOS keyboard shortcuts documented in Help.
v1.0.51 (2026-08-12) — YouTube video chapters longer than 3000 characters were being truncated in the search index and chat citation; now split into multiple parts, with continuity preserved.
v1.0.50 (2026-08-10) — Single LLM client factory in chat (fixes a real drift in the Gemini fallback list); full reset now available to every profile, not just Specialist.
v1.0.49 (2026-08-10) — Vector search (embeddings via Ollama nomic-embed-text) as a complement to BM25+FTS5+CrossEncoder in chat, active in Broad Search with optional one-click download.
v1.0.48 (2026-08-09) — Tagline unified to "Augment" across all logo art and text, deliberately referencing Douglas Engelbart's concept of Intelligence Augmentation.
v1.0.46–47 (2026-08-07/09) — Release fixes (race condition between Windows/macOS builders), full Overview translation, refreshed logo and README.
v1.0.44–45 (2026-08-07) — Real, persisted audio player for Study Mode summaries, fixed unwanted horizontal scroll in the side menu.
v1.0.42–43 (2026-08-06/07) — Complete Study Mode (flashcards with SM-2, summaries, post-its), playlist selection and date filter in extraction, support for Ollama models with native reasoning (thinking), "Agent" renamed to "Assistant" throughout the interface.
v1.0.41 (2026-07-31) — macOS installer (Apple Silicon), search public sources by knowledge area (arXiv, OpenAlex, FHIR and others), standalone web page reader, formalized as source-available (Elastic License 2.0).
v1.0.38–40 (2026-07-24/27) — Automatic legal document recognition, per-sentence graded confidence in chat, MCP Server exposed in the interface.
For the full detail — including internal CI/infra fixes — see the complete CHANGELOG.md.