0.4.0

Changed

• Added safer local launcher and connectivity defaults around 127.0.0.1, local CORS origins, and occupied-port handling.
• Reworked entity resolution into a durable incremental flow with exact scan/embed/apply phases, resumable saved progress, and stricter exact-only eligibility based on trustworthy unique-node coverage.
• Improved chat retrieval budgeting with ranked graph-node selection before the final node cap, final retained mutual neighbor caps, per-node description truncation, and an overall context-character limit.

Fixed

• Improved pooled-provider retry behavior and Gemini chat/query-embedding rate-limit handling so ambiguous Gemini 429 responses back off at the request level instead of falsely cooling down the whole key pool.
• Fixed entity-resolution startup and persistence issues including Chroma batch-limit handling, exact-only batch embedding behavior, already-committed merge durability, and Windows graph-save file-lock retries.
• Fixed local frontend/backend startup truthfulness, browser-side connectivity messaging, and shipped version metadata consistency.

No breaking changes are intended in this release.

v0.3.0

Changed

• Changed active-ingest graph handling to keep a committed in-memory graph session per world, serve graph/chat reads from the last durably saved snapshot during ingest, and acknowledge aborts without waiting for large graph saves to finish first.
• Changed chat on partial worlds to degrade gracefully instead of hard-blocking when chunk or unique-node vector coverage is incomplete, while still using whatever chunk and graph context is currently available.
• Changed entity resolution so Exact only can now run on aborted or partial worlds after ingest stops, while Exact + AI still waits for a fully completed world.
• Changed Safety Queue behavior after entity resolution so repaired chunks that were already made live and used by the last entity-resolution run become read-only until a later graph-changing ingest starts again.

Fixed

• Fixed resume/recovery so node-only unique-node embedding gaps no longer trigger chunk re-embedding.
• Fixed resume/retry cleanup so chunks that already have chunk vectors no longer delete and rebuild those vectors just because extraction coverage needs to be repaired.
• Fixed chat retrieval to reuse partial unique-node coverage and restore Context Graph results instead of dropping graph context whenever the unique-node index is incomplete.

v0.2.0

Changed

• Updated the default Graph Architect prompt to keep the app's nodes/edges JSON schema while asking for more detailed relationship descriptions.
• Refactored ingest to stream chunk embedding, graph extraction and glean, and unique-node embedding in parallel so chunk vectors start earlier and node vectors no longer wait until the end of the run.

Fixed

• Tightened ingest durability and progress tracking so checkpoints only advance after both chunk embedding and durable extraction succeed, resume no longer skips extraction for chunk-only vector hits, and staged node-vector work does not overstate committed graph coverage during active runs.

v0.1.1

Quick start

• Download the source zip from this release and extract it
• On Windows, run VySol.bat
• For API keys and first-run setup, read docs/SETUP.md
• For a full app walkthrough, read docs/WALKTHROUGH.md

Added

• Added a shared atomic JSON writer with Windows-friendly replace retries, plus focused regression coverage for atomic writes and active-ingest world snapshot responses.

Changed

• Changed world-card and ingest-config API payloads so the home and ingest UI can use lighter world summaries, live active-ingest audit snapshots, and provider-backed embedding model options more truthfully.
• Changed the world re-ingest setup UI to use provider-backed embedding model pickers instead of relying only on free-text model entry.

Fixed

• Fixed public settings payloads so /settings no longer returns stored API keys or internal provider-credential maps to the browser.
• Fixed the graph page to show a real loading state and retryable load error instead of failing silently when the graph request does not return.
• Fixed ingest-page runtime-state races so optimistic start state, live backend ingest ownership, aborting state, and stale snapshot refreshes no longer desync as easily.
• Fixed several metadata write paths to use the shared atomic JSON helper, reducing Windows file-lock failures during entity-resolution and world-duplication state writes.

Full Changelog: Vyce101/VySol@v0.1.0...v0.1.1

v0.1.0

Added

• Added a provider registry with first-class Groq support for chat, graph extraction, entity chooser, and entity combiner, plus provider-aware embedding settings that stay truthful when a backend is not actually available yet.
• Added a preset-backed global Configuration tab, a shared per-provider Key Library, and a locked baseline Default preset for saved global model/provider setups.
• Initial GraphRAG backend and frontend with a resilient local launcher.
• Split setup and usage docs into dedicated guides for setup, walkthrough, and features.
• Added a dedicated system diagram page and linked it from the README.
• Added branding assets and a README logo.
• Added a tracked frontend .env.local.example so clean clones can configure NEXT_PUBLIC_API_URL without guessing local setup.
• Added entity resolution run modes, including Exact only and Exact + chooser/combiner.
• Added per-key Gemini API key toggles and a dedicated repo-local pytest temp folder.
• Added the VySol browser/header branding icon using the square logo asset.
• Added a per-message Context Graph view in Context X-Ray for newer chat messages.
• Added a Safety Review Queue for extraction safety blocks, including one-shot recovery for already-collapsed blocked chunks and in-app chunk editing/testing.
• Added inline chat renaming in the sidebar, with conflict-safe saves that preserve Recent ordering.
• Added a world-specific Re-ingest settings editor for editing saved chunk settings, glean amount, world-local ingest prompts, and repaired-chunk reuse before starting a full rebuild.
• Added a New Nodes entity-resolution metric that tracks graph growth since the last completed entity-resolution run when a saved baseline exists.
• Added crash-safe world duplication from the home page, including copied chats/settings/graph data, temporary duplicate preview cards, and bottom-right duplication progress tracking.

Changed

• Changed global settings so provider-specific controls now only appear when the selected provider actually supports them, including Gemini safety/thinking controls and Groq reasoning controls.
• Changed Gemini and Groq reasoning controls to follow per-model capabilities more truthfully, with docs-backed presets for known supported models, visible unsupported-state messaging for known unsupported catalog models, and advanced manual entry only for custom or unknown model ids.
• Changed world ingest settings, vector manifests, and rebuild checks to track embedding provider as well as embedding model.
• Refreshed docs and aligned runtime defaults.
• Corrected project branding capitalization to VySol.
• Changed the README hero image to the tracked white-background social asset, and aligned backend/frontend public metadata with the VySol name.
• Refined the README positioning, added a Common Uses section to present roleplay in pre-existing fictional worlds as VySol's personal origin while broadening the documented use cases, and clarified the current macOS/Linux manual-setup support wording.
• Documented entity resolution run modes.
• Documented API key toggle behavior in the walkthrough and Google AI Studio key guide.
• Renamed the default embedding model to gemini-embedding-2-preview.

Fixed

• Fixed backend startup crashes caused by parser-level indentation damage so saved worlds/settings no longer appear to vanish when the API fails to boot, and the home/settings UI now reports backend-load failures explicitly instead of showing a false empty state.
• Fixed ingest and entity-resolution UI honesty so non-terminal ingest chunk errors no longer close the live progress stream, stale runtime refreshes now surface a visible warning with retry, and normal entity-resolution reason updates no longer render as false failure banners.
• Fixed ingest audit/progress truthfulness so unreadable vector collections now surface as explicit world blockers instead of fake missing coverage, Re-embed All keeps progressing through world-level unique-node rebuild and audit phases after chunk work finishes, and world-scope blockers are visible in the main progress UI instead of hiding only in the agent log.
• Fixed ingest truthfulness and abort responsiveness so chunk embedding no longer falsely reports 0 node vectors mid-run, final unique-node rebuild completion now reports its real count separately, slow model/embed awaits stop blocking abort responsiveness, and aborted runs ignore late provider results instead of writing stale data.
• Fixed Gemini chat payload assembly for the Gemini SDK request shape.
• Fixed launcher startup state detection.