Changelog — AI Playground

v1.2.5 latest

July 21, 2026

Syntax-highlighted code blocks with language labels and one-click copy buttons in Chat mode.
Collapsible thinking/reasoning tags — instead of stripping <think> blocks, they render as a toggleable "Thinking..." section you can expand to inspect the model's reasoning.
Edit & Regenerate — edit any user message inline and re-send, or regenerate the last assistant response with one click.
Animated thinking dots indicator while waiting for the first streaming chunk, replacing the previous empty state.
Live streaming speed indicator showing characters per second during response generation.
Context window usage bar now shown in Chat settings panel — see token consumption in real time.
Conversation pinning — pin important conversations to the top of the sidebar.
Message search — search across all messages in the current conversation from the sidebar.
Export conversations as Markdown, JSON, or plain text with one-click download.
Markdown table horizontal overflow — wide tables now scroll instead of breaking the layout.

v1.2.4

July 15, 2026

Fully migrated telemetry infrastructure to PostHog using the native Node SDK, replacing the manual offline queue system.
Background tracking is now crash-resilient and bypasses typical browser CORS/network restrictions affecting Electron's internal network stack.

Implemented strict web crawler filtering on the backend to prevent search engine bots from artificially inflating software download metrics.

v1.2.3

July 13, 2026

Model cache versioning — provider model caches now include a version number and automatically invalidate when the app updates, preventing stale model lists.
Improved AI/ML API provider parsing — more reliable model type detection and text/image capability inference across 400+ models.
"Save to Library" button relocated into each individual image output for quicker access.

Fixed Switch toggle visibility in Settings — toggles now render with correct background colors across all themes.
Empty model lists are no longer cached, preventing providers from showing zero models after a failed fetch.

v1.2.1

July 12, 2026

External links in Settings (Buy me a Coffee, Changelog) now open correctly on Linux.
Mac update download URL now points to the correct sync server.

v1.2.0

July 12, 2026

Three new providers — AI/ML (100+ models with image generation), NVIDIA (55 enterprise inference models), and Hugging Face (52 open models via router API).
Provider Health panel redesigned as a 3-column grid with brand-colored monograms for quick visual identification.
API Keys page redesigned — remove key button with confirmation dialog, show/hide key toggle (auto-hides after 10s), model count column, and last tested timestamp persisted in database.
Live Pricing Catalog now covers AI/ML API and NVIDIA NIM models with fallback pricing for all known models.
"New" badge system — newly added providers show a green badge for 6 days to highlight fresh integrations.
Hugging Face routing suffix toggle — choose between cheapest, fastest, or preferred model routing per request.
Vision-capable models only — Image Attach Zone in Text Playground now only appears for providers that support image input.
Full internationalization (i18n) — 6 languages (English, Spanish, French, German, Brazilian Portuguese, Simplified Chinese) with 487 translated keys per locale. Language picker on first launch and in Settings → Appearance.
Text Prompt Builder and Image Prompt Builder fully translated — all labels, toggles, placeholders, toasts, and the "About this Tool" reference table.
Shared ProviderSelector and ModelSelector components now display localized placeholders, search hints, and empty states.

Copy button in Chat is now icon-only with green checkmark feedback (1.5s).
Links in AI responses now open in your default browser instead of navigating within the app.
Thinking/reasoning tags from Hugging Face and other models are now stripped from rendered output.
Provider selector dropdown no longer overlaps the app title bar.
Sidebar provider count derived from configuration instead of hardcoded — automatically reflects new providers.
Fixed pricing catalog duplicate key for meta-llama/llama-3.1-70b-instruct across providers.

v1.1.0

July 3, 2026

Persistent Multi-turn Chat — standard conversation flows seamlessly across all supported providers (OpenAI, Claude, Gemini, Grok) with full chat history preserved via local SQLite.
Chat Sidebar Management — jump between threads instantly, rename active conversations via inline-editing, and continuously append to past run histories seamlessly.
Unified Theme — full conversion to a custom earthy "base-nova" dark theme across all Playgrounds and settings panels.
Revamped UI Dropdowns — native modern standard HTML Select inputs for cleaner UI consistency over previous comboboxes.

Resolved Tailwind module caching loops in Next.js backend to stabilize dev experience.
Chat message UI elements no longer show white borders when in dark mode.

v1.0.2

June 28, 2026

v1.0.1

June 15, 2026

v1.0.0 initial release

June 5, 2026