Restores earlier ChatGPT messages in unpaginated conversations. Complete conversation graphs are no longer cut; native paginated windows and upward history cursors remain intact.

Fix ChatGPT upward history loading while preserving native pagination cursors and the provider virtualizer. Stabilize code hydration without overriding CodeMirror editor/scroller geometry. Keep complete local timeline ordering safe against stale responses. Reduce duplicate full-graph processing and clarify the automatic recent-exchange window.

Threadwing 0.5.18 brings the current long-chat experience to ChatGPT, Claude, Gemini, Grok, DeepSeek, Kimi, and Doubao.

• ChatGPT now follows the provider's current conversation schema, keeps official turn geometry authoritative, and stabilizes compact-to-editor hydration so long answers do not create blank regions or repeated layout jumps.
• Export adds prompts-only and selected-message workflows, while Kimi web-search sources are preserved as safe, deduplicated links without leaking provider controls into exported content.
• The popup links to the official Threadwing website and adds a user-initiated browser update check with immediate, persistent feedback.
• Safe settings apply now restores drafts and scroll position against current ChatGPT turn anchors and rolls controls back cleanly if a refresh cannot be applied.

No tracking, remote code, payment SDK, conversation upload, or paid feature gate is included.

Fixed overlapping search results in the detached Timeline when long prompts or expanded answers change height. Small filtered sets now use natural document flow; larger searches and the full history retain virtualization. Search navigation and scroll restoration remain available.

Threadwing 0.5.11 makes the browser popup easier to scan: one active-conversation overview, clear Adaptive and Manual controls, and a direct Load all exchanges action in Manual mode. It retains the measured page/local-history counts and removes redundant healthy-state labels. This update does not replace the provider's native conversation renderer. Search, complete local timeline, selected-message export, and seven-provider support remain available. No analytics, conversation upload, remote code, or paid feature gate was added.

Threadwing 0.5.10 keeps the newest ChatGPT prompt and in-progress answer fully visible while a response is being generated, even when current ChatGPT markup omits per-turn streaming attributes. This prevents blank current turns and scroll jumps without disabling measured offscreen optimization for older completed content.

This version retains native upward cursor pagination, the local Timeline/search/export workflow across ChatGPT, Claude, Gemini, Grok, DeepSeek, Kimi, and Doubao, and local-only privacy. No tracking, remote code, payment SDKs, conversation upload, or paid feature gates.

Restores bounded ChatGPT rendering so long conversations keep only the current adaptive or manually selected exchange window on the official page while the complete history remains searchable and exportable in Timeline. Prevents upward scrolling from loading every earlier page, adds measured offscreen block containment for very large answers, and preserves the official ChatGPT interface. Also improves code rendering, message metadata, sources, provider coverage, and export readability across ChatGPT, Claude, Gemini, Grok, DeepSeek, Kimi, and Doubao.

Restores adaptive page relief for Claude, Gemini, Grok, Kimi, and Doubao without removing provider messages. Adds Manual control for 1–100 retained exchanges and a distinct Render all action. DeepSeek's native virtual list remains authoritative. Improves scroll stability and avoids full transcript rescans during streaming.

Threadwing 0.5.1 keeps each provider's native scrolling intact while using browser-native deferred rendering for older completed turns. It also folds reasoning and tool activity by default, removes popup whitespace, and improves role-distinct exports with selectable syntax-highlighted code, metadata, and safe sources.

Threadwing 0.5.0 adds complete multi-provider support for ChatGPT, Claude, Gemini, Grok, DeepSeek, Kimi, and Doubao. Adaptive rendering now remains bounded while scrolling, Manual mode provides a clear recent-turn limit plus one explicit Load all action, and provider-aware Timeline/export preserves prompt/reply roles, metadata, sources, code highlighting, and selectable message exports. This release also improves long-conversation performance, accessibility, bilingual UI, and print/PDF fidelity.

Threadwing 0.4.7 keeps Adaptive mode as the clear default and applies strategy changes immediately with draft-safe reload protection. Provider-triggered upward pagination can no longer refill the source page in bounded modes; complete display is one explicit action inside Manual. The render-cost model reacts more strongly to long text, dense lines, highlighted code, formulas, tables, and images. Inline formulas remain semantic even inside CJK emphasis or provider-double-escaped text. Full local history remains available after a partial page reload, while Prompt and Reply keep distinct role surfaces and local syntax highlighting across Timeline, Web Archive, and Print Studio.

No tracking, remote code, payment SDKs, conversation upload, or paid feature gates.

Threadwing 0.4.6 makes long AI conversations clearer and lighter. Page rendering and local history are now separate controls; collapsed replies no longer build Markdown, math, tables, images, or syntax-highlighted DOM. Prompt and Reply use distinct indigo/green surfaces in the timeline, HTML archive, and Print Studio. Code blocks now use bundled, language-aware highlighting with language labels and line counts. No tracking, remote code, conversation upload, or paid feature gates.

Threadwing 0.4.4 turns the shared timeline into a prompt-first index. Long prompts fold by default, replies stay compact and are not rich-rendered until Show reply is selected, and exchange details still show complete content. This reduces visual noise and rendering work across ChatGPT, Claude, and Gemini.

No tracking, remote code, payment SDKs, conversation upload, or paid feature gates.

Threadwing 0.4.3 keeps live conversation metrics stable while loading older history and collapses long assistant replies into compact previews with Show full reply and Collapse reply controls. It also includes complete ChatGPT, Claude, and Gemini support plus responsive-layout and performance improvements.

New in 0.4.2:
• Reduces main-thread work in long conversations with native offscreen rendering and mutation-local reconciliation.
• Avoids repeated full-transcript serialization while Claude or Gemini streams.
• Allows fixed mode to retain as few as one recent exchange.
• Preserves measured popup metrics when settings synchronize.

No tracking, remote code, payment SDKs, conversation upload, or paid feature gates.

New in 0.4.1:
• Preserves semantic headings, paragraphs, lists, emphasis, links, quotes, tables, code, math, images, and attachments captured from Claude and Gemini.
• Renders complete rich answers in the virtualized timeline instead of flattening or truncating them.
• Keeps compact highlighted search results and exact jump-to-source navigation.
• Refreshes the synchronized store screenshots to show the production rich-content experience.

No tracking, remote code, payment SDKs, conversation upload, or paid feature gates.

New in 0.4.0:
• Adds complete provider adapters for Claude and Gemini alongside ChatGPT.
• Unifies timeline, local search, exact navigation, draft-safe refresh, export, and reversible offscreen optimization across supported providers.
• Adds explicit provider identity throughout the popup, timeline, website, privacy policy, and review materials.
• Introduces a registry-based architecture for adding future providers without changing the normalized conversation core.

No tracking, remote code, payment SDKs, conversation upload, or paid feature gates.

Fixes a Firefox popup initialization issue that could leave Threadwing stuck on “Connecting…”. Runtime status now appears independently of transcript metadata, and rejected, malformed, or non-responsive page messages settle to an actionable state within a bounded time. No permissions were added.

Threadwing 0.3.10 brings the current three-browser release to Firefox:
• fixes popup status geometry and retained-node metrics;
• improves complete-history loading, local full-text search, branch inspection, and exact message navigation;
• adds Export Studio with Markdown, text, JSON, navigable HTML, and browser-native print/PDF output;
• adds Formal, Mint, and Monochrome export themes, prompt excerpts, embedded ChatGPT images, selectable text, clickable contents, and PDF bookmarks;
• clears stale conversation content when the active tab is not a supported ChatGPT page.

Threadwing remains local-first, with no backend, analytics, telemetry, advertising, account system, or conversation upload.