Changelog

A reply with several tools or specialists shows Trace beside the assistant name. Opening it lists the specialist, model, and each tool in a popover, so the answer stays in place. A reply with a single tool call has no entry.

A running reply auto-expanded a trace of the specialist, the model, and every tool step above the answer, repeating the tool cards and pushing the reply out of view. That trace is no longer shown in the chat. The decision, multi-step plan, and tool cards stay.

The copy button on the key list put the masked preview on the clipboard, and that value cannot call the API. The button is gone. The full key can still be copied right after creation, and the button then shows Copied. Status now reads Active, Expired, or Revoked.

English names for the competitor group, daily report, content feed, and brand knowledge were longer than the 220px sidebar, so the ends showed as ellipses. Those labels are shorter now, such as Competitors, Daily Report, and Brand Knowledge. Searching a group name still lists every item in that group.

The model button no longer has a border, and its label appears once, such as DeepSeek Chat. The description stays in the menu. Ask, Review, and Auto match that button's height and drop the extra outline.

On the English interface, agent routing notes, titles, outlines, and articles are generated in English. Chinese prompt overrides no longer apply to English sessions. Body text uses the free Inter typeface so long replies are easier to read.

Running tasks stay in the floating dock at the bottom right, where you can open progress and results. The agent chat no longer repeats the same list above the messages.

Before a send, publish, or submit leaves the product, the chat stops for Allow once or Deny. The composer can switch among Ask, Review, and Auto, and Auto stops only for high risk. Active tasks appear above the messages.

Multi-step runs now show a read-only plan and trace in the chat, including each specialist's status, time, and which step a tool call belongs to. A reply with a single tool call still shows only the tool card.

The homepage GEO intro and three-step flow no longer describe site structure, keyword rankings, or a single score report. Cards now cover page readability, four separate rates, and competitor retests.

The homepage grid and glow now move at different speeds while you scroll. The GEO orbit and feature art shift slightly against the copy. Body text, the input, stats, and FAQ stay in normal flow. Parallax is off on narrow screens and when reduced motion is on.

In the English UI, TDK, SEO details, keyword, SEO/GEO, sitemap, and robots.txt report cards now use English titles and body copy. Chinese reports stay in Chinese.

While a keyword or domain probe runs, each model shows waiting, asking, or returned. A row updates as soon as that model answers, instead of staying blank until every model finishes.

Answer probes now default to GPT-6 Astra, Gemini 3.1 Pro, and Claude Opus 5.5, the mainstream models behind ChatGPT, Google, and Claude, instead of GPT-5 nano, DeepSeek V4 Flash, and Qwen3 32B.

When every probe model fails authentication together, that is one rejected AI Gateway credential. OpenRouter keys are no longer sent as Gateway keys, the failure says to replace the credential, and the remaining models are not called again.

The icon-only site menu is gone. Overview, audit, keywords, GEO, interpretation, action plan, and competitors are text tabs across the top, so each section is visible without hovering.

The site overview splits the total, SEO audit, and GEO into three tiles, and puts missing analyses on one action row. A single score no longer opens an empty chart, and briefing plus deep analysis sit in the header.

The sidebar footer now collapses and expands with an arrow icon. Hover still shows the action name, and the account name has more room on the same row.

The header account menu now sits in the sidebar footer. Avatar, name, and collapse share one row; the menu still opens settings, help, feedback, and sign-out. The header keeps notifications.

Issue actions now copy a fix prompt an agent can apply directly, plus a verify button that rechecks only the affected groups and refreshes the list.

When a full re-audit finishes, the progress dialog closes and the audit overview opens with the updated health score, errors, and warnings.

The sidebar now follows daily work: sites, automation, and reports under Monitoring; competitor analysis under Competitors; writing and publishing under Content. Org, keys, prompts, channels, data sources, and agents move to Settings.

Global corner radii tighten from 12–16px to 6–8px. Cards, buttons, and inputs look sharper; avatars and pills stay circular.

Content management now starts from topics and runs outline, research, write, and SEO. Blank create is closed; research must be approved before writing. SEO and GEO stay on separate tracks, with demo persistence and Word export when no model is configured.

Workspace site analysis now awaits converted model messages so the run can start. Pasting a URL with site questions routes to the SEO analyst; failures appear in the chat instead of only the error banner.

Pressing Enter in the homepage prompt now starts analysis in the Agent workspace. Use Shift+Enter for a new line; IME composition no longer submits.

Answer probes no longer call retired deepseek-chat or default to gpt-5.6-sol, which the Gateway free tier blocks. They now use GPT-5 nano, DeepSeek V4 Flash, and Qwen3 32B, with readable failure copy.

Action buttons now follow brand teal for primary, outline for secondary, and red for danger. Mixed blue, cyan, purple, and emerald fills on the same screen are gone; status badges still use semantic colors.

Each probe now asks OpenAI, DeepSeek, and Qwen through AI Gateway. The roster is visible before a run; history compares mention, recommendation, citation, or recognition per model without merging them into one score.

GEO answer probes now fan out through Vercel AI Gateway. After $5 of spend the path disables, instead of calling native provider APIs.

Probe history no longer dumps JSON. Brand, category, competitors, and mention / recommend / cite now use icons and copy. GEO Optimization Center also gets a quick-start tour you can replay.

The marketing homepage no longer shows a competitor comparison table or placeholder testimonials. The page now runs Hero, GEO capabilities, product capabilities, then FAQ. Methodology and case pages remain available on their own routes.

When international checkout uses Creem, usage is billed as credit points instead of only local monthly quotas. Creem can be switched with an environment flag, and the usage page shows the credit balance. Signup, usage, renewal, and revoke all keep credits in sync.

Model calls now go through Vercel AI Gateway. The settings slot is gateway — bring your own key. Legacy OpenRouter config is merged into Gateway; OpenRouter is no longer a product provider.

Added named-recognition (D) and brand-free keyword (K) answer probes. Mention, recommend, citation, and recognition rates are reported separately and are not merged into one AI-visibility score. Watchsets and same-condition retests are included; methodology and how-to live in the site docs.

Shipped the missing content-answerability API, wired llms.txt and AI-crawler access checks, and added an exportable fix checklist on the GEO overview (including FAQPage / Organization JSON-LD snippets). Personal and small-team sites can now act on schema, Q&A structure, and robots setup instead of only seeing a heuristic score.

New multi-agent task execution panel with real-time task status, plans, results, and logs. Provides a more intuitive task management experience.

SEO Analysis Agent now supports saving execution result attachments, including analysis reports, charts, and data files. Supports both append and replace save modes.

Integrated Firecrawl API for powerful web scraping capabilities. Supports batch scraping, smart parsing, and structured data extraction.

Added Feishu and Notion integration support. Analysis results can now be synced directly to Feishu docs or Notion pages.

Optimized responsive layout of card components for better visual experience across different screen sizes.

Launched new Competitor Analysis Agent that automatically analyzes competitors' SEO strategies, content strategies, and keyword layouts.

Added BI data analysis feature with natural language query support, chart generation, and report export.

Optimized dark/light theme toggle experience, fixed theme flashing issues, and added smoother transition animations.

System now fully supports Chinese and English language switching, covering all interfaces and functional modules.

Strengthened user permission management system with new Role-Based Access Control (RBAC) for more granular permission control.