Public Crawler Page, Honest Limits & Interface Polish

Features

• Public Crawler Page — omniimpact.net/crawler documents the crawler by name: the exact user-agent string it sends, how it reads robots.txt, its crawl limits, and what to add to a host or firewall allowlist. It is a page you can hand to whoever runs your hosting.
• Report Branding — Agency and Enterprise workspaces can print their own name and logo on generated PDF reports, and switch on white label to remove the Omni Impact branding from them.

Improvements

• Plan Limits Come From Your Account — Usage badges, upgrade prompts, and the visibility dashboard now read your real limits from your account rather than a table compiled into the interface. A limit the app has not loaded yet reads as unknown instead of being rendered as zero and locking a feature you have.
• Honest Model Copy — The prompt editor no longer names a retired model as its default. What it lists is derived from the models your account can actually reach.
• Content Stays On Screen While Refreshing — Switching timeframes on the dashboard and on prompt metrics now keeps the current view mounted and refreshes it in place, instead of blanking to a spinner and back.
• Chart Empty States — Charts with nothing to plot now render a proper empty state explaining why, rather than an axis with no data or a degenerate shape.
• Faster First Preview — The PDF viewer is now loaded only when you first open a preview, and a large metrics drawer is loaded on first open rather than with the page.
• Readable Numbers and Labels — Response times are written in human units, onboarding step labels wrap instead of truncating mid-word, and the sidebar search shows a visible focus ring when tabbed to.
• Free Readiness Check Respects robots.txt — The crawler behind the public AEO Readiness Check now reads and honors robots.txt before fetching, the same as the authenticated crawler.
• Scheduled Runs Survive Deploys — The recurring prompt-tracking loop now checks for work it missed at start-up, so a deploy landing between ticks no longer skips a scheduled run.

Fixes

• Personal Workspace Access — Members of an organization can reach their own personal workspace again, and switching from an organization back to personal no longer signs them out.
• Workspace Switcher for Members — Members who are not administrators now see the workspaces they have been assigned, rather than an empty switcher.
• Retired Agent Streams — A chat that has been closed, replaced, or logged out from can no longer write into the conversation you are currently looking at, and leaving the code editor now retires its stream properly.
• Chart-Only Agent Replies — An agent turn that answers with a chart and no prose is no longer treated as an empty response.
• Agent Error Messages — Error bodies that arrive in a binary encoding are now decoded and shown instead of surfacing as an unreadable failure.
• Share of Voice Model Coverage — Share of Voice now runs every model your account is entitled to, keyed by the model itself; previously some entitled models were quietly skipped.
• Model Identifiers — Fixed a Gemini identifier that the gateway rejected, and refreshed the identifiers behind several models in the picker so selecting them executes the model you chose.
• Pricing Page Accuracy — Removed advertised figures the product does not actually grant.
• Stuck Chart Tooltip — A trend-chart tooltip that could remain on screen after the pointer left now clears, and the chart gained y-axis ticks and gridlines.
• Reliability Pass — Session-key verification no longer blocks other requests while it refreshes, every outbound model call now carries an explicit timeout instead of inheriting an unbounded one, and a set of unhandled failure paths found in error tracking were closed.