Small Fixes and Improvements

A few smaller changes that shipped this week:

• You can share a conversation with your workspace, read only. Teammates find it under "Shared with me" in the sidebar, can read the whole thread including what the agent did, and can't continue or edit it. Public links are still a separate switch in the same modal.
• The Brain is no longer read only from the web. You can create files and folders anywhere in the tree, edit any page in a real editor, and save when you're ready. Each save is one version, so the history stays readable, and the files the agent maintains itself are locked.
• You can tell the agent how to write to you. Hover any reply, open the controls next to Copy, and a model reads its own answer back and names what was wrong with it. Tick what you agree with or write your own rule, see the reply rewritten, and save. It applies to chat, Slack and routines from then on, and it's set per agency rather than per client.
• Every ad comment now has a link that opens it on Facebook or Instagram, with the channel's own glyph, so you can answer or hide it at the source without hunting for the ad.
• Videos play on shared links. Boards, live feeds and shared Top Creatives reports were showing a still frame to anyone opening them without an account.
• The chat model picker is grouped by provider and carries Fast, Balanced and Deep thinking controls, with Sonnet 5, Opus 5, Fable 5.1, Gemini 3.7 Flash and GPT-5.6 Sol available. Your old choice is carried over to its replacement.
• The Google Ads, LinkedIn and Meta account pickers ask the platform for the current list every time you open them. An account you were granted after connecting shows up without disconnecting and reconnecting.
• Connecting Google Ads tells you when the Google Ads permission wasn't ticked on the consent screen, instead of closing the popup with Google's own wording about authentication scopes.