New features

Boards you can shape

A new board used to arrive as a blank rectangle. It now starts with three columns you rename, so it reads as a kanban from the first click. Editing the schema looks like opening a ticket , title and description beside typed attributes , and renaming a column renames that value on every card in it, instead of drawing an empty column next to the old one.

Boards and cards also render in Brain, as boards, not as an empty prose editor that would write into the board if you typed.

Improvements

• Brain no longer lists every agent's own folders (output, config, memory) above the files you actually made
• The agent face in the sidebar blinks and glances when idle, and bobs when it is thinking
• Delete a document and you can undo it , restore puts the file back
• An approval card names what it is about to delete, instead of a generic confirm
• Closing an agent closes its tab, and Slack or LinkedIn rows in the inbox answer the keyboard again (R drafts from a chat row)
• The strategist writes its read into deal fields (on track, the play, how we win, risks) instead of a long assessment document. You can still customize it to keep a document if you want
• You can ask Cedar to file a webpage or an uploaded file into a knowledge-base entry, not only a Gmail attachment
• Ask Cedar to create, update, or delete a canvas from chat

Fixes

• Syncing or attaching a meeting now stores the transcript, so you can actually open it , the row used to claim there was one and the player said there wasn't
• Chat on accounts with many agents should stop failing mid-answer with a "prompt too long" error
• Chat tools (calendar, drafts, deal lookup) keep the user they are acting for, so they stop dying with a missing-user error
• A morning review of next-steps can no longer send a calendar invite to people outside Cedar without you approving it
• A chat id that is not a Gmail thread is no longer treated as one

Behind the scenes

• Field authoring can now see whether a field is actually kept live, find Slack channels the Cedar bot can post to, and test-run a subagent without sending for real
• Admins can roll up field and CRM-mapping drift across reps and propagate a change , a diverging target is refused unless you force it, and force will not run across more than one drifted target at once
• Creating a custom field can take manual options or one-step CRM picklist sourcing; removing an option shows how many values are affected
• Playbook webhooks can be listed, test-fired, and deleted; set-stage-criteria writes entry/exit text on the org playbook only
• CRM field mappings no longer show a per-field push toggle that never actually controlled pushes
• Conversation review no longer runs crm-updater or strategist (those agents need a live event a review does not have); next-steps and drafter go through the same curated tool path as a normal run
• Chat delegates share one tool server and prove who they are with a per-turn token, instead of cloning the full tool schema for every agent
• Manual meeting sync writes the transcript to storage before the meeting row, so a failed write can be retried instead of leaving a poisoned "has transcript" flag
• Staging deploys wait for CloudFormation to finish an in-flight update before starting the next one
• The automation model experiment now routes flagged traffic to Sol on its own workload
• Authoring docs were refreshed for field liveness, retired model/when-to-use frontmatter, and the new config-propagate rules