1.0.0-beta.26

published 1 October 2026

• Added xAI Grok models, with a Grok subscription sign-in or an API key from /logins; a subscription shows its weekly pool and purchased credits in /usage
• Added an xAI usage meter in /usage through an optional management key
• Added Meta Muse models with a Model API key from /logins
• Added /advise, a per-chat advisor switch that survives a resume, with /config → Advisor as the settings for every chat
• Added an advisor interval in minutes, and a chip in the status row while the advisor is on
• Added the credit balance beside every sign-in's usage meter
• Added deepseek, kimi and glm as model words for each family's newest model
• Added the DeepSeek, Kimi and GLM model names to the start screen's strip
• Added a warning on machines under 16 GB when a picked local window leaves too little memory
• Fixed the advisor running in every session at once; crewmates are never advised
• Fixed a full conversation getting stuck when Auto-compact was off
• Fixed automatic compaction starting too close to the context limit
• Fixed a full usage window not saying what takes your requests until it resets
• Fixed a tool row's counter running on after the tool had finished
• Fixed a very long single-line prompt hanging before anything was sent
• Fixed the exact token count running low after a thinking turn
• Fixed a very long GPT conversation getting stuck at the context limit instead of compacting
• Fixed /logins opening as a bottom panel instead of a popup
• Fixed an idle session polling a crew file when it had no crew
• Fixed a parked session's scheduled wake showing "due now" the whole time it was parked
• Fixed agent worktrees starting from a stale main or colliding on the repository lock
• Fixed mercury -p /usage (and the other screen-only commands) printing nothing instead of saying they need the interactive session
• Fixed the Gemini sign-in screens to say what Google ended and what the API bills, with Gemini's agent-only products kept out of the model picker