v1.1.3
LatestRicher AI chats and file review, plus Agent configuration and Markdown PDF export
New Features
- Steer and Queue Message: While the AI is still replying, Queue Message lines up follow-ups, and Steer steps into the current run right away. Queued messages can be edited, reordered, or withdrawn. Use them from the chat input; also available in Agent mode.
- Browse and jump to past prompts with arrow keys: In the chat input, use ↑ / ↓ to cycle through prompts you already sent. Use Alt+↑ / Alt+↓ (⌥↑ / ⌥↓ on macOS) to jump to the matching user message in the conversation.
- Export Markdown as PDF: Markdown can be exported as a same-named PDF in the same folder. Choose Export as PDF from the editor toolbar or the file-tree context menu, or ask in an AI chat to export it.
- Customize each Agent’s internal role models and generation preferences: On the Settings page, open an Agent under Agents to assign models to its internal roles in Role models, and set generation preferences in Preferences.
- Auto theme that follows system appearance: Theme now includes Auto, which follows your system light or dark appearance, and code highlighting follows the IDE theme more consistently. Set it under Menu → Settings.
- Choose whether AI file edits need your approval: When Require Approval is on, AI file edits stay pending until you accept or reject them. Turn it off to accept edits automatically. Find it under Menu → Settings → AI Edits.
- Add selected reply text to chat, or copy it: You can add selected text from an assistant reply to the chat as a quote, or copy it with Add to Chat / Copy.
- Return to your place after following a PDF link: After jumping via an in-PDF link, you can return to where you were reading with Back to previous location, or with Alt+← (⌥← on macOS).
Improvements & Fixes
- Chat stays closer to what you typed: Text you send is kept as typed instead of being turned into headings or lists, while code blocks still display as code. Pasted indentation and spaces are kept. Images from the web in assistant replies are no longer shown automatically.
- Large projects open more smoothly: Projects with many files are less likely to stall when you open them. A folder that suddenly gains a large batch of files is not stored in full in project history, and very large Agent file comparisons load only after you expand them.
- Built-in Gurobi license is now size-limited: Gurobi has restricted academic license distribution, so the built-in license changed from an unrestricted academic license to a size-limited one (about 2,000 variables and 2,000 constraints). Your own license can still be configured under Solver Settings in the sidebar.
- Jump to Settings sections from the avatar menu: You can open Account, AI Providers, User Rules, Agents, or Tools directly from the avatar menu, without browsing Settings first.
- TeX PDF preview no longer shows an older version while it first loads: When the TeX PDF preview first loads, or right after you switch files, it could briefly show a previous PDF. It now waits until the PDF for the current document is ready.
- Clearer separation of public and private Feedback: When submitting Feedback, public content and private information are kept apart more clearly.
- Review Agent file deletions before they take effect: If the Agent deletes a file, you can still review that deletion and choose to accept or reject it in the file review UI.
- More reliable Python and Lean completions: Python and Lean language completions are more reliable in everyday editing.
- Agent column alignment on the projects page: The Agent column on the projects page is aligned more consistently with the rest of the table.
- Improved management of Python compute resources: Python runs now time out after 1 hour and show a timeout message. A run that did not finish normally is cleaned up so it no longer holds compute resources.