Changes
- Added agent image generation: agents can create and edit images with cloud providers (OpenAI GPT Image and DALL-E, Google Gemini "Nano Banana"), configured through a new Image Generation settings panel for providers, API keys, account sources, and output targets.
- Added per-agent image generation budgets with daily and period limits, spend tracking, an action-on-limit policy, and a free option, surfaced in the agent Budget tab.
- Added the Gallery: a new tab to browse, preview, and manage images, a gallery picker modal, a slash command to insert gallery images into messages and editorial surfaces, and a gallery tool so agents can read images.
- Added file attachments across the app: attach files to private chat messages, task creation, note bodies, and note and task comments, persisted in a shared uploads directory and rendered in prompts and topic messages.
- Unified note and task comments into a single comments thread with agent-mention support.
- Added app reference links so agents, notes, and tasks render as click-to-navigate links in editorial surfaces.
- Added notes to the slash command menu with a note picker.
- Added a Community edition agent limit notice with an upgrade prompt, plus license-limit runtime notifications.
Fixes
- Reworked chat scroll memory to use height-stable, per-agent anchors, fixing jumpy position restoration when switching agents and after markdown renders.
- Corrected agent navigation and routing for reliable deep links.
- Isolated Chromium IPC sockets with a per-run TMPDIR and added SingletonLock diagnostics, fixing "browser already in use" startup failures.