Buda v1.4.0 Release Notes

Released: May 18, 2026

✨ New Features

• More free models are available: Free workspaces can now use stronger models including Claude Sonnet 4.6, GPT-5.4, and GPT-5.3 Codex.
• Clearer image tools: Image generation and editing tools are now separated by Gemini and GPT. We also added generate_image_gpt and edit_image_gpt, so you can use the stronger gpt-image-2 model to create and edit images.
• Video results appear directly in chat: When a video task finishes, the file card now appears in the chat automatically. The status is also easier to follow, from queued and running to succeeded, failed, or expired.
• One-click actions for project folders: When a Drive folder is a complete git project, Buda recognizes it automatically. The new action bar lets you open the related Git view or restart the dev server without switching panels or picking the project again.
• Git view works better with Drive projects: When you open Git from a project folder, Buda automatically uses the current repository path. Commit diffs also open more smoothly, making it easier to review past changes.

⚡ Improvements

• Drive feels smoother: Folder switching is optimized, so file operations no longer cause a brief pause afterward.
• Marketplace installs take effect right away: After you install a skill, hire an Agent, or recruit a team from the Marketplace, the new skills appear on your Agent immediately. No page refresh needed.
• Channels keep your chosen model: The model you choose in Web chat is saved as the Agent's default. New conversations from WeChat, Telegram, and other channels now use that same model instead of silently falling back to the default one.
• Cross-channel delivery is more reliable: When an Agent sends content from one channel to another, Buda can recover from an expired session and use an available channel under the same Agent, reducing target_not_found failures.
• WeCom quoted messages are understood: When you reply to a quoted message in WeCom, the Agent can now read that quoted context and answer with the right background.
• WeCom bot connections are steadier: If the WeCom connection drops, Buda now recovers automatically. If another connection interrupts it, Buda reconnects more carefully to avoid long periods with no response.
• Clearer onboarding: The welcome wizard and docs now include a "Space → Agent → Session + Drive" diagram, helping new users understand Buda before watching the intro video.
• Billing history is easier to read: Tool usage now shows a small "Tool" label and the tool name, such as Image generation, so you can quickly tell whether credits were used by a model or by a tool.
• Video errors are more useful: If video generation fails, the Job panel now shows the real provider error message, code, and parameter instead of a generic 400.
• Generated videos return to the original chat: When a video generation task starts from a channel, the result is sent back to that same chat automatically. You do not need to go look for it.
• Usage limit emails arrive sooner: When AI credits or member limits are reached, Buda now sends an email notification so you can understand why an action was blocked.

🐛 Bug Fixes

• Sidebar avatars stay round: When the sidebar is collapsed, the avatar in the lower-left corner stays circular. Avatar changes from account settings also update in the sidebar immediately.
• The first WeChat image can upload correctly: When a user sends an image as the first WeChat message, Buda now prepares the Agent's sandbox first and uploads the image to Drive correctly.
• Drag-and-drop upload text follows your language setting: Some upload prompts used to follow the browser language instead of your account language. They now respect your saved language preference.
• Friendlier network error messages: When a temporary network issue causes a chat message to fail, Buda now shows a clearer localized message telling you to try again later.
• Internal NO_REPLY text no longer appears in chat: If an Agent calls a tool and then decides not to reply, Buda keeps the tool result visible but hides the internal NO_REPLY control text.
• Sandbox and workspace matching is more accurate: File uploads, channels, and sandbox-related actions now resolve the correct workspace more reliably, reducing cases where files go to the wrong place or the workspace cannot be found.