This release focuses on AI enhancements and developer experience improvements, with better ways to provide context to AI chat and real-time feedback during package installation. Internal messaging improvements also make notebooks more responsive.
✨ Highlights
Enhanced AI context with files, cell outputs, and attachments
Provide richer context to AI chat by referencing files with #, cell outputs with @, and attaching images or other media (#6241, #6240, #6181, #6256). The AI can now "see" your code outputs, charts, tables, and files to provide more targeted assistance.
Real-time package installation progress
See live installation logs when marimo installs packages (#6279), giving feedback on what's happening during dependency installation without requiring the terminal.
✨ Enhancements
- Basic Debugpy integration (#6311)
- Add
pycapto module name map (#6315) - Visibility for
@in Generate with AI, also add model dropdown (#6300) - Stream packages installation logs (#6279)
- AI parse from parts in the backend, enable attachments for more providers (#6256)
- Support empty directories with
mo.ui.file_browser(#6261) - Migrate custom serde to
msgspec(#6234) - Cell output for
@context w/ attachments for media output (#6240) - Lazy-load command pallete and sidebar panels (#6232)
- Add Azure OpenAI provider support for existing models (#6231)
- SQL cells can't both define and refer to itself (#6216)
- Add warning when styling and sorting together (#6225)
- Add mapping for
upathtouniversal-pathlib(#6221) - Attachments in chat sidebar (#6181)
- OpenAI version 5 models go above 4 (#6212)
- Update Gemini2.5 models (#6215)
- Reference files as context with
#(#6241)
🐛 Bug fixes
- Marimo export script with headers (#6320)
- Attempt to fix intermittent CLI test failures (#6308)
- Fix
progress_bar.__iter__type hint to returnIteratorinstead ofIterable(#6292) - Fix tools conversion (#6289)
- Validate auth-token on websocket (#6284)
- AI endpoints from msgspec migration (#6273)
- AI endpoints from msgspec migration (more fixes) (#6278)
- Set SQL footer's background colour and border color (#6267)
- Create dataclass wrapper for VariableValue to resolve Pydantic compatibility (#6265)
- Vertical layout cleanup for read mode rendering pipeline (#6248)
- Fixed storybook setup for static layouts with window.MARIMO_STATIC (#6246)
- Output cell fails to wrap text in HTML export (#6245)
- Add additional test for lowest resolution bounds (#6198)
- Remove uneccesary keys in google tools ai call (#6201)
- If the gpu stats command fails, don't run it again on every usage check (#6202)
📚 Documentation
- Move
CLAUDE.mdto snippets (#6276) - Add some notes on deploying programmatically (#6217)
- Add
CLAUDE.mdto prompts docs (#6227)
🔬 Preview features
- Add
--server-startup-command(#6238) - Add MCP server integration with notebook inspection tools (#6222)
📝 Other changes
- Refactor backend tools to use abstract class (#6235)
- Upgrade AI SDK to v5 (#6242)
- Added vertical layout one column setup with long text in cells to storybook (#6233)
- Add asset URL support for templates and server (#6204)
- Implement timeout option (#6196)
- Fix storybook tailwind compatibility fixes (#6206)
- Fix storybook's missing
requestClientAtomdependency (#6208)
Contributors
Thanks to all our community and contributors who made this release possible: @bjoaquinc, @dmadisetti, @koaning, @Light2Dark, @loginoff, @lucharo, @manzt, @maxkoretskyi, @mscolnick, @tgbrooks, @wlach, @x2764tech, @zaneselvans.
And especially to our new contributors:
- @loginoff made their first contribution in #6202
- @zaneselvans made their first contribution in #6221
- @tgbrooks made their first contribution in #6196
- @wlach made their first contribution in #6217
- @x2764tech made their first contribution in #6315
Full Changelog: 0.15.2...0.15.3