Dalliance, An open-source coding agent for researchers
An open-source research notebook that turns plain-language questions into Python you can inspect and rerun.
| When | Selected project |
|---|---|
| Role | Creator & engineer |
| Status | Early preview · Open source |
| Link | dalliance.io ↗ |
I built Dalliance for researchers who want to work with data in a notebook using plain language. Each cell contains a question, the Python that answers it, and the result.
The agent writes and runs Python, checks the output, and retries when code fails. It shows its interpretation of a prompt as editable assumptions and asks for clarification when a wrong guess could change the result.
Working code stays in the notebook. Unchanged cells rerun their saved code without another model call, manual edits are preserved, and earlier versions can be restored. Notebooks save as standard .ipynb files that open in Jupyter.
Dalliance runs on macOS with a local Python kernel. Users connect their own Anthropic, Google, or OpenAI account; prompts and the context used by the agent go directly to the selected provider. Project instructions in AGENTS.md guide the agent across cells.
The app is open source and available as an early preview.