Skip to content
Dalliance
BlogGitHubDownload for macOS

Early preview

The Open-source Coding Agent for Researchers

Like a Jupyter notebook, but you write each cell in plain language. Dalliance writes the Python, runs it and keeps the code that works.

Download for macOSGitHub
Or in Terminal
curl -fsSL https://dalliance.io/install.sh | sh

Try it. Click an underlined guess to change it, open the code, or run a suggested prompt.

Survival analysisSonnet 5.5 · med
Notebook1 prompt

Overall survival

Does the new regimen extend survival over standard care once age and stage are accounted for?

Survival by treatment arm from patients.csv, with a log-rank test

Survival is time to death from , in , compared with a .

5.9s

Median survival was 31.1 months on the new regimen and 21.6 on standard care. The gap is significant, log-rank p = .004.

Overall survival by armMonths since randomisationSurvival probability0122436480.00.20.40.60.81.0New regimenStandard care

Createdpatients412 × 10

In short

Ask a question in plain words. Dalliance answers in a sentence and shows the table and the code behind it. When a prompt could mean two things, it asks you first.

A trial readout notebook. The cell "Median survival and 2-year survival by arm" shows the agent's reading of the prompt underlined, a one-line answer, the survival table and three suggested next prompts.
Ask in words
A cell mid-run. The agent found 38 patients who switched arms and asks whether to count them by intention to treat or per protocol before it writes any code.
It asks first
Correcting an underlined assumption in place, measuring survival from first dose instead of randomisation. The next run uses the correction.
Fix a guess
The Code tab of a cell, with the Kaplan-Meier code the agent wrote and ran.
Read the code
The History pane beside the notebook. An earlier version took a plain median that ignored censoring; the current one uses Kaplan-Meier.
Go back
The model picker, grouped by maker, with an effort setting below.
Any model

Apps

Turn your analysis into an app.

Describe what you want to explore. Dalliance builds an interactive app from your notebook data and opens it inside your project. The Python code is yours to keep.

Create an app to explore outcomes by study group.
AppsStudy explorer

Outcomes over time

A closer look at each study group
Records24
Latest mean score35.5
02550BaselineWeek 4Week 8
Interactive preview · illustrative data

Saved with your project. Built from a data snapshot you can refresh when your notebook changes.

What it does

  1. Runs and checks. The agent runs the code, reads the output or the error, and retries until the cell works. Failed code never lands in your notebook.
  2. Shows its guesses. Under each prompt it writes how it read it, with every assumption underlined. Click one to correct it and the next run uses your version.
  3. Asks when it matters. If a wrong guess would change the result, it checks the data and asks you before writing code.
  4. Keeps the code. An unchanged cell reruns its saved code directly, with no model call. Edit the code yourself and your version stays.
  5. Keeps every version. Each new version of a cell's code is saved. Restore any of them in a click.
  6. Reads your brief. Put notes like "months are from randomisation, not diagnosis" in AGENTS.md and the agent follows them in every cell.

What it needs

A Mac with Python. Notebook cells need Python 3.9 or newer. Apps use a separate environment with a compatible installed Python 3.11 or newer. Dalliance can create a private notebook environment with pandas, NumPy and matplotlib, or use an interpreter you already have.

Open source. Your own API key. Connect Anthropic, Google, or OpenAI. Requests go directly to your provider, and you pay for your own API usage.

What it is not

Not a chat. A chat answers once and scrolls away. Dalliance keeps each answer next to its code, in a notebook you can rerun top to bottom.

Not a black box. Every cell has a Code tab with the exact Python that ran. Notebooks save as standard .ipynb files and open in Jupyter.

Python runs locally. Your notebooks and Python kernel stay on your Mac. Your prompt and the context the agent reads go directly to your chosen provider, under your account's data policies.

Status

Dalliance is an early preview, so expect rough edges. Your notebooks are plain .ipynb files, so nothing is locked in.

From the blog

All posts

Write the question. Keep the code.

Download for macOSGitHub
Or in Terminal
curl -fsSL https://dalliance.io/install.sh | sh