runtimed 2.1.1a202604040259

Python toolkit for Jupyter runtimes, powered by runtimed Rust binaries

runtimed

Python bindings for the nteract runtime daemon. Execute code, manage kernels, and interact with notebooks programmatically.

Download the nteract desktop app — it ships the runtimed daemon and gives you a visual interface for your notebooks.

Using runtimed with agents? The nteract MCP server is built on runtimed and provides a ready-made agentic interface for AI assistants. It's also a great example of how to use runtimed in practice.

Installation

pip install runtimed

The stable release matches the nteract desktop stable app. If you're running the nightly desktop app, install the pre-release to match: pip install --pre runtimed (or uv pip install --prerelease allow runtimed). The nightly build automatically discovers the nightly daemon socket.

Client() and the high-level Python API use default_socket_path() by default. That helper respects RUNTIMED_SOCKET_PATH, so exported test or MCP sockets take precedence over the package's default channel.

Quick Start

All examples use await — run them inside asyncio.run(main()), a Jupyter notebook, or a Python REPL with top-level await (e.g. python -m asyncio).

import asyncio
import runtimed

async def main():
    client = runtimed.Client()
    notebook = await client.create_notebook()

    # Create and execute cells
    cell = await notebook.cells.create("print('hello')")
    result = await cell.run()
    print(result.stdout)  # "hello\n"

    # Read cell properties (sync — local CRDT replica)
    print(cell.source)      # "print('hello')"
    print(cell.cell_type)   # "code"

    # Edit cells
    await cell.set_source("x = 42")
    await cell.run()

    # Save the notebook
    path = await notebook.save_as("/tmp/my-notebook.ipynb")

asyncio.run(main())

Features

• Document-first model with Automerge CRDT sync
• Sync reads, async writes — reads from local replica, writes sync to peers
• Multi-client support for shared notebooks
• Rich output capture (stdout, stderr, display_data, errors)

API Overview

Client

client = runtimed.Client()

# Discover active notebooks
notebooks = await client.list_active_notebooks()
for info in notebooks:
    print(f"{info.name} [{info.status}] ({info.active_peers} peers)")

# Open, create, or join notebooks
notebook = await client.open_notebook("/path/to/notebook.ipynb")
notebook = await client.create_notebook(runtime="python")
notebook = await client.join_notebook(notebook_id)

If you need to target a specific release channel instead of the current process default:

import os
import runtimed

os.environ["RUNTIMED_SOCKET_PATH"] = runtimed.socket_path_for_channel("nightly")
client = runtimed.Client()

Use default_socket_path() for normal current-process behavior. Use socket_path_for_channel("stable"|"nightly") only for explicit channel targeting or cross-channel discovery because it intentionally ignores RUNTIMED_SOCKET_PATH.

Notebook

async with await client.create_notebook() as notebook:
    # Cells collection (sync reads, async writes)
    print(len(notebook.cells))
    for cell in notebook.cells:
        print(f"{cell.id[:8]}: {cell.source[:40]}")

    # Runtime state (sync read from local doc)
    print(notebook.runtime.kernel.status)

    # Runtime lifecycle
    await notebook.start(runtime="python")
    await notebook.restart()
    await notebook.interrupt()
    await notebook.save()
# Session closed automatically on exit

Cells

# Create cells
cell = await notebook.cells.create("import math")
cell = await notebook.cells.insert_at(0, "# Title", cell_type="markdown")

# Access cells
cell = notebook.cells.get_by_index(0)    # by position
cell = notebook.cells.get_by_id(cell_id) # by ID
matches = notebook.cells.find("import")  # search source

# Read properties (sync)
print(cell.source, cell.cell_type, cell.outputs)

# Mutate (async)
await cell.set_source("x = 2")
await cell.append("\ny = 3")
result = await cell.run()
await cell.delete()

Requirements

• The runtimed daemon, which ships with the nteract desktop app. For development, see the nteract/desktop repo.
• Python 3.10+

Documentation

See docs/python-bindings.md for full documentation.