Lib.rs

runtimed

RuntimeD is a daemon for REPLs built on top of Jupyter kernels. It's purpose built for exposing interactive computing primitives to large language models whether hosted or local.

The main CLI for interfacing with runtimed is runt.

Goal

The goal of runt is to provide simple, easy to use, and powerful access to interactive computing. We want to enable a new generation of builders to:

• Create new notebook applications
• Create new kinds of REPLs
• Allow large language models to reason about code and data

There are three main interfaces:

• runt - a CLI for managing runtimes
• runtimed - a daemon for working with the interactive computing runtimes
• runtimelib - a rust library for interfacing with runtimes directly

Getting Started with runtimelib

cargo install runtimelib

Asynchronous dispatch options

By default, runtimelib uses tokio. However, the async-dispatcher runtime can be selected at compile time with:

cargo build --feature async-dispatch-runtime

This will allow you to build GPUI apps with runtimelib.

Development - getting started

git clone git@github.com:runtimed/runtimed.git
cd runtimed
# Install the cli, `runt` into your path
cargo install --path runt
# Install the CLI for the `runtimed` daemon.
cargo install --path runtimed
# Start the daemon
runtimed

Usage

If you haven't already, start the runtimed daemon.

$ runtimed

List the available runtimes.

$ runt ps
╭─────────────┬────────────┬──────────────────────────────────────┬───────────┬───────────┬───────╮
│ Kernel Name │ Language   │ ID                                   │ IP        │ Transport │ State │
├─────────────┼────────────┼──────────────────────────────────────┼───────────┼───────────┼───────┤
│ python3     │ python     │ 6a090dec-08cb-5429-a6c7-ea19d71fc06e │ 127.0.0.1 │ tcp       │ alive │
│ deno        │ typescript │ 79c4c28f-1ffb-579a-b77c-23e4a1bb45ec │ 127.0.0.1 │ tcp       │ alive │
╰─────────────┴────────────┴──────────────────────────────────────┴───────────┴───────────┴───────╯

Submitting an Execution

$ runt exec 79c4c28f-1ffb-579a-b77c-23e4a1bb45ec $'function X() { return Math.random() };\nX()'
Execution "2d3827a2-4a7f-4a1f-bc41-5091f9ade2ab" submitted, run

runt get-results "2d3827a2-4a7f-4a1f-bc41-5091f9ade2ab"

to get the results of the execution.

Getting Results

$ runt get-results "2d3827a2-4a7f-4a1f-bc41-5091f9ade2ab"
╭────────────────────────────────────────────────────╮
│ Execution Results                                  │
├────────────────────────────────────────────────────┤
│ Execution ID: 2d3827a2-4a7f-4a1f-bc41-5091f9ade2ab │
│ Status: idle                                       │
│ Started: 2024-03-05T23:57:46.680992Z               │
│ Finished: 2024-03-05T23:57:46.688572Z              │
│                                                    │
│ -- Code --                                         │
│ function X() { return Math.random() };             │
│ X()                                                │
│                                                    │
│ -- Output --                                       │
│ 0.21616865512200545                                │
│                                                    │
╰────────────────────────────────────────────────────╯

The idea behind the runtimed API 💡

We're exposing a document oriented interface to working with kernels, as a REST API.

RuntimeD tracks executions of runtimes for recall and for working with interactive applications like notebooks and consoles. We track the association between Execution and Runtime (a running kernel). We also track for specific notebook apps with a Code Cell -> Execution.

Execution {
  id: ULID,
  execution_started: timestamp,
  execution_end: timestamp,
  status: running | queued | ...
  runtime: Runtime
}

Runtime {
  id: ULID,
  status: dead | alive | unresponsive,
  last_keepalive: timestamp
}

CodeCell {
  src: str,
  execution: Execution
}

Development

Working with the DB

The database is managed by the sqlx library. The db is created and any migrations are run automatically. If you are updating the schema or add more queries to the app, more tooling needs to be installed.

cargo install sqlx-cli

ln -s .env.example .env

Preparing new queries

After you write a new query, you will need to run

cargo sqlx prepare --workspace

Migrations

New migrations should be added with the following command:

cargo sqlx migrate add create_executions_table

This will create a new migration file in migrations/ that you can edit. After you are done, you can run the migration with:

cargo sqlx migrate run