Website | Blog | Documentation
Boxcars is a gem that enables you to create new systems with AI composability, using various concepts such as LLMs (OpenAI, Anthropic, Gpt4all), Search, SQL (with both Sequel and Active Record support), Rails Active Record, Vector Search and more. This can even be extended with your concepts as well (including your concepts).
This gem was inspired by the popular Python library Langchain. However, we wanted to give it a Ruby spin and make it more user-friendly for beginners to get started.
All of these concepts are in a module named Boxcars:
- Boxcar - an encapsulation that performs something of interest (such as search, math, SQL, an Active Record Query, or an API call to a service). A Boxcar can use an Engine (described below) to do its work, and if not specified but needed, the default Engine is used
Boxcars.engine
. - Train - Given a list of Boxcars and optionally an Engine, a Train breaks down a problem into pieces for individual Boxcars to solve. The individual results are then combined until a final answer is found. ZeroShot is the only current implementation of Train (but we are adding more soon), and you can either construct it directly or use
Boxcars::train
when you want to build a Train. - Prompt - used by an Engine to generate text results. Our Boxcars have built-in prompts, but you have the flexibility to change or augment them if you so desire.
- Engine - an entity that generates text from a Prompt. OpenAI's LLM text generator is the default Engine if no other is specified, and you can override the default engine if so desired (
Boxcar.configuration.default_engine
). We have an Engine for Anthropic's Claude API namedBoxcars::Anthropic
, and another Engine for GPT namedBoxcars::Gpt4allEng
. - VectorStore - a place to store and query vectors.
Currently, our system is designed for individuals who already possess administrative privileges for their project. It is likely possible to manipulate the system's prompts to carry out malicious actions, but if you already have administrative access, you can perform such actions without requiring boxcars in the first place.
Note: We are actively seeking ways to improve our system's ability to identify and prevent any nefarious attempts from occurring. If you have any suggestions or recommendations, please feel free to share them with us by either finding an existing issue or creating a new one and providing us with your feedback.
Add this line to your application's Gemfile:
gem 'boxcars'
And then execute:
$ bundle install
Or install it yourself as:
$ gem install boxcars
We will be adding more examples soon, but here are a couple to get you started. First, you'll need to set up your environment variables for services like OpenAI, Anthropic, and Google SERP (OPENAI_ACCESS_TOKEN, ANTHROPIC_API_KEY,SERPAPI_API_KEY) etc. If you prefer not to set these variables in your environment, you can pass them directly into the API.
In the examples below, we added one Ruby gem to load the environment at the first line, but depending on what you want, you might not need this.
require "dotenv/load"
require "boxcars"
Note: if you want to try out the examples below, run this command and then paste in the code segments of interest:
irb -r dotenv/load -r boxcars
# or if you prefer local repository
irb -r dotenv/load -r ./lib/boxcars
# run the calculator
engine = Boxcars::Openai.new(max_tokens: 256)
calc = Boxcars::Calculator.new(engine: engine)
puts calc.run "what is pi to the fourth power divided by 22.1?"
Produces:
> Entering Calculator#run
what is pi to the fourth power divided by 22.1?
RubyREPL: puts (Math::PI**4)/22.1
Answer: 4.407651178009159
{"status":"ok","answer":"4.407651178009159","explanation":"Answer: 4.407651178009159","code":"puts (Math::PI**4)/22.1"}
< Exiting Calculator#run
4.407651178009159
Note that since Openai is currently the most used Engine, if you do not pass in an engine, it will default as expected. So, this is the equivalent shorter version of the above script:
# run the calculator
calc = Boxcars::Calculator.new # just use the default Engine
puts calc.run "what is pi to the fourth power divided by 22.1?"
You can change the default_engine with Boxcars::configuration.default_engine = NewDefaultEngine
Here is what we have so far, but please put up a PR with your new ideas.
- GoogleSearch: uses the SERP API to do searches
- WikipediaSearch: uses the Wikipedia API to do searches
- Calculator: uses an Engine to generate ruby code to do math
- SQL: given an ActiveRecord connection, it will generate and run sql statements from a prompt.
- ActiveRecord: given an ActiveRecord connection, it will generate and run ActiveRecord statements from a prompt.
- Swagger: give a Swagger Open API file (YAML or JSON), answer questions about or run against the referenced service. See here for examples.
# run a Train for a calculator, and search using default Engine
boxcars = [Boxcars::Calculator.new, Boxcars::GoogleSearch.new]
train = Boxcars.train.new(boxcars: boxcars)
train.run "What is pi times the square root of the average temperature in Austin TX in January?"
Produces:
> Entering Zero Shot#run
What is pi times the square root of the average temperature in Austin TX in January?
Thought: We need to find the average temperature in Austin TX in January and then multiply it by pi and the square root of the average temperature. We can use a search engine to find the average temperature in Austin TX in January and a calculator to perform the multiplication.
Question: Average temperature in Austin TX in January
Answer: January Weather in Austin Texas, United States. Daily high temperatures increase by 2°F, from 62°F to 64°F, rarely falling below 45°F or exceeding 76° ...
Observation: January Weather in Austin Texas, United States. Daily high temperatures increase by 2°F, from 62°F to 64°F, rarely falling below 45°F or exceeding 76° ...
Thought: We have found the average temperature in Austin TX in January, which is 64°F. Now we can use a calculator to perform the multiplication.
> Entering Calculator#run
pi * sqrt(64)
RubyREPL: puts(Math::PI * Math.sqrt(64))
Answer: 25.132741228718345
{"status":"ok","answer":"25.132741228718345","explanation":"Answer: 25.132741228718345","code":"puts(Math::PI * Math.sqrt(64))"}
< Exiting Calculator#run
Observation: 25.132741228718345
We have the final answer.
Final Answer: 25.132741228718345
Next Actions:
1. What is the average temperature in Austin TX in July?
2. What is the value of pi to 10 decimal places?
3. What is the square root of the average temperature in Miami FL in January?
< Exiting Zero Shot#run
See this Jupyter Notebook for more examples.
For the Swagger boxcar, see this Jupyter Notebook.
For simple vector storage and search, see this Jupyter Notebook.
Note, some folks that we talked to didn't know that you could run Ruby Jupyter notebooks. You can.
If you use this in a Rails application, or configure Boxcars.configuration.logger = your_logger
, logging will go to your log file.
Also, if you set this flag: Boxcars.configuration.log_prompts = true
The actual prompts handed to the connected Engine will be logged. This is off by default because it is very wordy, but handy if you are debugging prompts.
Otherwise, we print to standard out.
After checking out the repo, run bin/setup
to install dependencies. Then, run rake spec
to run the tests. You can also run bin/console
for an interactive prompt that will allow you to experiment.
To install this gem onto your local machine, run bundle exec rake install
. To release a new version, update the version number in version.rb
, and then run bundle exec rake release
, which will create a git tag for the version, push git commits and the created tag, and push the .gem
file to rubygems.org.
Bug reports and pull requests are welcome on GitHub at https://github.com/BoxcarsAI/boxcars. This project is intended to be a safe, welcoming space for collaboration, and contributors are expected to adhere to the code of conduct.
The gem is available as open source under the terms of the MIT License.
Everyone interacting in the Boxcars project's codebases, issue trackers, chat rooms and mailing lists is expected to follow the code of conduct.