Add Rails engine for remote eval server

[Stephen Belanger (Qard)]

Summary

• Extracts Services::ListService and Services::EvalService from the existing Rack handlers so business logic is shared between the Rack and Rails adapters with no duplication
• Adds Braintrust::Server::Rails::Engine — a mountable Rails::Engine using ActionController::API and ActionController::Live for SSE streaming
• Reuses existing auth strategies (ClerkToken, NoAuth) and CORS middleware unchanged
• Adds a rails-server appraisal (actionpack ~> 8.0, rack-test) and full test coverage for services and engine controllers

Usage

# config/initializers/braintrust_server.rb
require "braintrust/server/rails"

Braintrust::Server::Rails::Engine.configure do |config|
  config.evaluators = { "my-eval" => MyEval.new }
  config.auth = :clerk_token  # or :none
end

# config/routes.rb
mount Braintrust::Server::Rails::Engine, at: "/braintrust"

Test plan

• bundle exec appraisal server rake test — existing Rack server tests pass (0 failures)
• bundle exec appraisal rails-server rake test — new Rails engine tests pass (0 failures)
• bundle exec standardrb — linting clean

🤖 Generated with Claude Code

[David Elner (delner)]

Overall I think this is off to a great start! I left some comments of some changes that are probably worth making, the most important of which would be sorting out names/namespaces.

I also would request that we add:

1. A readme update: probably to the existing Dev Server section, perhaps as its own subsection. We can highlight Running as a Rack app vs Running as a Rails Engine.
2. A Rails generator that scans the evaluators/ folder and generates the braintrust_server config for the Engine: presumably it would link "slug --> Evaluator class" for each found where the slug is a formatted class name by default. The sum UX would then be (1) define your Evaluators in evaluators/ then (2) run the generator.

[David Elner (delner)]

Looking really good! I particularly like the service object helpers. Most of my code feedback is minor. The only change I think we should make before merging this is adding something to the README. Otherwise I'm happy with this.

Resolved.

[David Elner (delner)]

Very happy with this. Thank you Stephen Belanger (@Qard) for your diligence (and patience!) getting this put together :)