We planted eight defects in our own CI. Three were caught. One gate was skipping 133,874 of 149,516 links and reporting green. Here is the command that proves it.
RubyLLM retries POSTs that Faraday would not, and honors Retry-After with no upper bound. A 429 can park a Rails request for as long as the provider says.
RubyLLM is a model client with Rails generators. Langchainrb is retrieval plumbing. Pick by whether you query your own documents, not by provider count.
Add LLM chat to Rails with RubyLLM: one API for OpenAI, Claude, and Ollama, acts_as_chat persistence, tool calling, streaming into Turbo, and real trade-offs.
LLM responses stream for 30 seconds while a Rails thread waits. The arithmetic of Puma threads vs open SSE streams, and when Falcon's fibers earn the switch.
Our production multi-agent LLM pipeline in Rails: eight RubyLLM agents on a 21-line base class, a 16-line loop, and the deploy that starved the database pool.
What a green Rails suite actually asserts when the thing under test is a hosted model: VCR matches on method and URI only, and the model can be retired.
Falcon production tuning: real benchmarks vs Puma, worker-count formulas, memory profiling, GC tuning, connection-pool sizing, and working config examples.
Reading this because something is going wrong?
A free code audit gives you a written assessment of your codebase in plain English.
