[BOT ISSUE] llm.rb gem not instrumented

[braintrust-bot]

Summary

The llm.rb gem (RubyGems: llm.rb) is an actively maintained Ruby toolkit for building LLM-powered systems that is not instrumented by this SDK. It is a distinct library from ruby_llm (crmne/ruby_llm), which is already instrumented.

llm.rb has 69K+ downloads, reached v4.11.0 (April 2026), and provides execution-oriented API surfaces across 8 providers.

What is missing

No instrumentation exists for any llm.rb execution surface. Key APIs that should be instrumented:

Chat / Completions

• LLM::Context#chat — stateful chat with conversation history
• Streaming responses via structured callbacks
• Stateless completions

Tool Calling

• Automatic tool orchestration across providers
• Concurrent tool execution (threads, fibers, async)
• Tool execution during streaming

Embeddings

• LLM::Provider#embed — vector embeddings generation

Agents

• Reusable, preconfigured assistants with tool auto-execution

Supported Providers

OpenAI, Anthropic, Google, DeepSeek, xAI, zAI, Ollama, LlamaCpp

Braintrust docs status

not_found — No mention of llm.rb or llmrb anywhere in Braintrust documentation at https://www.braintrust.dev/docs. The Braintrust SDK integrations page at https://www.braintrust.dev/docs/integrations/sdk-integrations lists "Ruby LLM" (which refers to the ruby_llm gem, a different library).

Upstream sources

• GitHub: https://github.com/llmrb/llm.rb
• RubyGems: https://rubygems.org/gems/llm.rb (69K+ downloads, v4.11.0, April 2026)
• Documentation: https://github.com/llmrb/llm.rb#readme

Local repo files inspected

• lib/braintrust/contrib.rb (lines 195–205) — registers only 4 integrations: OpenAI, RubyOpenAI, RubyLLM, Anthropic. No llm.rb integration.
• lib/braintrust/contrib/ — no directory or files for llm_rb or llmrb
• Appraisals — no appraisal scenarios for llm.rb
• braintrust.gemspec — no mention of llm.rb

Notes

llm.rb has built-in OpenTelemetry support, so users can get basic tracing via Braintrust's OTEL integration. However, dedicated instrumentation would capture richer provider-specific metadata (model parameters, token usage breakdowns, tool execution details, cost tracking) consistent with how the SDK instruments ruby_llm, openai, and anthropic today.