Summary
The dspy gem (dspy.rb, ~70K downloads, v1.0.0, April 2026) is the Ruby port of the DSPy LLM programming framework and is not instrumented by this SDK. Braintrust documents DSPy integration for Python, capturing module-level spans for Predict, ChainOfThought, and other DSPy modules — but there is no equivalent Ruby instrumentation for dspy.rb.
What is missing
No instrumentation exists for any dspy execution surface. Key APIs that should be instrumented:
Module Execution
DSPy::Predict#call(**inputs) — single-pass LLM call with typed input/output contract. The core building block of all DSPy programs.DSPy::ChainOfThought#call(**inputs) — LLM call that surfaces intermediate chain-of-thought reasoning steps before the final answer. Produces a richer trace than Predict due to the reasoning span.DSPy::ReAct#call(**inputs) — agentic reasoning loop that interleaves LLM calls with tool executions. May span multiple LLM turns and tool invocations.
Tool Execution (ReAct)
- Tool calls dispatched within
DSPy::ReAct loops should be captured as child spans, including tool name, inputs, and output.
LM Configuration
DSPy::LM.new('openai/gpt-4o-mini', ...) — LM adapter configuration, useful for capturing which model and provider is driving each module.
Expected instrumentation
Module execution spans should capture:
- Module type (Predict, ChainOfThought, ReAct)
- Signature type name (typed input/output schema)
- Input fields and values
- LLM calls made (model, provider, tokens)
- Intermediate reasoning steps (for ChainOfThought)
- Tool calls and results (for ReAct)
- Typed output fields
Braintrust docs status
unclear — Braintrust documents DSPy integration at https://www.braintrust.dev/docs/integrations/sdk-integrations/dspy and captures Predict, ChainOfThought, and other DSPy module spans with full tracing. However, the documented integration requires dspy>=2.6.0 (Python) and covers only Python. No Ruby-specific dspy.rb instrumentation is documented.
Upstream sources
Local repo files inspected
lib/braintrust/contrib.rb (lines 195–205) — registers only 4 integrations: OpenAI, RubyOpenAI, RubyLLM, Anthropic. No dspy integration.lib/braintrust/contrib/ — no directory or files for dspyAppraisals — no appraisal scenarios for dspybraintrust.gemspec — no mention of dspy- Grep for
dspy, DSPy, Predict, ChainOfThought, ReAct across entire codebase returns zero matches
Summary
The
dspygem (dspy.rb, ~70K downloads, v1.0.0, April 2026) is the Ruby port of the DSPy LLM programming framework and is not instrumented by this SDK. Braintrust documents DSPy integration for Python, capturing module-level spans forPredict,ChainOfThought, and other DSPy modules — but there is no equivalent Ruby instrumentation fordspy.rb.What is missing
No instrumentation exists for any
dspyexecution surface. Key APIs that should be instrumented:Module Execution
DSPy::Predict#call(**inputs)— single-pass LLM call with typed input/output contract. The core building block of all DSPy programs.DSPy::ChainOfThought#call(**inputs)— LLM call that surfaces intermediate chain-of-thought reasoning steps before the final answer. Produces a richer trace thanPredictdue to the reasoning span.DSPy::ReAct#call(**inputs)— agentic reasoning loop that interleaves LLM calls with tool executions. May span multiple LLM turns and tool invocations.Tool Execution (ReAct)
DSPy::ReActloops should be captured as child spans, including tool name, inputs, and output.LM Configuration
DSPy::LM.new('openai/gpt-4o-mini', ...)— LM adapter configuration, useful for capturing which model and provider is driving each module.Expected instrumentation
Module execution spans should capture:
Braintrust docs status
unclear— Braintrust documents DSPy integration at https://www.braintrust.dev/docs/integrations/sdk-integrations/dspy and capturesPredict,ChainOfThought, and other DSPy module spans with full tracing. However, the documented integration requiresdspy>=2.6.0(Python) and covers only Python. No Ruby-specificdspy.rbinstrumentation is documented.Upstream sources
Local repo files inspected
lib/braintrust/contrib.rb(lines 195–205) — registers only 4 integrations: OpenAI, RubyOpenAI, RubyLLM, Anthropic. Nodspyintegration.lib/braintrust/contrib/— no directory or files fordspyAppraisals— no appraisal scenarios fordspybraintrust.gemspec— no mention ofdspydspy,DSPy,Predict,ChainOfThought,ReActacross entire codebase returns zero matches