Summary
The Braintrust Go SDK instruments google.golang.org/genai for direct Gemini model inference and cloud.google.com/go/aiplatform is already tracked as a gap for Vertex AI Reasoning Engine / RAG execution (issue #168) — but neither covers Google Cloud Discovery Engine (cloud.google.com/go/discoveryengine), the Go client for Vertex AI Search / Agentspace's grounded-generation and conversational-search product. This is a distinct Go module with its own execution surface that is not reachable through either existing integration.
What is missing
A trace/contrib/discoveryengine/ integration module wrapping the generative execution methods on ConversationalSearchClient (package cloud.google.com/go/discoveryengine/apiv1):
AnswerQuery — grounded answer generation: retrieves from a Discovery Engine data store and generates a natural-language answer via an LLM (the core "RAG-as-a-service" execution call for this product)
StreamAnswerQuery — streaming variant of the above
ConverseConversation — multi-turn conversational search execution, maintaining conversation state across turns
These are genuinely new execution surfaces distinct from google.golang.org/genai (direct Gemini calls) and from cloud.google.com/go/aiplatform's ReasoningEngineExecutionClient/VertexRagClient (Vertex AI Agent Builder reasoning engines and RAG corpora) — Discovery Engine is a separate Google Cloud product (Vertex AI Search / Agentspace) with its own data stores, serving configs, and grounded-generation pipeline.
Braintrust docs status
not_found. Braintrust's Go SDK docs list openai, anthropic, bedrockruntime, genai, genkit, adk, cloudwego/eino, langchaingo, and sashabaranov/go-openai as supported trace/contrib integrations. Neither "Discovery Engine" nor "Vertex AI Search" nor "Agentspace" nor "Conversational Search" appears anywhere in the Braintrust tracing docs.
Upstream sources
Weekly downloads: 3,354,124 (as of 2026-08-17; source: https://pypistats.org/api/packages/google-cloud-discoveryengine/recent — the official sibling PyPI package google-cloud-discoveryengine, published by Google Cloud and covering the identical Discovery Engine API surface, used here as the primary-package adoption proxy since the Go module itself is not tracked by a download registry, following the same substitution pattern as prior issues in this repo for Go-only client libraries)
Braintrust docs sources
Local repo files inspected
go.mod — no cloud.google.com/go/discoveryengine dependency
trace/contrib/ — no discoveryengine/ directory; only genai/ and adk/ exist for Google integrations
trace/contrib/all/all.go — no discoveryengine import
trace/contrib/genai/tracegenai.go — routes on :generateContent and :embedContent only; no Discovery Engine paths
examples/ — no Discovery Engine example
- Grep for
discoveryengine, DiscoveryEngine, AnswerQuery, ConversationalSearch across all repo files — zero matches
Summary
The Braintrust Go SDK instruments
google.golang.org/genaifor direct Gemini model inference andcloud.google.com/go/aiplatformis already tracked as a gap for Vertex AI Reasoning Engine / RAG execution (issue #168) — but neither covers Google Cloud Discovery Engine (cloud.google.com/go/discoveryengine), the Go client for Vertex AI Search / Agentspace's grounded-generation and conversational-search product. This is a distinct Go module with its own execution surface that is not reachable through either existing integration.What is missing
A
trace/contrib/discoveryengine/integration module wrapping the generative execution methods onConversationalSearchClient(packagecloud.google.com/go/discoveryengine/apiv1):AnswerQuery— grounded answer generation: retrieves from a Discovery Engine data store and generates a natural-language answer via an LLM (the core "RAG-as-a-service" execution call for this product)StreamAnswerQuery— streaming variant of the aboveConverseConversation— multi-turn conversational search execution, maintaining conversation state across turnsThese are genuinely new execution surfaces distinct from
google.golang.org/genai(direct Gemini calls) and fromcloud.google.com/go/aiplatform'sReasoningEngineExecutionClient/VertexRagClient(Vertex AI Agent Builder reasoning engines and RAG corpora) — Discovery Engine is a separate Google Cloud product (Vertex AI Search / Agentspace) with its own data stores, serving configs, and grounded-generation pipeline.Braintrust docs status
not_found. Braintrust's Go SDK docs list
openai,anthropic,bedrockruntime,genai,genkit,adk,cloudwego/eino,langchaingo, andsashabaranov/go-openaias supportedtrace/contribintegrations. Neither "Discovery Engine" nor "Vertex AI Search" nor "Agentspace" nor "Conversational Search" appears anywhere in the Braintrust tracing docs.Upstream sources
cloud.google.com/go/discoveryengine/apiv1on pkg.go.dev: https://pkg.go.dev/cloud.google.com/go/discoveryengine/apiv1ConversationalSearchClient.AnswerQuery/StreamAnswerQuery/ConverseConversation: https://pkg.go.dev/cloud.google.com/go/discoveryengine/apiv1#ConversationalSearchClientgoogleapis/google-cloud-gomonorepo: https://pkg.go.dev/cloud.google.com/go/discoveryengine?tab=versionsWeekly downloads: 3,354,124 (as of 2026-08-17; source: https://pypistats.org/api/packages/google-cloud-discoveryengine/recent — the official sibling PyPI package
google-cloud-discoveryengine, published by Google Cloud and covering the identical Discovery Engine API surface, used here as the primary-package adoption proxy since the Go module itself is not tracked by a download registry, following the same substitution pattern as prior issues in this repo for Go-only client libraries)Braintrust docs sources
trace/contribGo integrations; nodiscoveryengineentry)Local repo files inspected
go.mod— nocloud.google.com/go/discoveryenginedependencytrace/contrib/— nodiscoveryengine/directory; onlygenai/andadk/exist for Google integrationstrace/contrib/all/all.go— no discoveryengine importtrace/contrib/genai/tracegenai.go— routes on:generateContentand:embedContentonly; no Discovery Engine pathsexamples/— no Discovery Engine examplediscoveryengine,DiscoveryEngine,AnswerQuery,ConversationalSearchacross all repo files — zero matches