Hi! I'm building a Next.js app using BAML with the

vertex-ai

provider and I want to leverage some Google-specific features directly from my

clients.baml

configuration. My use cases — I want to know if and how these are supported: 1. Google Search grounding 2. URL Context 3. RAG Engine Grounding (Vertex RAG Engine with

ragCorpora

) 4. Cached Content (explicit caching with a pre-created cache name) 5. File Inputs — both from GCS (

gs://

URLs) and runtime uploads (user-uploaded files) What I understand so far: - BAML's

vertex-ai

provider does pass-through of unknown

options

fields to the Vertex REST API -

generationConfig

and

safetySettings

are documented and work - For the features above, I'm not sure if the pass-through covers them, or if there's a different recommended approach in BAML My question: Is there an officially supported way to use these Vertex AI features directly from

clients.baml

? If so, what's the correct syntax? If not, what's the recommended pattern for integrating them in a BAML-based project? Thanks! 🙏

let me get back to you on this!

Hey Aaron! Just following up on this — no rush, I know you're busy. For additional context: I'm looking for something similar to what the Vercel AI SDK does with its Google provider, where you can pass tools like google.tools.googleSearch({}), google.tools.urlContext({}) or vertex.tools.vertexRagStore({}) directly in the call config. Ideally I'd love to have that same kind of first-class support declaratively in clients.baml — either as named tool options inside the client block, or through any other BAML-native pattern. Is this something on the roadmap, or is there a recommended way to handle it today? Thanks again! 🙏

the passthrough should cover any websearch tools as well, I dont have the format handy yet though