suspend fun queryGemmaAgent() {
    val gemmaServiceUrl = System.getenv("GEMMA_SERVICE_URL") ?: "https://sjjuuy.run.app"

    // 1. Fetch fresh OIDC identity token for this invocation
    val oidcToken = fetchCloudRunIdToken(gemmaServiceUrl)

    // 2. Initialize HTTP Client
    val client = HttpClient(CIO) {
        install(ContentNegotiation) {
            json(Json { ignoreUnknownKeys = true })

        }
        install(HttpTimeout) {
            requestTimeoutMillis = 30_000L  // Total time for the request (30s)
            connectTimeoutMillis = 10_000L  // Time to establish TCP connection (10s)
            socketTimeoutMillis = 30_000L   // Max time between two data packets (30s)
        }
    }



    // 3. Construct the payload
    val payload = ChatRequest(
        model = "google/gemma-4-E4B-it",
        messages = listOf(
            ChatMessage(role = "system", content = "You are a Kotlin-based automation agent."),
            ChatMessage(role = "user", content = "What is the optimal concurrency strategy for Ktor coroutines?")
        )
    )

    // 4. Send authenticated POST request to the inference server
    val response: HttpResponse = client.post("$gemmaServiceUrl/v1/chat/completions") {
        contentType(ContentType.Application.Json)
        // Pass the OIDC token into the header to clear Cloud Run IAM security
        header(HttpHeaders.Authorization, "Bearer $oidcToken")
        setBody(payload)
    }

    println("Gemma 4 Response: ${response.bodyAsText()}")
    client.close()
}