Generative AI & ML

Emulates managed AI/ML platforms: Bedrock and SageMaker (AWS) and Vertex AI (GCP). Each ships a control plane (model catalog, endpoints, jobs) and a deterministic runtime so inference calls return predictable, echo-based responses — no GPUs, no accounts, no cost.

Responses are deterministic emulations, not real model output. Runtime calls echo the prompt back (with model-native envelopes and plausible token counts) so you can test wiring, retries, streaming setup, and error handling without a live model.

Use the real SDK (recommended)

import (
    "github.com/aws/aws-sdk-go-v2/service/bedrockruntime"
    "github.com/stackshy/cloudemu"
    awsserver "github.com/stackshy/cloudemu/server/aws"
)

cloud := cloudemu.NewAWS()
ts := httptest.NewServer(awsserver.New(awsserver.Drivers{
    Bedrock:   cloud.Bedrock,
    SageMaker: cloud.SageMaker,
}))

rt := bedrockruntime.NewFromConfig(cfg, func(o *bedrockruntime.Options) {
    o.BaseEndpoint = aws.String(ts.URL)
})

rt.InvokeModel(ctx, &bedrockruntime.InvokeModelInput{
    ModelId: aws.String("anthropic.claude-3-sonnet-20240229-v1:0"),
    Body:    []byte(`{"messages":[{"role":"user","content":"hello"}]}`),
})

Vertex AI speaks the aiplatform REST API (generateContent, predict). See the SDK-Compat Server page.

Operations supported via SDK-compat

Bedrock — control plane: ListFoundationModels, GetFoundationModel; custom-model customization jobs; custom models (List/Get/Delete); guardrails (Create/Get/List/Update/Delete); provisioned throughput (Create/Get/List/Delete); model-invocation logging config (Put/Get/Delete). Runtime: InvokeModel, Converse. Seeded model families: Anthropic Claude, Amazon Titan, Meta Llama, Cohere Command, plus Titan embedding models.

SageMaker — control plane (X-Amz-Target: SageMaker.*): models, endpoint configs, endpoints (incl. UpdateEndpointWeightsAndCapacities), inference components, training / processing / transform / tuning / AutoML / labeling / compilation jobs (complete synchronously), model registry, Studio, notebook instances, HyperPod, Feature Store, pipelines, tagging. Runtime: InvokeEndpoint (sync), InvokeEndpointAsync (returns a synthetic S3 output URI).

Vertex AI — control plane (REST /v1/projects/{p}/locations/{l}/…): models, endpoints (incl. DeployModel/UndeployModel), datasets, custom / batch-prediction / hyperparameter-tuning / training-pipeline / pipeline / tuning jobs, cached contents, Feature Store, Vector Search, metadata, tensorboards, schedules, notebook runtimes. Runtime: GenerateContent (Gemini), CountTokens, Predict, RawPredict.

Realistic behaviors

• Deterministic runtime: InvokeModel / Converse / GenerateContent / InvokeEndpoint echo the input with model-family-specific response envelopes and whitespace-based token counts, so assertions are stable across runs.
• Embeddings: Bedrock embedding models return fixed-dimension vectors seeded by input length.
• Jobs complete synchronously: training, tuning, batch-prediction, and customization jobs are driven straight to their terminal success state — no polling loops in tests.
• LROs done-on-arrival: Vertex AI long-running operations return with done: true and the result inlined.
• Endpoint validation: InvokeEndpoint requires the endpoint to exist and be InService.

Alternative: Portable Go API

import bedrockdriver "github.com/stackshy/cloudemu/bedrock/driver"

models, _ := aws.Bedrock.ListFoundationModels(ctx)

out, _ := aws.Bedrock.InvokeModel(ctx, bedrockdriver.InvokeModelInput{
    ModelID: "anthropic.claude-3-sonnet-20240229-v1:0",
    Body:    []byte(`{"prompt":"hello"}`),
})