diff --git a/llama_stack/apis/datatypes.py b/llama_stack/apis/datatypes.py
index 597572758..a373d8165 100644
--- a/llama_stack/apis/datatypes.py
+++ b/llama_stack/apis/datatypes.py
@@ -22,6 +22,7 @@ class Api(Enum):
     datasetio = "datasetio"
     post_training = "post_training"
     tool_runtime = "tool_runtime"
+    evaluation = "evaluation"
 
     telemetry = "telemetry"
 
diff --git a/llama_stack/providers/inline/evaluation/meta_reference/__init__.py b/llama_stack/providers/inline/evaluation/meta_reference/__init__.py
new file mode 100644
index 000000000..bf5f5a6fa
--- /dev/null
+++ b/llama_stack/providers/inline/evaluation/meta_reference/__init__.py
@@ -0,0 +1,27 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+from typing import Any, Dict
+
+from llama_stack.distribution.datatypes import Api
+
+from .config import MetaReferenceEvaluationConfig
+
+
+async def get_provider_impl(
+    config: MetaReferenceEvaluationConfig,
+    deps: Dict[Api, Any],
+):
+    from .evaluation import MetaReferenceEvaluationImpl
+
+    impl = MetaReferenceEvaluationImpl(
+        config,
+        deps[Api.datasetio],
+        deps[Api.datasets],
+        deps[Api.inference],
+        deps[Api.agents],
+    )
+    await impl.initialize()
+    return impl
diff --git a/llama_stack/providers/inline/evaluation/meta_reference/config.py b/llama_stack/providers/inline/evaluation/meta_reference/config.py
new file mode 100644
index 000000000..024f0c168
--- /dev/null
+++ b/llama_stack/providers/inline/evaluation/meta_reference/config.py
@@ -0,0 +1,26 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+from typing import Any, Dict
+
+from llama_stack.providers.utils.kvstore.config import (
+    KVStoreConfig,
+    SqliteKVStoreConfig,
+)
+
+from pydantic import BaseModel
+
+
+class MetaReferenceEvaluationConfig(BaseModel):
+    kvstore: KVStoreConfig
+
+    @classmethod
+    def sample_run_config(cls, __distro_dir__: str, **kwargs: Any) -> Dict[str, Any]:
+        return {
+            "kvstore": SqliteKVStoreConfig.sample_run_config(
+                __distro_dir__=__distro_dir__,
+                db_name="meta_reference_evaluation.db",
+            )
+        }
diff --git a/llama_stack/providers/inline/evaluation/meta_reference/evaluation.py b/llama_stack/providers/inline/evaluation/meta_reference/evaluation.py
new file mode 100644
index 000000000..1e5ffc3b5
--- /dev/null
+++ b/llama_stack/providers/inline/evaluation/meta_reference/evaluation.py
@@ -0,0 +1,77 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+import json
+from typing import Any, Dict, List, Optional
+
+from llama_stack.apis.agents import Agents
+from llama_stack.apis.datasetio import DatasetIO
+from llama_stack.apis.datasets import Datasets
+from llama_stack.apis.inference import Inference
+from llama_stack.providers.datatypes import BenchmarksProtocolPrivate
+
+from tqdm import tqdm
+
+from .....apis.benchmarks import Benchmark
+
+from .....apis.common.job_types import Job
+from .....apis.evaluation.evaluation import (
+    Evaluation,
+    EvaluationCandidate,
+    EvaluationJob,
+    EvaluationResponse,
+    EvaluationTask,
+)
+from .config import MetaReferenceEvaluationConfig
+
+EVAL_TASKS_PREFIX = "benchmarks:"
+
+
+class MetaReferenceEvaluationImpl(
+    Evaluation,
+    BenchmarksProtocolPrivate,
+):
+    def __init__(
+        self,
+        config: MetaReferenceEvaluationConfig,
+        datasetio_api: DatasetIO,
+        datasets_api: Datasets,
+        inference_api: Inference,
+        agents_api: Agents,
+    ) -> None:
+        self.config = config
+        self.datasetio_api = datasetio_api
+        self.datasets_api = datasets_api
+        self.inference_api = inference_api
+        self.agents_api = agents_api
+
+    async def initialize(self) -> None:
+        pass
+
+    async def shutdown(self) -> None:
+        pass
+
+    async def register_benchmark(self, benchmark: Benchmark) -> None:
+        pass
+
+    async def run(
+        self,
+        task: EvaluationTask,
+        candidate: EvaluationCandidate,
+    ) -> EvaluationJob:
+        raise NotImplementedError("Run is not implemented yet")
+
+    async def run_sync(
+        self,
+        task: EvaluationTask,
+        candidate: EvaluationCandidate,
+    ) -> EvaluationResponse:
+        raise NotImplementedError("Run sync is not implemented yet")
+
+    async def grade(self, task: EvaluationTask) -> EvaluationJob:
+        raise NotImplementedError("Grade is not implemented yet")
+
+    async def grade_sync(self, task: EvaluationTask) -> EvaluationResponse:
+        raise NotImplementedError("Grade sync is not implemented yet")
diff --git a/llama_stack/providers/registry/evaluation.py b/llama_stack/providers/registry/evaluation.py
new file mode 100644
index 000000000..2481cea5a
--- /dev/null
+++ b/llama_stack/providers/registry/evaluation.py
@@ -0,0 +1,36 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from typing import List
+
+from llama_stack.providers.datatypes import Api, InlineProviderSpec, ProviderSpec
+from llama_stack.providers.utils.kvstore import kvstore_dependencies
+
+
+def available_providers() -> List[ProviderSpec]:
+    return [
+        InlineProviderSpec(
+            api=Api.evaluation,
+            provider_type="inline::meta-reference",
+            pip_packages=[
+                "matplotlib",
+                "pillow",
+                "pandas",
+                "scikit-learn",
+            ]
+            + kvstore_dependencies(),
+            module="llama_stack.providers.inline.evaluation.meta_reference",
+            config_class="llama_stack.providers.inline.evaluation.meta_reference.MetaReferenceEvaluationImplConfig",
+            api_dependencies=[
+                Api.inference,
+                Api.safety,
+                Api.vector_io,
+                Api.vector_dbs,
+                Api.tool_runtime,
+                Api.tool_groups,
+            ],
+        ),
+    ]
diff --git a/pyproject.toml b/pyproject.toml
index da29e0b9b..807da4337 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -215,6 +215,7 @@ exclude = [
     "^llama_stack/providers/inline/agents/meta_reference/agent_instance\\.py$",
     "^llama_stack/providers/inline/agents/meta_reference/agents\\.py$",
     "^llama_stack/providers/inline/agents/meta_reference/safety\\.py$",
+    "^llama_stack/providers/inline/evaluation/meta_reference/evaluation\\.py$",
     "^llama_stack/providers/inline/datasetio/localfs/",
     "^llama_stack/providers/inline/eval/meta_reference/eval\\.py$",
     "^llama_stack/providers/inline/inference/meta_reference/config\\.py$",