agents to use tools api (#673)

# What does this PR do?

PR #639 introduced the notion of Tools API and ability to invoke tools
through API just as any resource. This PR changes the Agents to start
using the Tools API to invoke tools. Major changes include:
1) Ability to specify tool groups with AgentConfig
2) Agent gets the corresponding tool definitions for the specified tools
and pass along to the model
3) Attachements are now named as Documents and their behavior is mostly
unchanged from user perspective
4) You can specify args that can be injected to a tool call through
Agent config. This is especially useful in case of memory tool, where
you want the tool to operate on a specific memory bank.
5) You can also register tool groups with args, which lets the agent
inject these as well into the tool call.
6) All tests have been migrated to use new tools API and fixtures
including client SDK tests
7) Telemetry just works with tools API because of our trace protocol
decorator


## Test Plan
```
pytest -s -v -k fireworks llama_stack/providers/tests/agents/test_agents.py  \
   --safety-shield=meta-llama/Llama-Guard-3-8B \
   --inference-model=meta-llama/Llama-3.1-8B-Instruct

pytest -s -v -k together  llama_stack/providers/tests/tools/test_tools.py \
   --safety-shield=meta-llama/Llama-Guard-3-8B \
   --inference-model=meta-llama/Llama-3.1-8B-Instruct

LLAMA_STACK_CONFIG="/Users/dineshyv/.llama/distributions/llamastack-together/together-run.yaml" pytest -v tests/client-sdk/agents/test_agents.py
```
run.yaml:
https://gist.github.com/dineshyv/0365845ad325e1c2cab755788ccc5994

Notebook:
https://colab.research.google.com/drive/1ck7hXQxRl6UvT-ijNRZ-gMZxH1G3cN2d?usp=sharing

llama_stack/templates/bedrock/bedrock.py

@@ -9,8 +9,7 @@ from pathlib import Path
 from llama_models.sku_list import all_registered_models
 
 from llama_stack.apis.models import ModelInput
-from llama_stack.distribution.datatypes import Provider
-
+from llama_stack.distribution.datatypes import Provider, ToolGroupInput
 from llama_stack.providers.inline.memory.faiss.config import FaissImplConfig
 from llama_stack.providers.remote.inference.bedrock.bedrock import MODEL_ALIASES
 from llama_stack.templates.template import DistributionTemplate, RunConfigSettings
@@ -26,6 +25,12 @@ def get_distribution_template() -> DistributionTemplate:
         "eval": ["inline::meta-reference"],
         "datasetio": ["remote::huggingface", "inline::localfs"],
         "scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::memory-runtime",
+        ],
     }
     name = "bedrock"
     memory_provider = Provider(
@@ -46,6 +51,20 @@ def get_distribution_template() -> DistributionTemplate:
         )
         for m in MODEL_ALIASES
     ]
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::memory",
+            provider_id="memory-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
 
     return DistributionTemplate(
         name=name,
@@ -61,6 +80,7 @@ def get_distribution_template() -> DistributionTemplate:
                     "memory": [memory_provider],
                 },
                 default_models=default_models,
+                default_tool_groups=default_tool_groups,
             ),
         },
         run_config_env_vars={

llama_stack/templates/bedrock/build.yaml

@@ -2,7 +2,6 @@ version: '2'
 name: bedrock
 distribution_spec:
   description: Use AWS Bedrock for running LLM inference and safety
-  docker_image: null
   providers:
     inference:
     - remote::bedrock
@@ -25,4 +24,9 @@ distribution_spec:
     - inline::basic
     - inline::llm-as-judge
     - inline::braintrust
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::memory-runtime
 image_type: conda

llama_stack/templates/bedrock/run.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: bedrock
-docker_image: null
 conda_env: bedrock
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: bedrock
@@ -65,8 +65,24 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/bedrock}/registry.db
 models:
@@ -90,3 +106,10 @@ memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/cerebras/build.yaml

@@ -2,7 +2,6 @@ version: '2'
 name: cerebras
 distribution_spec:
   description: Use Cerebras for running LLM inference
-  docker_image: null
   providers:
     inference:
     - remote::cerebras
@@ -14,4 +13,9 @@ distribution_spec:
     - inline::meta-reference
     telemetry:
     - inline::meta-reference
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::memory-runtime
 image_type: conda

llama_stack/templates/cerebras/cerebras.py

@@ -9,8 +9,12 @@ from pathlib import Path
 from llama_models.sku_list import all_registered_models
 
 from llama_stack.apis.models.models import ModelType
-
-from llama_stack.distribution.datatypes import ModelInput, Provider, ShieldInput
+from llama_stack.distribution.datatypes import (
+    ModelInput,
+    Provider,
+    ShieldInput,
+    ToolGroupInput,
+)
 from llama_stack.providers.inline.inference.sentence_transformers import (
     SentenceTransformersInferenceConfig,
 )
@@ -26,6 +30,12 @@ def get_distribution_template() -> DistributionTemplate:
         "memory": ["inline::meta-reference"],
         "agents": ["inline::meta-reference"],
         "telemetry": ["inline::meta-reference"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::memory-runtime",
+        ],
     }
 
     inference_provider = Provider(
@@ -58,6 +68,20 @@ def get_distribution_template() -> DistributionTemplate:
             "embedding_dimension": 384,
         },
     )
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::memory",
+            provider_id="memory-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
 
     return DistributionTemplate(
         name="cerebras",
@@ -74,6 +98,7 @@ def get_distribution_template() -> DistributionTemplate:
                 },
                 default_models=default_models + [embedding_model],
                 default_shields=[ShieldInput(shield_id="meta-llama/Llama-Guard-3-8B")],
+                default_tool_groups=default_tool_groups,
             ),
         },
         run_config_env_vars={

llama_stack/templates/cerebras/run.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: cerebras
-docker_image: null
 conda_env: cerebras
 apis:
 - agents
@@ -8,6 +7,7 @@ apis:
 - memory
 - safety
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: cerebras
@@ -45,8 +45,24 @@ providers:
       service_name: ${env.OTEL_SERVICE_NAME:llama-stack}
       sinks: ${env.TELEMETRY_SINKS:console,sqlite}
       sqlite_db_path: ${env.SQLITE_DB_PATH:~/.llama/distributions/cerebras/trace_store.db}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/cerebras}/registry.db
 models:
@@ -64,14 +80,17 @@ models:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields:
-- params: null
-  shield_id: meta-llama/Llama-Guard-3-8B
-  provider_id: null
-  provider_shield_id: null
+- shield_id: meta-llama/Llama-Guard-3-8B
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/fireworks/build.yaml

@@ -2,7 +2,6 @@ version: '2'
 name: fireworks
 distribution_spec:
   description: Use Fireworks.AI for running LLM inference
-  docker_image: null
   providers:
     inference:
     - remote::fireworks
@@ -25,4 +24,9 @@ distribution_spec:
     - inline::basic
     - inline::llm-as-judge
     - inline::braintrust
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::memory-runtime
 image_type: conda

llama_stack/templates/fireworks/fireworks.py

@@ -9,8 +9,12 @@ from pathlib import Path
 from llama_models.sku_list import all_registered_models
 
 from llama_stack.apis.models.models import ModelType
-
-from llama_stack.distribution.datatypes import ModelInput, Provider, ShieldInput
+from llama_stack.distribution.datatypes import (
+    ModelInput,
+    Provider,
+    ShieldInput,
+    ToolGroupInput,
+)
 from llama_stack.providers.inline.inference.sentence_transformers import (
     SentenceTransformersInferenceConfig,
 )
@@ -30,6 +34,12 @@ def get_distribution_template() -> DistributionTemplate:
         "eval": ["inline::meta-reference"],
         "datasetio": ["remote::huggingface", "inline::localfs"],
         "scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::memory-runtime",
+        ],
     }
 
     name = "fireworks"
@@ -69,6 +79,20 @@ def get_distribution_template() -> DistributionTemplate:
             "embedding_dimension": 384,
         },
     )
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::memory",
+            provider_id="memory-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
 
     return DistributionTemplate(
         name=name,
@@ -86,6 +110,7 @@ def get_distribution_template() -> DistributionTemplate:
                 },
                 default_models=default_models + [embedding_model],
                 default_shields=[ShieldInput(shield_id="meta-llama/Llama-Guard-3-8B")],
+                default_tool_groups=default_tool_groups,
             ),
         },
         run_config_env_vars={

llama_stack/templates/fireworks/run.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: fireworks
-docker_image: null
 conda_env: fireworks
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: fireworks
@@ -70,8 +70,24 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/fireworks}/registry.db
 models:
@@ -129,14 +145,17 @@ models:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields:
-- params: null
-  shield_id: meta-llama/Llama-Guard-3-8B
-  provider_id: null
-  provider_shield_id: null
+- shield_id: meta-llama/Llama-Guard-3-8B
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/hf-endpoint/build.yaml

@@ -2,7 +2,6 @@ version: '2'
 name: hf-endpoint
 distribution_spec:
   description: Use (an external) Hugging Face Inference Endpoint for running LLM inference
-  docker_image: null
   providers:
     inference:
     - remote::hf::endpoint
@@ -25,4 +24,9 @@ distribution_spec:
     - inline::basic
     - inline::llm-as-judge
     - inline::braintrust
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::memory-runtime
 image_type: conda

llama_stack/templates/hf-endpoint/hf_endpoint.py

@@ -5,7 +5,12 @@
 # the root directory of this source tree.
 
 from llama_stack.apis.models.models import ModelType
-from llama_stack.distribution.datatypes import ModelInput, Provider, ShieldInput
+from llama_stack.distribution.datatypes import (
+    ModelInput,
+    Provider,
+    ShieldInput,
+    ToolGroupInput,
+)
 from llama_stack.providers.inline.inference.sentence_transformers import (
     SentenceTransformersInferenceConfig,
 )
@@ -24,6 +29,12 @@ def get_distribution_template() -> DistributionTemplate:
         "eval": ["inline::meta-reference"],
         "datasetio": ["remote::huggingface", "inline::localfs"],
         "scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::memory-runtime",
+        ],
     }
     name = "hf-endpoint"
     inference_provider = Provider(
@@ -58,6 +69,20 @@ def get_distribution_template() -> DistributionTemplate:
             "embedding_dimension": 384,
         },
     )
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::memory",
+            provider_id="memory-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
 
     return DistributionTemplate(
         name=name,
@@ -74,6 +99,7 @@ def get_distribution_template() -> DistributionTemplate:
                     "memory": [memory_provider],
                 },
                 default_models=[inference_model, embedding_model],
+                default_tool_groups=default_tool_groups,
             ),
             "run-with-safety.yaml": RunConfigSettings(
                 provider_overrides={
@@ -96,6 +122,7 @@ def get_distribution_template() -> DistributionTemplate:
                     embedding_model,
                 ],
                 default_shields=[ShieldInput(shield_id="${env.SAFETY_MODEL}")],
+                default_tool_groups=default_tool_groups,
             ),
         },
         run_config_env_vars={

llama_stack/templates/hf-endpoint/run-with-safety.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: hf-endpoint
-docker_image: null
 conda_env: hf-endpoint
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: hf-endpoint
@@ -75,33 +75,50 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/hf-endpoint}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: hf-endpoint
-  provider_model_id: null
   model_type: llm
 - metadata: {}
   model_id: ${env.SAFETY_MODEL}
   provider_id: hf-endpoint-safety
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields:
-- params: null
-  shield_id: ${env.SAFETY_MODEL}
-  provider_id: null
-  provider_shield_id: null
+- shield_id: ${env.SAFETY_MODEL}
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/hf-endpoint/run.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: hf-endpoint
-docker_image: null
 conda_env: hf-endpoint
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: hf-endpoint
@@ -70,24 +70,45 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/hf-endpoint}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: hf-endpoint
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields: []
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/hf-serverless/build.yaml

@@ -2,7 +2,6 @@ version: '2'
 name: hf-serverless
 distribution_spec:
   description: Use (an external) Hugging Face Inference Endpoint for running LLM inference
-  docker_image: null
   providers:
     inference:
     - remote::hf::serverless
@@ -25,4 +24,9 @@ distribution_spec:
     - inline::basic
     - inline::llm-as-judge
     - inline::braintrust
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::memory-runtime
 image_type: conda

llama_stack/templates/hf-serverless/hf_serverless.py

@@ -5,7 +5,12 @@
 # the root directory of this source tree.
 
 from llama_stack.apis.models.models import ModelType
-from llama_stack.distribution.datatypes import ModelInput, Provider, ShieldInput
+from llama_stack.distribution.datatypes import (
+    ModelInput,
+    Provider,
+    ShieldInput,
+    ToolGroupInput,
+)
 from llama_stack.providers.inline.inference.sentence_transformers import (
     SentenceTransformersInferenceConfig,
 )
@@ -24,6 +29,12 @@ def get_distribution_template() -> DistributionTemplate:
         "eval": ["inline::meta-reference"],
         "datasetio": ["remote::huggingface", "inline::localfs"],
         "scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::memory-runtime",
+        ],
     }
 
     name = "hf-serverless"
@@ -59,6 +70,20 @@ def get_distribution_template() -> DistributionTemplate:
             "embedding_dimension": 384,
         },
     )
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::memory",
+            provider_id="memory-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
 
     return DistributionTemplate(
         name=name,
@@ -97,6 +122,7 @@ def get_distribution_template() -> DistributionTemplate:
                     embedding_model,
                 ],
                 default_shields=[ShieldInput(shield_id="${env.SAFETY_MODEL}")],
+                default_tool_groups=default_tool_groups,
             ),
         },
         run_config_env_vars={

llama_stack/templates/hf-serverless/run-with-safety.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: hf-serverless
-docker_image: null
 conda_env: hf-serverless
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: hf-serverless
@@ -75,33 +75,50 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/hf-serverless}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: hf-serverless
-  provider_model_id: null
   model_type: llm
 - metadata: {}
   model_id: ${env.SAFETY_MODEL}
   provider_id: hf-serverless-safety
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields:
-- params: null
-  shield_id: ${env.SAFETY_MODEL}
-  provider_id: null
-  provider_shield_id: null
+- shield_id: ${env.SAFETY_MODEL}
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/hf-serverless/run.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: hf-serverless
-docker_image: null
 conda_env: hf-serverless
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: hf-serverless
@@ -70,24 +70,39 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/hf-serverless}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: hf-serverless
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields: []
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups: []

llama_stack/templates/meta-reference-gpu/build.yaml

@@ -2,7 +2,6 @@ version: '2'
 name: meta-reference-gpu
 distribution_spec:
   description: Use Meta Reference for running LLM inference
-  docker_image: null
   providers:
     inference:
     - inline::meta-reference
@@ -25,4 +24,9 @@ distribution_spec:
     - inline::basic
     - inline::llm-as-judge
     - inline::braintrust
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::memory-runtime
 image_type: conda

llama_stack/templates/meta-reference-gpu/meta_reference.py

@@ -7,8 +7,12 @@
 from pathlib import Path
 
 from llama_stack.apis.models.models import ModelType
-
-from llama_stack.distribution.datatypes import ModelInput, Provider, ShieldInput
+from llama_stack.distribution.datatypes import (
+    ModelInput,
+    Provider,
+    ShieldInput,
+    ToolGroupInput,
+)
 from llama_stack.providers.inline.inference.meta_reference import (
     MetaReferenceInferenceConfig,
 )
@@ -29,6 +33,12 @@ def get_distribution_template() -> DistributionTemplate:
         "eval": ["inline::meta-reference"],
         "datasetio": ["remote::huggingface", "inline::localfs"],
         "scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::memory-runtime",
+        ],
     }
     name = "meta-reference-gpu"
     inference_provider = Provider(
@@ -66,6 +76,20 @@ def get_distribution_template() -> DistributionTemplate:
         model_id="${env.SAFETY_MODEL}",
         provider_id="meta-reference-safety",
     )
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::memory",
+            provider_id="memory-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
 
     return DistributionTemplate(
         name=name,
@@ -104,6 +128,7 @@ def get_distribution_template() -> DistributionTemplate:
                     embedding_model,
                 ],
                 default_shields=[ShieldInput(shield_id="${env.SAFETY_MODEL}")],
+                default_tool_groups=default_tool_groups,
             ),
         },
         run_config_env_vars={

llama_stack/templates/meta-reference-gpu/run-with-safety.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: meta-reference-gpu
-docker_image: null
 conda_env: meta-reference-gpu
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: meta-reference-inference
@@ -77,33 +77,50 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/meta-reference-gpu}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: meta-reference-inference
-  provider_model_id: null
   model_type: llm
 - metadata: {}
   model_id: ${env.SAFETY_MODEL}
   provider_id: meta-reference-safety
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields:
-- params: null
-  shield_id: ${env.SAFETY_MODEL}
-  provider_id: null
-  provider_shield_id: null
+- shield_id: ${env.SAFETY_MODEL}
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/meta-reference-gpu/run.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: meta-reference-gpu
-docker_image: null
 conda_env: meta-reference-gpu
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: meta-reference-inference
@@ -71,24 +71,39 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/meta-reference-gpu}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: meta-reference-inference
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields: []
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups: []

llama_stack/templates/meta-reference-quantized-gpu/build.yaml

@@ -2,7 +2,6 @@ version: '2'
 name: meta-reference-quantized-gpu
 distribution_spec:
   description: Use Meta Reference with fp8, int4 quantization for running LLM inference
-  docker_image: null
   providers:
     inference:
     - inline::meta-reference-quantized
@@ -25,4 +24,9 @@ distribution_spec:
     - inline::basic
     - inline::llm-as-judge
     - inline::braintrust
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::memory-runtime
 image_type: conda

llama_stack/templates/meta-reference-quantized-gpu/meta_reference.py

@@ -7,8 +7,7 @@
 from pathlib import Path
 
 from llama_stack.apis.models.models import ModelType
-
-from llama_stack.distribution.datatypes import ModelInput, Provider
+from llama_stack.distribution.datatypes import ModelInput, Provider, ToolGroupInput
 from llama_stack.providers.inline.inference.meta_reference import (
     MetaReferenceQuantizedInferenceConfig,
 )
@@ -29,7 +28,27 @@ def get_distribution_template() -> DistributionTemplate:
         "eval": ["inline::meta-reference"],
         "datasetio": ["remote::huggingface", "inline::localfs"],
         "scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::memory-runtime",
+        ],
     }
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::memory",
+            provider_id="memory-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
     name = "meta-reference-quantized-gpu"
     inference_provider = Provider(
         provider_id="meta-reference-inference",
@@ -76,6 +95,7 @@ def get_distribution_template() -> DistributionTemplate:
                     "memory": [memory_provider],
                 },
                 default_models=[inference_model, embedding_model],
+                default_tool_groups=default_tool_groups,
             ),
         },
         run_config_env_vars={

llama_stack/templates/meta-reference-quantized-gpu/run.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: meta-reference-quantized-gpu
-docker_image: null
 conda_env: meta-reference-quantized-gpu
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: meta-reference-inference
@@ -73,24 +73,45 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/meta-reference-quantized-gpu}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: meta-reference-inference
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields: []
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/ollama/build.yaml

@@ -2,7 +2,6 @@ version: '2'
 name: ollama
 distribution_spec:
   description: Use (an external) Ollama server for running LLM inference
-  docker_image: null
   providers:
     inference:
     - remote::ollama
@@ -25,4 +24,9 @@ distribution_spec:
     - inline::basic
     - inline::llm-as-judge
     - inline::braintrust
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::memory-runtime
 image_type: conda

llama_stack/templates/ollama/ollama.py

@@ -7,8 +7,12 @@
 from pathlib import Path
 
 from llama_stack.apis.models.models import ModelType
-
-from llama_stack.distribution.datatypes import ModelInput, Provider, ShieldInput
+from llama_stack.distribution.datatypes import (
+    ModelInput,
+    Provider,
+    ShieldInput,
+    ToolGroupInput,
+)
 from llama_stack.providers.inline.inference.sentence_transformers import (
     SentenceTransformersInferenceConfig,
 )
@@ -27,6 +31,12 @@ def get_distribution_template() -> DistributionTemplate:
         "eval": ["inline::meta-reference"],
         "datasetio": ["remote::huggingface", "inline::localfs"],
         "scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::memory-runtime",
+        ],
     }
     name = "ollama"
     inference_provider = Provider(
@@ -61,6 +71,20 @@ def get_distribution_template() -> DistributionTemplate:
             "embedding_dimension": 384,
         },
     )
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::memory",
+            provider_id="memory-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
 
     return DistributionTemplate(
         name=name,
@@ -92,6 +116,7 @@ def get_distribution_template() -> DistributionTemplate:
                     embedding_model,
                 ],
                 default_shields=[ShieldInput(shield_id="${env.SAFETY_MODEL}")],
+                default_tool_groups=default_tool_groups,
             ),
         },
         run_config_env_vars={

llama_stack/templates/ollama/run-with-safety.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: ollama
-docker_image: null
 conda_env: ollama
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: ollama
@@ -69,33 +69,50 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/ollama}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: ollama
-  provider_model_id: null
   model_type: llm
 - metadata: {}
   model_id: ${env.SAFETY_MODEL}
   provider_id: ollama
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields:
-- params: null
-  shield_id: ${env.SAFETY_MODEL}
-  provider_id: null
-  provider_shield_id: null
+- shield_id: ${env.SAFETY_MODEL}
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/ollama/run.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: ollama
-docker_image: null
 conda_env: ollama
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: ollama
@@ -69,24 +69,39 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/ollama}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: ollama
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields: []
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups: []

llama_stack/templates/remote-vllm/build.yaml

@@ -2,7 +2,6 @@ version: '2'
 name: remote-vllm
 distribution_spec:
   description: Use (an external) vLLM server for running LLM inference
-  docker_image: null
   providers:
     inference:
     - remote::vllm
@@ -16,4 +15,9 @@ distribution_spec:
     - inline::meta-reference
     telemetry:
     - inline::meta-reference
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::memory-runtime
 image_type: conda

llama_stack/templates/remote-vllm/run-with-safety.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: remote-vllm
-docker_image: null
 conda_env: remote-vllm
 apis:
 - agents
@@ -8,6 +7,7 @@ apis:
 - memory
 - safety
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: vllm-inference
@@ -52,33 +52,50 @@ providers:
       service_name: ${env.OTEL_SERVICE_NAME:llama-stack}
       sinks: ${env.TELEMETRY_SINKS:console,sqlite}
       sqlite_db_path: ${env.SQLITE_DB_PATH:~/.llama/distributions/remote-vllm/trace_store.db}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/remote-vllm}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: vllm-inference
-  provider_model_id: null
   model_type: llm
 - metadata: {}
   model_id: ${env.SAFETY_MODEL}
   provider_id: vllm-safety
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields:
-- params: null
-  shield_id: ${env.SAFETY_MODEL}
-  provider_id: null
-  provider_shield_id: null
+- shield_id: ${env.SAFETY_MODEL}
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/remote-vllm/run.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: remote-vllm
-docker_image: null
 conda_env: remote-vllm
 apis:
 - agents
@@ -8,6 +7,7 @@ apis:
 - memory
 - safety
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: vllm-inference
@@ -46,24 +46,39 @@ providers:
       service_name: ${env.OTEL_SERVICE_NAME:llama-stack}
       sinks: ${env.TELEMETRY_SINKS:console,sqlite}
       sqlite_db_path: ${env.SQLITE_DB_PATH:~/.llama/distributions/remote-vllm/trace_store.db}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/remote-vllm}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: vllm-inference
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields: []
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups: []

llama_stack/templates/remote-vllm/vllm.py

@@ -7,8 +7,12 @@
 from pathlib import Path
 
 from llama_stack.apis.models.models import ModelType
-
-from llama_stack.distribution.datatypes import ModelInput, Provider, ShieldInput
+from llama_stack.distribution.datatypes import (
+    ModelInput,
+    Provider,
+    ShieldInput,
+    ToolGroupInput,
+)
 from llama_stack.providers.inline.inference.sentence_transformers import (
     SentenceTransformersInferenceConfig,
 )
@@ -24,6 +28,12 @@ def get_distribution_template() -> DistributionTemplate:
         "safety": ["inline::llama-guard"],
         "agents": ["inline::meta-reference"],
         "telemetry": ["inline::meta-reference"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::memory-runtime",
+        ],
     }
     name = "remote-vllm"
     inference_provider = Provider(
@@ -60,6 +70,20 @@ def get_distribution_template() -> DistributionTemplate:
             "embedding_dimension": 384,
         },
     )
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::memory",
+            provider_id="memory-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
 
     return DistributionTemplate(
         name=name,
@@ -97,6 +121,7 @@ def get_distribution_template() -> DistributionTemplate:
                     embedding_model,
                 ],
                 default_shields=[ShieldInput(shield_id="${env.SAFETY_MODEL}")],
+                default_tool_groups=default_tool_groups,
             ),
         },
         run_config_env_vars={

llama_stack/templates/template.py

@@ -20,6 +20,7 @@ from llama_stack.distribution.datatypes import (
     Provider,
     ShieldInput,
     StackRunConfig,
+    ToolGroupInput,
 )
 from llama_stack.distribution.distribution import get_provider_registry
 from llama_stack.distribution.utils.dynamic import instantiate_class_type
@@ -30,6 +31,7 @@ class RunConfigSettings(BaseModel):
     provider_overrides: Dict[str, List[Provider]] = Field(default_factory=dict)
     default_models: Optional[List[ModelInput]] = None
     default_shields: Optional[List[ShieldInput]] = None
+    default_tool_groups: Optional[List[ToolGroupInput]] = None
 
     def run_config(
         self,
@@ -91,6 +93,7 @@ class RunConfigSettings(BaseModel):
             ),
             models=self.default_models or [],
             shields=self.default_shields or [],
+            tool_groups=self.default_tool_groups or [],
         )
 
 
@@ -159,14 +162,22 @@ class DistributionTemplate(BaseModel):
 
         build_config = self.build_config()
         with open(yaml_output_dir / "build.yaml", "w") as f:
-            yaml.safe_dump(build_config.model_dump(), f, sort_keys=False)
+            yaml.safe_dump(
+                build_config.model_dump(exclude_none=True),
+                f,
+                sort_keys=False,
+            )
 
         for yaml_pth, settings in self.run_configs.items():
             run_config = settings.run_config(
                 self.name, self.providers, self.docker_image
             )
             with open(yaml_output_dir / yaml_pth, "w") as f:
-                yaml.safe_dump(run_config.model_dump(), f, sort_keys=False)
+                yaml.safe_dump(
+                    run_config.model_dump(exclude_none=True),
+                    f,
+                    sort_keys=False,
+                )
 
         if self.template_path:
             docs = self.generate_markdown_docs()

llama_stack/templates/tgi/build.yaml

@@ -2,7 +2,6 @@ version: '2'
 name: tgi
 distribution_spec:
   description: Use (an external) TGI server for running LLM inference
-  docker_image: null
   providers:
     inference:
     - remote::tgi
@@ -25,4 +24,9 @@ distribution_spec:
     - inline::basic
     - inline::llm-as-judge
     - inline::braintrust
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::memory-runtime
 image_type: conda

llama_stack/templates/tgi/run-with-safety.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: tgi
-docker_image: null
 conda_env: tgi
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: tgi-inference
@@ -70,27 +70,45 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/tgi}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: tgi-inference
-  provider_model_id: null
   model_type: llm
 - metadata: {}
   model_id: ${env.SAFETY_MODEL}
   provider_id: tgi-safety
-  provider_model_id: null
   model_type: llm
 shields:
-- params: null
-  shield_id: ${env.SAFETY_MODEL}
-  provider_id: null
-  provider_shield_id: null
+- shield_id: ${env.SAFETY_MODEL}
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/tgi/run.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: tgi
-docker_image: null
 conda_env: tgi
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: tgi-inference
@@ -69,24 +69,39 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/tgi}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: tgi-inference
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields: []
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups: []

llama_stack/templates/tgi/tgi.py

@@ -7,8 +7,12 @@
 from pathlib import Path
 
 from llama_stack.apis.models.models import ModelType
-
-from llama_stack.distribution.datatypes import ModelInput, Provider, ShieldInput
+from llama_stack.distribution.datatypes import (
+    ModelInput,
+    Provider,
+    ShieldInput,
+    ToolGroupInput,
+)
 from llama_stack.providers.inline.inference.sentence_transformers import (
     SentenceTransformersInferenceConfig,
 )
@@ -27,6 +31,12 @@ def get_distribution_template() -> DistributionTemplate:
         "eval": ["inline::meta-reference"],
         "datasetio": ["remote::huggingface", "inline::localfs"],
         "scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::memory-runtime",
+        ],
     }
     name = "tgi"
     inference_provider = Provider(
@@ -63,6 +73,20 @@ def get_distribution_template() -> DistributionTemplate:
         model_id="${env.SAFETY_MODEL}",
         provider_id="tgi-safety",
     )
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::memory",
+            provider_id="memory-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
 
     return DistributionTemplate(
         name=name,
@@ -99,6 +123,7 @@ def get_distribution_template() -> DistributionTemplate:
                     safety_model,
                 ],
                 default_shields=[ShieldInput(shield_id="${env.SAFETY_MODEL}")],
+                default_tool_groups=default_tool_groups,
             ),
         },
         run_config_env_vars={

llama_stack/templates/together/build.yaml

@@ -2,7 +2,6 @@ version: '2'
 name: together
 distribution_spec:
   description: Use Together.AI for running LLM inference
-  docker_image: null
   providers:
     inference:
     - remote::together
@@ -25,4 +24,9 @@ distribution_spec:
     - inline::basic
     - inline::llm-as-judge
     - inline::braintrust
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::memory-runtime
 image_type: conda

llama_stack/templates/together/run.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: together
-docker_image: null
 conda_env: together
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: together
@@ -70,8 +70,24 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/together}/registry.db
 models:
@@ -124,14 +140,17 @@ models:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields:
-- params: null
-  shield_id: meta-llama/Llama-Guard-3-8B
-  provider_id: null
-  provider_shield_id: null
+- shield_id: meta-llama/Llama-Guard-3-8B
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/together/together.py

@@ -9,8 +9,12 @@ from pathlib import Path
 from llama_models.sku_list import all_registered_models
 
 from llama_stack.apis.models.models import ModelType
-
-from llama_stack.distribution.datatypes import ModelInput, Provider, ShieldInput
+from llama_stack.distribution.datatypes import (
+    ModelInput,
+    Provider,
+    ShieldInput,
+    ToolGroupInput,
+)
 from llama_stack.providers.inline.inference.sentence_transformers import (
     SentenceTransformersInferenceConfig,
 )
@@ -30,6 +34,12 @@ def get_distribution_template() -> DistributionTemplate:
         "eval": ["inline::meta-reference"],
         "datasetio": ["remote::huggingface", "inline::localfs"],
         "scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::memory-runtime",
+        ],
     }
     name = "together"
     inference_provider = Provider(
@@ -59,6 +69,20 @@ def get_distribution_template() -> DistributionTemplate:
         )
         for m in MODEL_ALIASES
     ]
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::memory",
+            provider_id="memory-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
     embedding_model = ModelInput(
         model_id="all-MiniLM-L6-v2",
         provider_id="sentence-transformers",
@@ -83,6 +107,7 @@ def get_distribution_template() -> DistributionTemplate:
                     "memory": [memory_provider],
                 },
                 default_models=default_models + [embedding_model],
+                default_tool_groups=default_tool_groups,
                 default_shields=[ShieldInput(shield_id="meta-llama/Llama-Guard-3-8B")],
             ),
         },

llama_stack/templates/vllm-gpu/build.yaml

@@ -2,7 +2,6 @@ version: '2'
 name: vllm-gpu
 distribution_spec:
   description: Use a built-in vLLM engine for running LLM inference
-  docker_image: null
   providers:
     inference:
     - inline::vllm
@@ -25,4 +24,9 @@ distribution_spec:
     - inline::basic
     - inline::llm-as-judge
     - inline::braintrust
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::memory-runtime
 image_type: conda

llama_stack/templates/vllm-gpu/run.yaml

@@ -1,6 +1,5 @@
 version: '2'
 image_name: vllm-gpu
-docker_image: null
 conda_env: vllm-gpu
 apis:
 - agents
@@ -11,6 +10,7 @@ apis:
 - safety
 - scoring
 - telemetry
+- tool_runtime
 providers:
   inference:
   - provider_id: vllm
@@ -73,24 +73,45 @@ providers:
     provider_type: inline::braintrust
     config:
       openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: memory-runtime
+    provider_type: inline::memory-runtime
+    config: {}
 metadata_store:
-  namespace: null
   type: sqlite
   db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/vllm-gpu}/registry.db
 models:
 - metadata: {}
   model_id: ${env.INFERENCE_MODEL}
   provider_id: vllm
-  provider_model_id: null
   model_type: llm
 - metadata:
     embedding_dimension: 384
   model_id: all-MiniLM-L6-v2
   provider_id: sentence-transformers
-  provider_model_id: null
   model_type: embedding
 shields: []
 memory_banks: []
 datasets: []
 scoring_fns: []
 eval_tasks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::memory
+  provider_id: memory-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter

llama_stack/templates/vllm-gpu/vllm.py

@@ -11,7 +11,11 @@ from llama_stack.providers.inline.inference.sentence_transformers import (
 )
 from llama_stack.providers.inline.inference.vllm import VLLMConfig
 from llama_stack.providers.inline.memory.faiss.config import FaissImplConfig
-from llama_stack.templates.template import DistributionTemplate, RunConfigSettings
+from llama_stack.templates.template import (
+    DistributionTemplate,
+    RunConfigSettings,
+    ToolGroupInput,
+)
 
 
 def get_distribution_template() -> DistributionTemplate:
@@ -24,7 +28,14 @@ def get_distribution_template() -> DistributionTemplate:
         "eval": ["inline::meta-reference"],
         "datasetio": ["remote::huggingface", "inline::localfs"],
         "scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::memory-runtime",
+        ],
     }
+
     name = "vllm-gpu"
     inference_provider = Provider(
         provider_id="vllm",
@@ -54,6 +65,20 @@ def get_distribution_template() -> DistributionTemplate:
             "embedding_dimension": 384,
         },
     )
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::memory",
+            provider_id="memory-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
 
     return DistributionTemplate(
         name=name,
@@ -70,6 +95,7 @@ def get_distribution_template() -> DistributionTemplate:
                     "memory": [memory_provider],
                 },
                 default_models=[inference_model, embedding_model],
+                default_tool_groups=default_tool_groups,
             ),
         },
         run_config_env_vars={