feat: allow building distro with external providers (#1967)

# What does this PR do?

We can now build a distribution that includes external providers.
Closes: https://github.com/meta-llama/llama-stack/issues/1948

## Test Plan

Build a distro with an external provider following the doc instructions.

[//]: # (## Documentation)

Added.

Rendered:


![Screenshot 2025-04-18 at 11 26
39](https://github.com/user-attachments/assets/afcf3d50-8d30-48c3-8d24-06a4b3662881)

Signed-off-by: Sébastien Han <seb@redhat.com>

.github/workflows/test-external-providers.yml

@@ -9,6 +9,11 @@ on:
 jobs:
   test-external-providers:
     runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        image-type: [venv]
+        # We don't do container yet, it's tricky to install a package from the host into the
+        # container and point 'uv pip install' to the correct path...
     steps:
       - name: Checkout repository
         uses: actions/checkout@v4
@@ -35,17 +40,25 @@ jobs:
           uv sync --extra dev --extra test
           uv pip install -e .
 
-      - name: Install Ollama custom provider
+      - name: Apply image type to config file
+        run: |
+          yq -i '.image_type = "${{ matrix.image-type }}"' tests/external-provider/llama-stack-provider-ollama/custom-distro.yaml
+          cat tests/external-provider/llama-stack-provider-ollama/custom-distro.yaml
+
+      - name: Setup directory for Ollama custom provider
         run: |
           mkdir -p tests/external-provider/llama-stack-provider-ollama/src/
           cp -a llama_stack/providers/remote/inference/ollama/ tests/external-provider/llama-stack-provider-ollama/src/llama_stack_provider_ollama
-          uv pip install tests/external-provider/llama-stack-provider-ollama
 
       - name: Create provider configuration
         run: |
           mkdir -p /tmp/providers.d/remote/inference
           cp tests/external-provider/llama-stack-provider-ollama/custom_ollama.yaml /tmp/providers.d/remote/inference/custom_ollama.yaml
 
+      - name: Build distro from config file
+        run: |
+          USE_COPY_NOT_MOUNT=true LLAMA_STACK_DIR=. uv run llama stack build --config tests/external-provider/llama-stack-provider-ollama/custom-distro.yaml
+
       - name: Wait for Ollama to start
         run: |
           echo "Waiting for Ollama..."
@@ -62,11 +75,13 @@ jobs:
           exit 1
 
       - name: Start Llama Stack server in background
+        if: ${{ matrix.image-type }} == 'venv'
         env:
           INFERENCE_MODEL: "meta-llama/Llama-3.2-3B-Instruct"
         run: |
-          source .venv/bin/activate
+          source ci-test/bin/activate
-          nohup uv run llama stack run tests/external-provider/llama-stack-provider-ollama/run.yaml --image-type venv > server.log 2>&1 &
+          uv run pip list
+          nohup uv run --active llama stack run tests/external-provider/llama-stack-provider-ollama/run.yaml --image-type ${{ matrix.image-type }} > server.log 2>&1 &
 
       - name: Wait for Llama Stack server to be ready
         run: |

docs/source/distributions/building_distro.md

@@ -176,7 +176,11 @@ distribution_spec:
     safety: inline::llama-guard
     agents: inline::meta-reference
     telemetry: inline::meta-reference
+image_name: ollama
 image_type: conda
+
+# If some providers are external, you can specify the path to the implementation
+external_providers_dir: /etc/llama-stack/providers.d
 ```
 
 ```
@@ -184,6 +188,57 @@ llama stack build --config llama_stack/templates/ollama/build.yaml
 ```
 :::
 
+:::{tab-item} Building with External Providers
+
+Llama Stack supports external providers that live outside of the main codebase. This allows you to create and maintain your own providers independently or use community-provided providers.
+
+To build a distribution with external providers, you need to:
+
+1. Configure the `external_providers_dir` in your build configuration file:
+
+```yaml
+# Example my-external-stack.yaml with external providers
+version: '2'
+distribution_spec:
+  description: Custom distro for CI tests
+  providers:
+    inference:
+    - remote::custom_ollama
+# Add more providers as needed
+image_type: container
+image_name: ci-test
+# Path to external provider implementations
+external_providers_dir: /etc/llama-stack/providers.d
+```
+
+Here's an example for a custom Ollama provider:
+
+```yaml
+adapter:
+  adapter_type: custom_ollama
+  pip_packages:
+  - ollama
+  - aiohttp
+  - llama-stack-provider-ollama # This is the provider package
+  config_class: llama_stack_ollama_provider.config.OllamaImplConfig
+  module: llama_stack_ollama_provider
+api_dependencies: []
+optional_api_dependencies: []
+```
+
+The `pip_packages` section lists the Python packages required by the provider, as well as the
+provider package itself. The package must be available on PyPI or can be provided from a local
+directory or a git repository (git must be installed on the build environment).
+
+2. Build your distribution using the config file:
+
+```
+llama stack build --config my-external-stack.yaml
+```
+
+For more information on external providers, including directory structure, provider types, and implementation requirements, see the [External Providers documentation](../providers/external.md).
+:::
+
 :::{tab-item} Building Container
 
 ```{admonition} Podman Alternative

llama_stack/cli/stack/_build.py

@@ -210,16 +210,9 @@ def run_stack_build_command(args: argparse.Namespace) -> None:
                 )
                 sys.exit(1)
 
-        if build_config.image_type == LlamaStackImageType.CONTAINER.value and not args.image_name:
-            cprint(
-                "Please specify --image-name when building a container from a config file",
-                color="red",
-            )
-            sys.exit(1)
-
     if args.print_deps_only:
         print(f"# Dependencies for {args.template or args.config or image_name}")
-        normal_deps, special_deps = get_provider_dependencies(build_config.distribution_spec.providers)
+        normal_deps, special_deps = get_provider_dependencies(build_config)
         normal_deps += SERVER_DEPENDENCIES
         print(f"uv pip install {' '.join(normal_deps)}")
         for special_dep in special_deps:
@@ -274,9 +267,10 @@ def _generate_run_config(
         image_name=image_name,
         apis=apis,
         providers={},
+        external_providers_dir=build_config.external_providers_dir if build_config.external_providers_dir else None,
     )
     # build providers dict
-    provider_registry = get_provider_registry()
+    provider_registry = get_provider_registry(build_config)
     for api in apis:
         run_config.providers[api] = []
         provider_types = build_config.distribution_spec.providers[api]
@@ -290,8 +284,22 @@ def _generate_run_config(
             if p.deprecation_error:
                 raise InvalidProviderError(p.deprecation_error)
 
-            config_type = instantiate_class_type(provider_registry[Api(api)][provider_type].config_class)
+            try:
-            if hasattr(config_type, "sample_run_config"):
+                config_type = instantiate_class_type(provider_registry[Api(api)][provider_type].config_class)
+            except ModuleNotFoundError:
+                # HACK ALERT:
+                # This code executes after building is done, the import cannot work since the
+                # package is either available in the venv or container - not available on the host.
+                # TODO: use a "is_external" flag in ProviderSpec to check if the provider is
+                # external
+                cprint(
+                    f"Failed to import provider {provider_type} for API {api} - assuming it's external, skipping",
+                    color="yellow",
+                )
+                # Set config_type to None to avoid UnboundLocalError
+                config_type = None
+
+            if config_type is not None and hasattr(config_type, "sample_run_config"):
                 config = config_type.sample_run_config(__distro_dir__=f"~/.llama/distributions/{image_name}")
             else:
                 config = {}
@@ -323,6 +331,7 @@ def _run_stack_build_command_from_build_config(
     template_name: Optional[str] = None,
     config_path: Optional[str] = None,
 ) -> str:
+    image_name = image_name or build_config.image_name
     if build_config.image_type == LlamaStackImageType.CONTAINER.value:
         if template_name:
             image_name = f"distribution-{template_name}"

llama_stack/distribution/build.py

@@ -7,16 +7,16 @@
 import importlib.resources
 import logging
 from pathlib import Path
-from typing import Dict, List
 
 from pydantic import BaseModel
 from termcolor import cprint
 
-from llama_stack.distribution.datatypes import BuildConfig, Provider
+from llama_stack.distribution.datatypes import BuildConfig
 from llama_stack.distribution.distribution import get_provider_registry
 from llama_stack.distribution.utils.exec import run_command
 from llama_stack.distribution.utils.image_types import LlamaStackImageType
 from llama_stack.providers.datatypes import Api
+from llama_stack.templates.template import DistributionTemplate
 
 log = logging.getLogger(__name__)
 
@@ -37,19 +37,24 @@ class ApiInput(BaseModel):
 
 
 def get_provider_dependencies(
-    config_providers: Dict[str, List[Provider]],
+    config: BuildConfig | DistributionTemplate,
 ) -> tuple[list[str], list[str]]:
     """Get normal and special dependencies from provider configuration."""
-    all_providers = get_provider_registry()
+    # Extract providers based on config type
+    if isinstance(config, DistributionTemplate):
+        providers = config.providers
+    elif isinstance(config, BuildConfig):
+        providers = config.distribution_spec.providers
     deps = []
+    registry = get_provider_registry(config)
 
-    for api_str, provider_or_providers in config_providers.items():
+    for api_str, provider_or_providers in providers.items():
-        providers_for_api = all_providers[Api(api_str)]
+        providers_for_api = registry[Api(api_str)]
 
         providers = provider_or_providers if isinstance(provider_or_providers, list) else [provider_or_providers]
 
         for provider in providers:
-            # Providers from BuildConfig and RunConfig are subtly different – not great
+            # Providers from BuildConfig and RunConfig are subtly different – not great
             provider_type = provider if isinstance(provider, str) else provider.provider_type
 
             if provider_type not in providers_for_api:
@@ -71,8 +76,8 @@ def get_provider_dependencies(
     return list(set(normal_deps)), list(set(special_deps))
 
 
-def print_pip_install_help(providers: Dict[str, List[Provider]]):
+def print_pip_install_help(config: BuildConfig):
-    normal_deps, special_deps = get_provider_dependencies(providers)
+    normal_deps, special_deps = get_provider_dependencies(config)
 
     cprint(
         f"Please install needed dependencies using the following commands:\n\nuv pip install {' '.join(normal_deps)}",
@@ -91,7 +96,7 @@ def build_image(
 ):
     container_base = build_config.distribution_spec.container_image or "python:3.10-slim"
 
-    normal_deps, special_deps = get_provider_dependencies(build_config.distribution_spec.providers)
+    normal_deps, special_deps = get_provider_dependencies(build_config)
     normal_deps += SERVER_DEPENDENCIES
 
     if build_config.image_type == LlamaStackImageType.CONTAINER.value:

llama_stack/distribution/build_container.sh

@@ -90,7 +90,7 @@ WORKDIR /app
 
 RUN apt-get update && apt-get install -y \
        iputils-ping net-tools iproute2 dnsutils telnet \
-       curl wget telnet \
+       curl wget telnet git\
        procps psmisc lsof \
        traceroute \
        bubblewrap \

llama_stack/distribution/datatypes.py

@@ -326,3 +326,12 @@ class BuildConfig(BaseModel):
         default="conda",
         description="Type of package to build (conda | container | venv)",
     )
+    image_name: Optional[str] = Field(
+        default=None,
+        description="Name of the distribution to build",
+    )
+    external_providers_dir: Optional[str] = Field(
+        default=None,
+        description="Path to directory containing external provider implementations. The providers packages will be resolved from this directory. "
+        "pip_packages MUST contain the provider package name.",
+    )

llama_stack/distribution/distribution.py

@@ -12,7 +12,6 @@ from typing import Any, Dict, List
 import yaml
 from pydantic import BaseModel
 
-from llama_stack.distribution.datatypes import StackRunConfig
 from llama_stack.log import get_logger
 from llama_stack.providers.datatypes import (
     AdapterSpec,
@@ -97,7 +96,9 @@ def _load_inline_provider_spec(spec_data: Dict[str, Any], api: Api, provider_nam
     return spec
 
 
-def get_provider_registry(config: StackRunConfig | None = None) -> Dict[Api, Dict[str, ProviderSpec]]:
+def get_provider_registry(
+    config=None,
+) -> Dict[Api, Dict[str, ProviderSpec]]:
     """Get the provider registry, optionally including external providers.
 
     This function loads both built-in providers and external providers from YAML files.
@@ -122,7 +123,7 @@ def get_provider_registry(config: StackRunConfig | None = None) -> Dict[Api, Dic
           llama-guard.yaml
 
     Args:
-        config: Optional StackRunConfig containing the external providers directory path
+        config: Optional object containing the external providers directory path
 
     Returns:
         A dictionary mapping APIs to their available providers
@@ -142,7 +143,8 @@ def get_provider_registry(config: StackRunConfig | None = None) -> Dict[Api, Dic
         except ImportError as e:
             logger.warning(f"Failed to import module {name}: {e}")
 
-    if config and config.external_providers_dir:
+    # Check if config has the external_providers_dir attribute
+    if config and hasattr(config, "external_providers_dir") and config.external_providers_dir:
         external_providers_dir = os.path.abspath(config.external_providers_dir)
         if not os.path.exists(external_providers_dir):
             raise FileNotFoundError(f"External providers directory not found: {external_providers_dir}")

scripts/distro_codegen.py

@@ -98,7 +98,7 @@ def collect_template_dependencies(template_dir: Path) -> tuple[str | None, list[
 
         if template_func := getattr(module, "get_distribution_template", None):
             template = template_func()
-            normal_deps, special_deps = get_provider_dependencies(template.providers)
+            normal_deps, special_deps = get_provider_dependencies(template)
             # Combine all dependencies in order: normal deps, special deps, server deps
             all_deps = sorted(set(normal_deps + SERVER_DEPENDENCIES)) + sorted(set(special_deps))
 

tests/external-provider/llama-stack-provider-ollama/custom-distro.yaml

@@ -0,0 +1,9 @@
+version: '2'
+distribution_spec:
+  description: Custom distro for CI tests
+  providers:
+    inference:
+    - remote::custom_ollama
+image_type: container
+image_name: ci-test
+external_providers_dir: /tmp/providers.d

tests/external-provider/llama-stack-provider-ollama/custom_ollama.yaml

@@ -1,6 +1,6 @@
 adapter:
   adapter_type: custom_ollama
-  pip_packages: ["ollama", "aiohttp"]
+  pip_packages: ["ollama", "aiohttp", "tests/external-provider/llama-stack-provider-ollama"]
   config_class: llama_stack_provider_ollama.config.OllamaImplConfig
   module: llama_stack_provider_ollama
 api_dependencies: []

tests/external-provider/llama-stack-provider-ollama/run.yaml

@@ -1,14 +1,10 @@
 version: '2'
 image_name: ollama
 apis:
-- agents
-- datasetio
-- eval
 - inference
-- safety
-- scoring
 - telemetry
 - tool_runtime
+- datasetio
 - vector_io
 providers:
   inference:
@@ -24,19 +20,6 @@ providers:
         type: sqlite
         namespace: null
         db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/ollama}/faiss_store.db
-  safety:
-  - provider_id: llama-guard
-    provider_type: inline::llama-guard
-    config:
-      excluded_categories: []
-  agents:
-  - provider_id: meta-reference
-    provider_type: inline::meta-reference
-    config:
-      persistence_store:
-        type: sqlite
-        namespace: null
-        db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/ollama}/agents_store.db
   telemetry:
   - provider_id: meta-reference
     provider_type: inline::meta-reference
@@ -44,14 +27,6 @@ providers:
       service_name: ${env.OTEL_SERVICE_NAME:llama-stack}
       sinks: ${env.TELEMETRY_SINKS:console,sqlite}
       sqlite_db_path: ${env.SQLITE_DB_PATH:~/.llama/distributions/ollama/trace_store.db}
-  eval:
-  - provider_id: meta-reference
-    provider_type: inline::meta-reference
-    config:
-      kvstore:
-        type: sqlite
-        namespace: null
-        db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/ollama}/meta_reference_eval.db
   datasetio:
   - provider_id: huggingface
     provider_type: remote::huggingface
@@ -67,17 +42,6 @@ providers:
         type: sqlite
         namespace: null
         db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/ollama}/localfs_datasetio.db
-  scoring:
-  - provider_id: basic
-    provider_type: inline::basic
-    config: {}
-  - provider_id: llm-as-judge
-    provider_type: inline::llm-as-judge
-    config: {}
-  - provider_id: braintrust
-    provider_type: inline::braintrust
-    config:
-      openai_api_key: ${env.OPENAI_API_KEY:}
   tool_runtime:
   - provider_id: brave-search
     provider_type: remote::brave-search