Merge pull request #52 from workfloworchestrator/support-llm-features

Support for search/agent features in the core package

Author: pboers1988

.env.example

@@ -3,4 +3,8 @@ export PYTHONUNBUFFERED=1
 export OAUTH2_ACTIVE=0
 export LOG_LEVEL=DEBUG
 export WEBSOCKET_BROADCASTER_URL="redis://localhost:6379"
+
+export OPENAI_API_KEY=None
+export AGENT_ENABLED=False
+export SEARCH_ENABLED=False
 # export COMPOSE_PROFILES=lso # unhash the start of this line to also start up lso container

docker-compose.yml

@@ -71,6 +71,25 @@ services:
     volumes:
       - netbox-redis-cache-data:/data
 
+  embeddings:
+    container_name: embeddings
+    image: ghcr.io/huggingface/text-embeddings-inference:cpu-1.8
+    command: --model-id sentence-transformers/all-MiniLM-L6-v2
+    profiles:
+      - embeddings
+    env_file:
+      - ./docker/embeddings/embeddings.env
+      - path: ./docker/overrides/embeddings/embeddings.env
+        required: false
+    ports:
+      - "8081:80"
+    healthcheck:
+      test: ["CMD", "curl", "-f", "http://localhost:80/health"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+      start_period: 30s
+
   rover-compose:
     container_name: rover-compose
     build:
@@ -160,10 +179,13 @@ services:
     image: ${ORCH_BACKEND_TAG:-ghcr.io/workfloworchestrator/orchestrator-core:latest}
     env_file:
       - ./docker/orchestrator/orchestrator.env
+      - ./docker/embeddings/embeddings.env
       - path: ./docker/overrides/orchestrator/orchestrator.env
         required: false
     environment:
       LSO_ENABLED: ${COMPOSE_PROFILES:+True}
+      AGENT_ENABLED: ${AGENT_ENABLED:-False}
+      SEARCH_ENABLED: ${SEARCH_ENABLED:-False}
     ports:
       - "8080:8080"
       - "5678:5678" #Enable Python debugger
@@ -179,7 +201,7 @@ services:
       - ./graphql_federation.py:/home/orchestrator/graphql_federation.py
       - ./utils:/home/orchestrator/utils
       - ./services:/home/orchestrator/services
-      - ./requirements.txt:/home/orchestrator/requirements.txt
+      - ./pyproject.toml:/home/orchestrator/pyproject.toml
       - ./alembic.ini:/home/orchestrator/alembic.ini
       - ./translations:/home/orchestrator/translations
       - ./templates:/home/orchestrator/templates
@@ -191,6 +213,9 @@ services:
         condition: service_healthy
       redis:
         condition: service_healthy
+      embeddings:
+        condition: service_healthy
+        required: false
     healthcheck:
       start_period: 15s
       timeout: 5s

docker/embeddings/README.md

@@ -0,0 +1,88 @@
+# Embeddings Service
+
+This directory configures a local embedding server using Hugging Face Text Embeddings Inference (TEI) with the `sentence-transformers/all-MiniLM-L6-v2` model.
+
+## Overview
+
+The embeddings service provides an OpenAI-compatible API for generating text embeddings locally.
+
+- **Model**: sentence-transformers/all-MiniLM-L6-v2
+- **Embedding Dimension**: 384
+- **API Endpoint**: http://embeddings:80/v1 (internal), http://localhost:8081/v1 (external)
+
+## Prerequisites
+
+To use embeddings for search and agent features, add these to your `.env` file (see `.env.example`):
+
+```env
+AGENT_ENABLED=True
+SEARCH_ENABLED=True
+OPENAI_API_KEY=your-api-key-here  # Optional: only needed for agent features or when using OpenAI embeddings
+```
+
+## Local Embeddings (Default)
+
+This setup uses a local embedding service with no external API required. The default configuration in `embeddings.env` is:
+
+- `OPENAI_BASE_URL=http://embeddings:80/v1`
+- `EMBEDDING_DIMENSION=384`
+
+### Start the orchestrator
+
+Start the orchestrator with the local embeddings service:
+
+```bash
+docker compose --profile embeddings up orchestrator
+```
+
+## Alternative: Using OpenAI Embeddings
+
+If you prefer to use OpenAI's embedding service instead of running a local model:
+
+### Configuration
+
+Override the embedding settings by editing `docker/overrides/embeddings/embeddings.env`:
+
+```env
+OPENAI_BASE_URL=https://api.openai.com/v1
+EMBEDDING_DIMENSION=1536
+```
+
+### Start the orchestrator
+
+Start only the orchestrator (skips the local embeddings service):
+
+```bash
+docker compose up orchestrator
+```
+
+## Post-Setup Steps
+
+After starting the services and making sure you have data for the entity you want to index:
+
+### 1. Apply the schema change
+
+This will resize the vector dimension to match your embedding configuration (384 for local, 1536 for OpenAI) and delete existing records:
+
+```bash
+docker compose exec orchestrator /home/orchestrator/.venv/bin/python main.py embedding resize
+```
+
+⚠️ **Note**: This command will delete all existing embedding records.
+
+### 2. Re-index your data
+
+Example Index subscriptions:
+
+```bash
+docker compose exec orchestrator /home/orchestrator/.venv/bin/python main.py index subscriptions
+```
+
+## Advanced Configuration
+
+The following configurations use conservative defaults for local/unknown models:
+
+- `EMBEDDING_FALLBACK_MAX_TOKENS=512`: Maximum tokens per embedding request
+- `EMBEDDING_MAX_BATCH_SIZE=32`: Maximum batch size for embedding requests
+
+**Note**: These settings are only used as fallbacks for local or unknown models (like the example in this setup). For known providers and models, the system automatically retrieves the correct values via LiteLLM. The fallback values are already configured safely for local models, but can be adjusted if needed in `docker/overrides/embeddings/embeddings.env`.

docker/embeddings/embeddings.env

@@ -0,0 +1,11 @@
+# Default: Local embeddings service using sentence-transformers/all-MiniLM-L6-v2
+# Provides OpenAI-compatible embeddings API locally with no external dependencies
+
+OPENAI_BASE_URL=http://embeddings:80/v1
+EMBEDDING_DIMENSION=384
+EMBEDDING_MAX_BATCH_SIZE=32 # Not required when using OpenAI
+
+# Alternative: Uncomment below to use OpenAI embeddings instead
+# (Requires OPENAI_API_KEY in your .env file)
+# OPENAI_BASE_URL=https://api.openai.com/v1
+# EMBEDDING_DIMENSION=1536

docker/orchestrator/entrypoint.sh

@@ -4,20 +4,40 @@
 PATH=$PATH:~/.local/bin
 
 # Install extra requirements for orchestrator
-pip install -r requirements.txt
+pip install uv
+uv sync
+source .venv/bin/activate
 
 if [ -f ${CORE_OVERRIDE}/pyproject.toml ]; then
-    echo "⏭️ Use editable install of orchestrator-core with dev and test dependencies"
-    pip install -e $CORE_OVERRIDE[dev,test]
+    echo "⏭️ Use editable install of orchestrator-core"
+
+    extras=""
+    [ "${AGENT_ENABLED,,}" = "true" ] && extras+="agent,"
+    [ "${SEARCH_ENABLED,,}" = "true" ] && extras+="search,"
+
+    install_spec="$CORE_OVERRIDE"
+    if [ -n "$extras" ]; then
+        # Example: '/path/to/core[agent,search]'
+        install_spec="$CORE_OVERRIDE"'['"${extras%,}"']'
+    fi
+
+    echo "Installing with spec: '$install_spec'"
+    uv pip install -e "$install_spec"
 
     # Run any missing migrations on the database
     python main.py db upgrade heads
 
-    uvicorn --host 0.0.0.0 --port 8080 $UVICORN_ARGS main:app --reload --proxy-headers --reload-dir $CORE_OVERRIDE
+    uvicorn --host 0.0.0.0 --port 8080 $UVICORN_ARGS main:app --reload --proxy-headers \
+        --reload-dir $CORE_OVERRIDE \
+        --reload-dir products \
+        --reload-dir services \
+        --reload-dir translations \
+        --reload-dir utils \
+        --reload-dir workflows
 else
     # Run any missing migrations on the database
     python main.py db upgrade heads
 
-    echo "⏭️ Use pip installed orchestrator-core $(pip freeze | grep orchestrator-core)"
+    echo "⏭️ Use orchestrator-core as specified in pyproject.toml $(uv pip freeze | grep orchestrator-core)"
     uvicorn --host 0.0.0.0 --port 8080 $UVICORN_ARGS main:app --reload --proxy-headers
-fi
+fi