Lightspeed-Test Project

This project creates a Docker Compose setup with two main services:

• llama-stack: A LLaMA Stack instance connected to Ollama (port 8321)
• lightspeed-stack: A Lightspeed Stack instance for AI interactions (port 8080)

Project Structure

├── docker-compose.yaml      # Main docker-compose configuration
├── config.yaml             # Lightspeed-stack configuration
├── run-llama-stack.yaml    # LLaMA stack runtime configuration
└── test-functions.sh       # Testing utility functions

How to Start Services

Requirements:

• podman-compose v1.5.0+

OLLAMA (running locally)

export VLLM_URL="http://host.containers.internal:11434/v1"

More info: https://llama-stack.readthedocs.io/en/latest/providers/inference/remote_vllm.html

Vertex AI running in GCP

# create ~/.config/gcloud/application_default_credentials.json:
gcloud auth application-default login   

# Verify login:
gcloud auth list

# From https://console.cloud.google.com/home/dashboard
export VERTEXAI_PROJECT="myproject"

More info: https://llama-stack.readthedocs.io/en/latest/providers/inference/remote_vertexai.html

Using podman-compose

podman-compose up -d

Check running services

podman-compose ps

Output:

CONTAINER ID  IMAGE                                            COMMAND     CREATED         STATUS                   PORTS                   NAMES
7099fbbad9a1  docker.io/llamastack/distribution-ollama:latest              23 seconds ago  Up 23 seconds (healthy)  0.0.0.0:8321->8321/tcp  llama-stack
8204fde2b4f3  quay.io/lightspeed-core/lightspeed-stack:latest              23 seconds ago  Up 12 seconds            0.0.0.0:8080->8080/tcp  lightspeed-stack

Service Endpoints

Port 8080 (Lightspeed-Stack)

The Lightspeed-Stack service provides the following functionality:

• Service Info: GET /v1/info

  {
    "name": "Test LSCore",
    "version": "0.1.3"
  }

• Configuration: GET /v1/config - Returns complete service configuration

• Available Models: GET /v1/models - Lists available AI models

• Query Endpoint: POST /v1/query - Main AI interaction endpoint

Port 8321 (LLaMA-Stack)

The LLaMA-Stack service provides:

• Health Check: GET /v1/health

  {"status":"OK"}

• Available Models: GET /v1/models

  {
    "data": [
      {
        "identifier": "gemma3:27b-it-qat",
        "provider_resource_id": "gemma3:27b-it-qat",
        "provider_id": "ollama",
        "type": "model",
        "metadata": {},
        "model_type": "llm"
      }
    ]
  }

• Providers: GET /v1/providers - Lists available providers (ollama, model-context-protocol, meta-reference)

• Chat Completion: POST /v1/inference/chat-completion - AI inference endpoint

Test Functions

The test-functions.sh file contains utility functions for testing both services. To use them:

source test-functions.sh

LLaMA-Stack Functions

Function	Description
llama::list_models()	List available models (identifiers only)
llama::list_models_full()	List available models (full details)
llama::list_providers()	List provider IDs
llama::list_providers_full()	List providers (full details)
llama::list_tools()	List available tools (may return error if tools not configured)
llama::list_toolgroups()	List available tools groups
llama::chat_completion()	Test chat completion with sample query
llama::list_agents()	List created agents in llamastack

Lightspeed-Stack Functions

Function	Description
ls::info()	Get service information
ls::config()	Get complete service configuration
ls::models()	List available models
ls::test()	Test query endpoint with sample request
ls::stest()	Test query endpoint with sample request (streaming)
ls::conversation()	Retrieve conversation request

Testing LLaMA-Stack with Functions

# Source the functions
source test-functions.sh

# List providers
llama::list_providers
# Output: 
# meta-reference
# vllm-inference
# google-vertex
# model-context-protocol
#meta-reference

# List available models from providers
llama::list_models
# Output: vertex_ai/gemini-2.5-flash

# set model for next function calls
export LLAMA_MODEL=vertex_ai/gemini-2.5-flash

# Test chat completion (requires correct LLAMA_MODEL env variable)
llama::chat_completion

Testing Lightspeed-Stack with Functions

# Source the functions
source test-functions.sh

# Get service info
ls::info
# Output: {"name": "Test LSCore", "version": "0.1.3"}

# Get full configuration
ls::config

# List available models
ls::models

# Test query endpoint
ls::test

Configuration Details

LLaMA-Stack Configuration

• Model: gemma3:27b-it-qat
• Provider: Ollama (requires external Ollama instance)
• APIs: inference, safety, tool_runtime, telemetry
• Storage: SQLite databases in /tmp/

Lightspeed-Stack Configuration

• Default Model: gemma3:27b-it-qat
• Default Provider: ollama
• Auth: Disabled
• Data Collection: Feedback and transcripts enabled
• MCP Server: Configured for model-context-protocol

Notes

• The LLaMA-Stack connects to an external Ollama instance (configure via OLLAMA_HOST environment variable)
• Both services use the lightspeednet bridge network for communication
• The Lightspeed-Stack waits for LLaMA-Stack to be healthy before starting
• Tool runtime endpoint may return errors if MCP servers are not properly configured