Skip to content

README.md

Latest commit

 

History

History
107 lines (80 loc) · 2.94 KB

README.md

File metadata and controls

107 lines (80 loc) · 2.94 KB
Raw

Qiskit Studio Backend API

Getting Started

Agent Workflow API

The API is based on Maestro you can install it as follows (you need to have uv installed)

Clone this repository and switch to the api directory

git clone https://github.com/AI4quantum/qiskit-studio
cd qiskit-studio/api/

Create an .env file for each maestro agent by copying .env.template into each agent directory

cp .env.template chat-agent/.env
cp .env.template codegen-agent/.env

Start each agent in separate terminals

cd chat-agent/
uv run maestro serve agents.yaml workflow.yaml
cd codegen-agent/
uv run maestro serve agents.yaml workflow.yaml --port 8001
cd coderun-agent/
uv run python agent.py --port 8002

Setup the vector database using maestro-knowledge

git clone https://github.com/AI4quantum/maestro-knowledge.git
cd maestro-knowledge
CUSTOM_EMBEDDING_URL=http://127.0.0.1:11434/v1 CUSTOM_EMBEDDING_MODEL=nomic-embed-text CUSTOM_EMBEDDING_VECTORSIZE=768 CUSTOM_EMBEDDING_API_KEY=dummy uv run ./start.sh

Populate the vector database for use by the chat agent

cd chat-agent/
uv run python scripts/add-rag-docs-remote-embed.py

LLM

Be sure that you have installed Ollama or similar local LLM provider and download granite3.3:8b

You can change the endpoint in .env or switch this model for other by editing agents.yaml before starting maestro.

Usage

After installing dependencies and starting the Maestro workflow servers, you can call the APIs at:

http://127.0.0.1:8000/chat # chat agent
http://127.0.0.1:8001/chat # code generation agent
http://127.0.0.1:8002/run  # coderun agent

Building and running in Docker

Each agent has a Dockerfile and can be built and run with the following commands:

cd chat-agent/
docker build -t chat-agent:latest .
docker run -p 8000:8000 --env-file .env chat-agent:latest
cd codegen-agent/
docker build -t codegen-agent:latest .
docker run -p 8001:8000 --env-file .env codegen-agent:latest

Note: You may need to update the urls in .env and chat-agent/agent.py to use docker compatible endpoints.

Running in Kubernetes

Follow the instructions in charts/qiskit-studio/README.md to run the entire qiskit studio stack in a kubernetes cluster using a helm chart

Creating new maestro workflows

Below is an example maestro agent with notes on different possible fields and values:

apiVersion: maestro/v1alpha1
kind: Agent
metadata:
  name: llm-agent
  labels:
    app: qiskit-studio
spec:
  model: granite3.3:8b # add a 'ollama/' prefix when using dspy
  framework: openai # or dspy or beeai
  mode: local
  url: "http://localhost:11434" # http://host.docker.internal:11434 # only used by dspy
  description: Generates text using LLMs
  instructions: <instructions to add to the system prompt>