SciPhi AI example

This is an example based on SciPhi-AI/R2R-qna-rag-prebuilt.

How to execute it

Different modes:

Local with pgvector
Docker with pgvector (own infrastructure)
SciPhi

Required environment variables for all modes

OPENAI_API_KEY
ANTHROPIC_API_KEY: anthropic.com
CONFIDENTAI_API_KEY: For Confident AI's deepeval prompt evaluations/monitoring. An alternative for this is Parea AI.

$ export OPENAI_API_KEY=xxxxxx
$ export ANTHROPIC_API_KEY=yyyyyy
$ export CONFIDENT_API_KEY=zzzzzz

Local with pgvector

Uses src/app-external.py application.

pgvector is a vectorial database based on PostgreSQL.

CONFIG_PATH: JSON config file name.
POSTGRES_USER
POSTGRES_PASSWORD
POSTGRES_HOST
POSTGRES_PORT
POSTGRES_DBNAME

Example:

# Using Python >=3.12
$ python3 -m venv env
$ source env/bin/activate
(env) $ pip install -r docker/requirements.txt
(env) $ export CONFIG_PATH="./docker/config-pgvector.json"
(env) $ export POSTGRES_USER=vectordb
(env) $ export POSTGRES_PASSWORD=vectordb
(env) $ export POSTGRES_HOST=localhost
(env) $ export POSTGRES_PORT=5432
(env) $ export POSTGRES_DBNAME=vectordb
(env) $ ./start.sh

Docker with pgvector

Uses src/app-external.py application.

Modify the docker/docker-compose.yml settings for the sciphi container. This repository contains different service combinations:

docker/config-litellm-pgvector.json: Uses the pgvector db container and litellm as LLM provider (proxy to providers). Uses OpenAI's embeddings. See Providers documentation.
docker/config-litellm-local_embeddings-pgvector.json: Uses pgvector db container, litellm as proxy providers and sentence_transformers with HuggingFace models locally for embeddings.
docker/config-llama-cpp-pgvector.json: Uses the pgvector db container and llama-cpp as LLM provider, using local Ollama models. Uses OpenAI's embeddings.

# Run all containers
$ docker compose -f docker/docker-compose.yml up  # -d to detach

# Run only sciphi-related containers
$ docker compose -f docker/docker-compose.yml up sciphi  # -d to detach

SciPhi

⚠️ This provider requires that no Docker-related files to be present in the root path.

This environment uses the config.json and src/app.py files since those are expected in their infrastructure.

Add required keys in deployments config:

How to run queries

See postman/QUERIES.md.

ariel17 / sciphi-example