Skip to content

Latest commit

 

History

History
92 lines (73 loc) · 3.02 KB

README.md

File metadata and controls

92 lines (73 loc) · 3.02 KB

Stream RAG Agent — Python side

This package is the query / serving half of the project. The Go agent handles high-throughput streaming ingestion (Kafka → windowing → embeddings → Elasticsearch); this Python code provides:

  • strag-service — a FastAPI RAG query service (POST /query, GET /health), a Python peer of the Go API. Listens on :8000 so it can run next to the Go API on :8080.
  • strag-producer — a Kafka producer that can emit both financial_transactions and sensor_data (the Go producer only emits the former).
  • strag-query — a CLI client that works against either the Python (:8000) or Go (:8080) endpoint.

Everything reads the same configs/configs.yml and talks to the same Elasticsearch index and Ollama instance as the Go side, so the two stacks are fully interoperable — documents written by Go are queried by Python and vice versa.

Install

cd python
python -m venv .venv
# Windows: .venv\Scripts\activate   |   Unix: source .venv/bin/activate
pip install -r requirements.txt
pip install -e .           # exposes the strag-* commands

Run

Start the infrastructure and Ollama first (see the root README), then:

# RAG query service on http://localhost:8000
strag-service

# Feed data (both topics, ~10 msg/s)
strag-producer --topic all --rate 10

# Ask a question (defaults to the Python service on :8000)
strag-query "Are there any transactions in EUR?"

# Or query the Go API instead
strag-query --url http://localhost:8080/query "Get data for account_id ACC-0833"

Configuration & environment overrides

The config file is located via, in order: an explicit path, $STRAG_CONFIG, then a walk up the directory tree looking for configs/configs.yml.

These environment variables override the file (useful in containers):

Variable Overrides
KAFKA_BROKERS kafka.brokers (comma-separated)
OLLAMA_URL ollama.url
OLLAMA_EMBEDDING_MODEL ollama.embedding_model
OLLAMA_LLM_MODEL ollama.llm_model
ELASTICSEARCH_ADDRESSES elasticsearch.addresses (comma-separated)
ELASTICSEARCH_INDEX elasticsearch.index_name
STRAG_HOST / STRAG_PORT service bind address (default 0.0.0.0:8000)

Docker

docker build -t strag-agent-py ./python
docker run --rm -p 8000:8000 \
  -v "$(pwd)/configs:/config:ro" \
  -e ELASTICSEARCH_ADDRESSES=http://host.docker.internal:9200 \
  -e OLLAMA_URL=http://host.docker.internal:11434 \
  strag-agent-py

Layout

python/
  strag/
    config.py         # loads configs/configs.yml (+ env overrides)
    ollama_client.py  # /api/embeddings and /api/generate
    es_client.py      # same index/mapping/kNN as the Go vectordb package
    rag.py            # build_rag_prompt + RagPipeline
    service.py        # FastAPI app (strag-service)
    producer.py       # multi-topic Kafka producer (strag-producer)
    client.py         # CLI query client (strag-query)
  pyproject.toml
  requirements.txt
  Dockerfile