ClickHouse Persistence

Columnar vector store for large-scale knowledge retrieval and analytics

ClickHouse stores knowledge embeddings for fast vector search at scale — ideal for large document collections and analytics workloads.

import os
from praisonaiagents import Agent

os.environ["PRAISONAI_KNOWLEDGE_STORE"] = "clickhouse"
os.environ["CLICKHOUSE_HOST"] = "localhost"

agent = Agent(
    name="Analyst",
    instructions="Answer from our knowledge base",
    knowledge=["docs/"],
)
agent.start("Summarise our refund policy")

The user asks from your knowledge base; ClickHouse vector search retrieves relevant chunks at scale.

Quick Start

1. Simple Usage

   pip install clickhouse-connect praisonai
   export PRAISONAI_KNOWLEDGE_STORE=clickhouse
   export CLICKHOUSE_HOST=localhost
   export CLICKHOUSE_PORT=8123

   from praisonaiagents import Agent
   
   agent = Agent(
       name="Analyst",
       instructions="Answer from docs",
       knowledge=["docs/"],
   )
   agent.start("What are our key metrics?")

2. With Configuration

   Create the store directly for full control:

   from praisonai.persistence import create_knowledge_store
   
   store = create_knowledge_store(
       "clickhouse",
       host="localhost",
       port=8123,
       username="default",
       password="",
       database="praisonai",
       secure=False,
   )
   store.create_collection("docs", dimension=1536)

How It Works

ClickHouse is a knowledge store (vector search), not a primary conversation backend. Pair it with SQLite or PostgreSQL for chat history via db().

Configuration Options

Environment variables: CLICKHOUSE_HOST, CLICKHOUSE_PORT, CLICKHOUSE_USER, CLICKHOUSE_PASSWORD.

Docker Setup

docker run -d \
  --name praisonai-clickhouse \
  --ulimit nofile=262144:262144 \
  -p 8123:8123 \
  -p 9000:9000 \
  clickhouse/clickhouse-server:latest

curl "http://localhost:8123/?query=SELECT%20version()"

Best Practices

Related