TiDB

TiDB speaks the MySQL wire protocol, so Memori integrates with it through the same SQLAlchemy / PyMySQL path as MySQL. Memori auto-detects TiDB from SELECT VERSION() and routes it through a dedicated TiDB integration path that reuses the MySQL storage family.

Install

Install TiDB Driver

pip install memori pymysql sqlalchemy certifi

Quick Start

TiDB Connection

from memori import Memori
import certifi
from sqlalchemy import create_engine
from sqlalchemy.orm import sessionmaker

engine = create_engine(
    "mysql+pymysql://user:password@host:4000/memori_db?charset=utf8mb4",
    connect_args={"ssl": {"ca": certifi.where()}},
    pool_pre_ping=True,
    pool_recycle=1800,
)
SessionLocal = sessionmaker(bind=engine)

mem = Memori(conn=SessionLocal)
mem.config.storage.build()

TiDB Cloud Zero Provisioning

For disposable development databases, Memori can provision a TiDB Cloud Zero instance and then use the existing TiDB/MySQL BYODB path:

pip install "memori[tidb-zero]"

from memori import Memori

mem = Memori.provision(provider="tidb-zero", build=True)

You can also use the CLI:

python -m memori provision tidb-zero
python -m memori provision --provider tidb-zero

The provider defaults to https://zero.tidbapi.com/v1beta1/instances. You can override it with MEMORI_TIDB_ZERO_URL if TiDB changes the preview endpoint, and pass TIDB_ZERO_API_KEY if bearer authentication is required.

Provisioned results are cached in $MEMORI_HOME/.memori/provisioning.json or ~/.memori/provisioning.json. This file contains the full database DSN, including credentials. Memori writes it with owner-only permissions where supported, but you should treat it as a secrets file and rotate or delete it when the database is no longer needed. If you switch TiDB Zero endpoints, accounts, or projects on the same machine, pass a distinct tag / cache_key or set cache=False so Memori does not reuse a cached development database from another context.

Connection Strings

Environment	Connection String
Local / Self-hosted	`mysql+pymysql://user:pass@host:4000/memori_db?charset=utf8mb4`
TiDB Cloud	`mysql+pymysql://user:pass@gateway.example.tidbcloud.com:4000/memori_db?charset=utf8mb4`
With TLS params	`mysql+pymysql://user:pass@host:4000/memori_db?charset=utf8mb4&ssl_verify_cert=true`

Use the connection string provided by your TiDB deployment or TiDB Cloud instance. For TiDB Cloud Serverless with SQLAlchemy + PyMySQL, pass TLS through connect_args={"ssl": {"ca": certifi.where()}} so the driver uses a trusted CA bundle. Memori does not require a separate adapter.

Complete Example

import os
import certifi
from sqlalchemy import create_engine
from sqlalchemy.orm import sessionmaker
from memori import Memori
from openai import OpenAI

engine = create_engine(
    os.getenv("DATABASE_CONNECTION_STRING"),
    connect_args={"ssl": {"ca": certifi.where()}} if os.getenv("DATABASE_USE_TLS") else {},
    pool_pre_ping=True,
    pool_recycle=1800,
)
SessionLocal = sessionmaker(bind=engine)

client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
mem = Memori(conn=SessionLocal).llm.register(client)
mem.attribution(entity_id="user_123", process_id="my_agent")
mem.config.storage.build()

response = client.chat.completions.create(
    model="gpt-4.1-mini",
    messages=[{"role": "user", "content": "I work on distributed systems."}]
)
print(response.choices[0].message.content)

mem.augmentation.wait()
facts = mem.recall("distributed systems")
print(facts)

Why TiDB?

MySQL compatibility: Works with the same client stack most Python apps already use.
Distributed SQL: Good match for multi-tenant memory services that need horizontal growth.
Operational simplicity: Easy path from local prototyping to TiDB Cloud without changing Memori code.

Notes

TiDB currently follows the same Memori storage family as MySQL. This keeps the integration low-risk and smooth for other MySQL-compatible systems.
TiDB Cloud Serverless requires secure transport. With SQLAlchemy + PyMySQL, the simplest working setup is certifi plus connect_args={"ssl": {"ca": certifi.where()}}.
If you later want TiDB-specific optimizations, such as native vector support, those can be layered in separately without changing the basic BYODB path.