Python & cross-language interop

monlite is a TypeScript library, but a monlite database is plain SQLite. So a Python program can read and write the same .db file with the standard library — no port required. This makes the classic AI split first-class: Python ingests and embeds, Node serves, over one file.

Documents

import sqlite3, json

db = sqlite3.connect("app.db")
for (_id, data) in db.execute("SELECT _id, data FROM users"):
    doc = json.loads(data)
    print(doc["name"])

# write a document Node will read
import time, uuid
now = int(time.time() * 1000)
db.execute(
    "INSERT INTO users(_id, data, created_at, updated_at) VALUES (?, ?, ?, ?)",
    (str(uuid.uuid4()), json.dumps({"name": "Ada"}), now, now),
)
db.commit()

Full-text search

FTS5 ships with Python's sqlite3, so you query the same index Node wrote:

rows = db.execute("SELECT doc_id FROM posts_fts WHERE posts_fts MATCH ?", ("hello",)).fetchall()

Vectors

sqlite-vec has official Python bindings:

import sqlite_vec
db.enable_load_extension(True)
sqlite_vec.load(db)
hits = db.execute(
    "SELECT doc_id, distance FROM docs WHERE embedding MATCH ? AND k = 5 ORDER BY distance",
    (json.dumps(query_vector),),
).fetchall()

Cache & queue handoff

The kv and queue tables are plain SQLite, so a Python worker and a Node API can hand work back and forth: Python enqueues a job (INSERT into the _jobs table) and Node's queue.process(...) picks it up, or vice-versa — sharing one cache and one queue with no extra infrastructure.

A native Python package

Because the whole family is "conventions over one SQLite file," a single pip install monlite can mirror the entire family — documents, kv, queue, cron, fts, vector — over sqlite3. That's on the roadmap; until then, the file format is the interop contract.