MonaDB

A collection is a MutableMapping of keys to documents. Subscript, in, len, and iteration go to storage; get, pop, setdefault, and clear come from the ABC. Assignment is an upsert and missing keys raise KeyError. Every operation is its own commit, and update() commits a whole mapping at once.

import monadb

db = monadb.open("app.db")
users = db["users"]                  # a handle; made on first write

users["alice"] = {"age": 30}         # upsert, one commit
users["alice"]                       # {'age': 30}
"alice" in users                     # True
len(users)                           # 1

users.update({"bob": {"age": 41},    # one commit for the mapping
              "cy": {"age": 9}})
del users["bob"]                     # KeyError if absent

events = db["events"]
for ts in [3, 1, 2]:
    events[ts] = {"at": ts}

list(events)                     # [1, 2, 3]
list(reversed(events))           # [3, 2, 1]
events.first()                   # (1, {'at': 1})

list(events.range(1, 3))         # half-open; None is unbounded
[k for k, _ in db["logs"].prefix("2026-08")]

Keys are str, int, bytes, or a flat tuple of those, held in an order-preserving encoding: encode(a) < encode(b) exactly when a < b. Iteration is key order, not insertion order, so range and prefix scans are b-tree seeks rather than filters. Across types the order is int < str < bytes.

A document is any mapping. Bind a handle to a dataclass or a pydantic model and it converts on write and rebuilds instances on read. The binding lives on the handle, not in the file, so the same documents still read as dicts. Values store as BSON: None, bool, int, float, str, bytes, datetime, list, and nested mappings.

from dataclasses import dataclass

@dataclass
class User:
    age: int
    email: str

users = db.collection("users", User)
users["gwen"] = User(age=44, email="g@corp.io")

users["gwen"]         # User(age=44, email='g@corp.io')
db["users"]["gwen"]   # {'age': 44, 'email': 'g@corp.io'}