wayyDB - High-Performance Time-Series Database

Performance That Speaks for Itself

Median of 5 runs, single-threaded. Smaller is better.

As-of Join

Create Table

Aggregation

Window Fns

Persistence

AS-OF JOIN: 1M trades x 1M quotes, 10 symbols

wayyDB

142 ms

1.0x

DuckDB

345 ms

2.4x

Polars

568 ms

4.0x

pandas

8,234 ms

58.0x

CREATE TABLE: 1M rows, 7 columns (OHLCV + symbol)

wayyDB

12 ms

1.0x

Polars

35 ms

2.8x

DuckDB

89 ms

7.2x

pandas

145 ms

11.8x

AGGREGATIONS: sum + mean + min + max + std on 1M rows

wayyDB

0.8 ms

1.0x (SIMD)

NumPy

2.9 ms

3.6x

Polars

4.1 ms

5.1x

DuckDB

5.6 ms

7.0x

pandas

16.2 ms

20.3x

WINDOW FUNCTIONS: mavg(20) + ema(0.1) + mstd(20) on 1M rows

wayyDB

5.2 ms

1.0x

Polars

22 ms

4.2x

pandas

65 ms

12.5x

LOAD FROM DISK: 1M rows, 7 columns

wayyDB

0.05 ms

mmap (instant)

Polars

18 ms

360x

DuckDB

32 ms

640x

pandas

62 ms

1240x

Reproduce it yourself:

pip install wayy-db[bench] && python -m benchmarks.benchmark --compare all

Quick Start

From pip install to as-of join in 30 seconds.

import wayy_db as wdb
import numpy as np

# Create trades table from numpy arrays
trades = wdb.from_dict({
    "timestamp": np.array([1000, 2000, 3000, 4000, 5000], dtype=np.int64),
    "symbol":    np.array([0, 1, 0, 1, 0], dtype=np.uint32),
    "price":     np.array([150.25, 380.50, 151.00, 381.25, 152.00]),
}, name="trades", sorted_by="timestamp")

# As-of join -- O(n log m), the join quant finance needs
result = wdb.ops.aj(trades, quotes,
                     on=["symbol"],
                     as_of="timestamp")

# Zero-copy NumPy (no data copied!)
prices = trades["price"].to_numpy()

# SIMD-accelerated aggregations
avg_price = wdb.ops.avg(trades["price"])

# Window functions
mavg = wdb.ops.mavg(trades["price"], window=20)
ema  = wdb.ops.ema(trades["price"], alpha=0.1)
      

Built for Time-Series

Every feature designed for tick data, OHLCV, and quant workflows.

SIMD Aggregations

AVX2-accelerated sum, avg, min, max, std. Processes 8 doubles per cycle. 10x faster than pure Python loops.

Zero-Copy NumPy

Memory-mapped columns share directly with NumPy arrays via the buffer protocol. No serialization overhead.

As-of Joins

O(n log m) via binary search on sorted indices. The join quant finance lives on. Native, not a workaround.

Window Functions

Moving average, EMA, rolling std, mmin, mmax -- all O(n) single-pass. Built for tick-level data.

Persistent Storage

Tables saved as mmap files. Load 10M rows in microseconds. No deserialization, no parsing, just page faults.

Streaming API

FastAPI REST + WebSocket. Real-time ingestion at 1M ticks/sec. Pub/sub with InMemory or Redis backend.

How It Compares

Honest feature comparison. We show what we have and what we don't.

	wayyDB	kdb+	DuckDB	Polars	pandas
License	MIT	$$$$$	MIT	MIT	BSD
Language	C++20	q/k	C++	Rust	Python
Python Bindings	yes	limited	yes	yes	native
As-of Join	yes	yes	yes	yes	yes
Window Join	yes	yes	no	no	no
SIMD	AVX2	yes	yes	yes	no
Zero-Copy NumPy	yes	no	partial	partial	no
Streaming API	yes	no	no	no	no
Memory-Mapped	yes	yes	partial	no	no
Price	Free	~$100k/yr	Free	Free	Free

Install

pip install and go. No system dependencies required.

$ pip install wayy-db # Core library

$ pip install wayy-db[api] # + FastAPI server

$ pip install wayy-db[bench] # + Benchmark suite