Djehuti
TOWER
One self-contained engine. One file in - the analysis out.
Cross-industry analysis, zero install, zero disk writes.
What it is
A single engine. One file in, the analysis out.
Djehuti maps your data file read-only (no copy, no full decompress needed) and runs a cross-sector analysis mix or a one or more specialized modules : Monte Carlo, quant models, time series, genomics, and more then prints a human-readable log plus a machine-readable JSON block. It writes nothing to disk. There is no install and no source: you run one binary.
Built for what you already have. Djehuti runs on an 8GB RAM, 10-year-old Ryzen and still delivers 213M simulations per second, that's a personal supercomputer in a single binary. Offline, read-only, zero disk writes.
Cross-industry analysis
Finance, insurance, energy, telecom, agriculture, climate, genomics, manufacturing, retail, economics, mining, the same engine adapts to your domain. Run the modules that match your question.
Amnesiac mmap, zero disk writes
The file is mapped read-only using MAP_PRIVATE | MAP_NORESERVE.
It is never copied or fully decompressed. The engine writes nothing to disk, no temp
files, no logs, no caches. Every run is independent and leaves no trace.
Read from chunk, no full decompress
Large files (VCF, FASTA, HDF5) are processed in streaming mode. The engine reads from the mmap as needed, never loading the whole file into memory. Swap-friendly, RAM-efficient, and fast.
Performance, proven
Runs on what you already own.
No cloud. No cluster. No monthly bill. Just your computer at full speed.
simulations per second on a 10-year-old Ryzen
50M N(μ,σ) draws in 234 ms, real DJIA params
faster than pandas on real financial data
177×177 Pearson matrix in 10.99 ms vs 136.50 ms
correlations per second
2534 observations, full 177×177 matrix
Tested on an AMD Ryzen 5 1600X (6 cores, 12 threads) with 8GB DDR4 a 10-year-old consumer CPU. Pure AVX2, no custom kernel, no special hardware.
> The more cores and RAM you have, the faster it runs. AVX-512 builds coming soon.
Quick start
Run it in seconds.
# Linux — run the AppImage
$ ./Djehuti-10M-x86_64.AppImage data.npy
# Windows — run the .exe
C:\> djehuti2-10M.exe data.npy
# Pick modules
$ ./Djehuti-10M-x86_64.AppImage data.npy --modules heston,evt,copula
# See all options
$ ./Djehuti-10M-x86_64.AppImage --help
Get the engine.
Compiled, capped at 10,000,000 simulations per run. Linux and Windows builds.