Open distributed research

Find Bitcoin with one command.

OpenPuzzle coordinates Bitcoin Puzzle ranges between anonymous clients. Computation and private keys remain entirely on each participant’s machine.

$ openpuzzle run

Live database

Project statistics

Public statistics are calculated from the central range database.

Active clients

2

Ranges running

6

Ranges completed

9 389

Ranges cancelled

0

Range completion

0.003522662952%
41 588 263 645 168 009 of 1 180 591 620 717 411 303 424 keys

Search keyspace

Start
400000000000000000
End
7FFFFFFFFFFFFFFFFF

Range database

Total ranges
9 401
Failed ranges
3

Distributed network

Network overview

Live aggregate information from OpenPuzzle clients active during the last two minutes.

Clients online

0 Recently connected

Clients running

0 Currently searching

Clients idle

0 Ready for work

Active assignments

0 Valid active leases

CUDA workers

0 Actively searching

OpenCL workers

0 Actively searching

CPU threads

0 Actively searching

Network speed

0 MKey/s Waiting for network data

Latest activity

Recently completed ranges

Range Start End Completed
#9427 69xxxxxxxxxxxxxxxx 69xxxxxxxxxxxxxxxx 2026-11-02 19:47:32
#9426 65xxxxxxxxxxxxxxxx 65xxxxxxxxxxxxxxxx 2026-11-02 19:47:16
#9425 5Fxxxxxxxxxxxxxxxx 5Fxxxxxxxxxxxxxxxx 2026-11-02 19:42:43
#9424 5Fxxxxxxxxxxxxxxxx 5Fxxxxxxxxxxxxxxxx 2026-11-02 19:29:38
#9423 6Exxxxxxxxxxxxxxxx 6Exxxxxxxxxxxxxxxx 2026-11-02 19:26:33
#9422 6Exxxxxxxxxxxxxxxx 6Exxxxxxxxxxxxxxxx 2026-11-02 19:18:18
#9421 7Bxxxxxxxxxxxxxxxx 7Bxxxxxxxxxxxxxxxx 2026-11-02 18:48:49
#9420 51xxxxxxxxxxxxxxxx 51xxxxxxxxxxxxxxxx 2026-11-02 18:48:12
#9419 69xxxxxxxxxxxxxxxx 69xxxxxxxxxxxxxxxx 2026-11-02 18:44:35
#9418 76xxxxxxxxxxxxxxxx 76xxxxxxxxxxxxxxxx 2026-11-02 18:33:07

Open source client

Install. Run. Win.

The website assigns the work. Your computer decides how to execute it using its locally configured engine and GPU profile.

Ubuntu / Linux

Version 1.0.53 · x86_64

Stable release

Vanity now uses blocking CUDA synchronization to release host CPU while waiting for GPU work. The exact candidate passed known-address tests on five RTX 3080 cards and a 30-second simultaneous local measurement. Search kernels, saved profiles, one-hour ranges and common controls are preserved.

Download 1.0.53 for Ubuntu

Before you start: install your GPU driver

NVIDIA · CUDANVIDIA driver for your GPU ↗
A driver compatible with the included CUDA engine is required.

AMD · OpenCLAMD driver with OpenCL support ↗
Confirm support with the local engine self-test.

Prebuilt Windows package: no Visual Studio or full CUDA Toolkit / AMD SDK required. Development SDKs are for building from source. AMD Windows hardware validation is still pending.

Windows installation and first run
  1. Download and run the Windows installer .exe. Follow the graphical setup.
  2. Open OpenPuzzle Test - Auto Benchmark from the Desktop, or OpenPuzzle Test in the new Start Menu folder. Normal startup opens only the interface.
  3. For your first profile, choose CPU, CUDA or OpenCL and the device under Tools, then click Benchmark. Review and confirm the local measurement. GPU parameters are automatic; allow about 10–20 minutes for CUDA/OpenCL.
  4. A successful benchmark saves and selects the profile. Start the search explicitly when ready.

The coordinator is automatic. The public package includes no personal profiles or client data. Finish a previous search with Safe Stop before starting another search on the same GPU.

NVIDIA power and fan controls are included; AMD controls are not enabled in this Windows build. Automatic benchmarks were confirmed on Windows 10 and 11. Setup 0.1.5 was reviewed on Windows 11; clean-machine compatibility remains under testing.

Verify the Windows download

Compare the installer with the published Setup 0.1.5 SHA-256 checksums.

cb8c15587be2f87c6ff607a33fab7de1f69a0c77e8d588caaad7ed55c5a4e3a5

Alternative ZIP SHA-256: ff52d451d0e513f53df8fcf037853e52cd93fa5d910b646eac6e137211fb897a.

Linux terminal commands
1 openpuzzle run
2 openpuzzle status
3 openpuzzle update --check
What changed in Linux 1.0.53
OpenPuzzle 1.0.53

Vanity now uses blocking CUDA synchronization to release host CPU while waiting for GPU work. The exact candidate passed known-address tests on five RTX 3080 cards and a 30-second simultaneous local measurement. Search kernels, saved profiles, one-hour ranges and common controls are preserved.

Compiled targets include sm_50 through sm_90 and compute_50 PTX. Other GPU families are not all hardware-validated.

Read the release history →

Security principles

Your discovery belongs to you.

OpenPuzzle coordinates work but never receives a discovered private key.

No registration

Participation is anonymous. Each installation generates only a random local UUID.

Keys remain local

A discovered private key is stored exclusively on the computer that found it.

Blockchain verification

A potential solution creates only a metadata report pending review. A puzzle is marked solved only after independent confirmation through the Bitcoin blockchain.

Local engine configuration

GPU, backend, executable and tuning settings stay under the participant’s control.

Documentation

A small and open protocol.

Clients anonymously claim a range, report non-sensitive progress and confirm when that range has been searched. A potential solution report contains only assignment and anonymous client identifiers. Private keys and raw engine output are never transmitted.