wrecover was born from a single conviction: no one should permanently lose access to their digital assets because of a forgotten word, a misplaced note, or a corrupted backup.
Every year, billions of dollars in cryptocurrency become permanently inaccessible — not because of theft or fraud, but because of simple human memory. wrecover exists to solve that problem with precision, speed, and integrity.
Help legitimate owners regain access to their own wallets through intelligent, high-performance brute-force recovery using every available clue.
Leverage the full potential of modern hardware — multi-core CPUs, NVIDIA & AMD GPUs — to reduce recovery time from years to hours or minutes.
Every operation runs entirely on your local machine. No seed phrase, no private key, no recovery result ever leaves your hardware.
The catalyst was straightforward: someone close to me had lost access to a Bitcoin wallet from 2018. The seed phrase was partially written down — eight of twelve words were legible, but four had faded beyond recognition. The funds were significant.
I started looking for tools that could help. What I found was either abandoned open-source projects that no longer compiled, slow scripts written in python, or closed-source commercial services that required you to hand over your seed phrase to a stranger. None of these were acceptable.
"If you know eight of your twelve seed words, you shouldn't need to trust a stranger with the other four. You just need a fast, reliable tool that runs on your own machine."
I spent months studying the BIP-32, BIP-39, and BIP-44 specifications in depth — understanding the precise derivation path that transforms a mnemonic into a hierarchical deterministic wallet. I wrote the cryptographic primitives from scratch: SHA-512 for PBKDF2, HMAC-SHA512 for child key derivation, the full secp256k1 elliptic curve arithmetic for address generation.
The first version was CPU-only and could brute-force a single missing word in under a second. But two missing words took minute. Three was impractical. That's when I began the GPU work.
Porting the entire BIP-39 derivation pipeline to CUDA and OpenCL was the hardest engineering challenge of the project — keeping the cryptographic logic bit-perfect while parallelising it across thousands of GPU cores. The speedup was transformative: 650× to 6,000× faster than a single CPU thread (measured: ~2.6M seeds/sec on Tesla T4 vs 3,000–4,000 seeds/sec per CPU thread for 12-word mnemonics; ~21M seeds/sec on Tesla T4 vs 30,000–32,000 seeds/sec per CPU thread for 24-word mnemonics).
What started as a single-mode missing-word finder evolved into a complete recovery platform. Users needed to test passphrase variations, not just missing words. They needed to handle leet-speak substitutions, masked patterns, lists of candidates from exported password managers. Each new use case became a new recovery mode.
Today, wrecover covers 10 distinct recovery algorithms spanning every realistic scenario: from configuration-file-based seed constraints to full cross-product testing of seed lists with passphrase lists. The work-file system — pause, resume, distribute across multiple machines — came from seeing users give up on long jobs because their computer needed to restart. The work-file is compatible with all other CPU or GPU systems.
wrecover is, and will remain, a tool built for legitimate owners recovering legitimate access. The architecture is deliberately local-first: there is no telemetry, no cloud API call, no license server that sees your data. Your seed phrase never leaves your machine.
Every component of wrecover was written with cryptographic correctness and raw performance as co-equal goals. There are no shortcuts in the math.
Full, specification-compliant implementation of the BIP-39 mnemonic-to-seed derivation (PBKDF2-HMAC-SHA512, 2048 iterations) and BIP-32 hierarchical deterministic key derivation with secp256k1 elliptic curve arithmetic. Supports all standard derivation paths and 4 BIP-39 wordlist languages (English, Spanish, French, Italian).
The recovery engine is written in modern C++20. The architecture is template-heavy but compile-time resolved, with zero virtual dispatch in the hot path. Supports up to 10 threads (Free) or unlimited (Pro).
A fully parallelised CUDA kernel runs the complete BIP-39 derivation pipeline on NVIDIA GPUs. The kernel is tuned for maximum occupancy with configurable grid and thread dimensions. Achieves 50–1000× speedup over CPU depending on hardware and recovery mode. Measured throughput: ~2.6M seeds/sec (12-word) to ~21M seeds/sec (24-word) on a Tesla T4; ~26.4M (12-word) to ~264M seeds/sec (24-word) on a RTX PRO 6000. The Free version exposes GPU acceleration for Mode 1 (up to 3 missing words) so users can benchmark their hardware. The Pro version unlocks full multi-GPU acceleration across all compatible modes with linear multi-GPU scaling.
A portable OpenCL implementation enables GPU acceleration on AMD graphics cards and other OpenCL-compatible devices. The same algorithmic pipeline as the CUDA backend, cross-compiled via the OpenCL runtime with device-specific optimisations for memory bandwidth and compute throughput.
Binary work files capture the full recovery state — search space parameters, progress counters, and configuration — in an encrypted, tamper-evident format. Any interrupted job can be resumed exactly where it stopped, on the same machine or a different one, enabling distributed recovery across multiple computers.
From combinatorial seed-model constraints (Mode 0) to cross-product seed × passphrase file testing (Mode 9), each algorithm is purpose-built for a distinct recovery scenario. The substitution engine (Mode 4) uses a precomputed variant database for leet-speak and character-substitution recovery without redundant recomputation.
Every architectural and product decision in wrecover is grounded in four non-negotiable principles.
The Software operates in complete isolation from the network. No telemetry, no analytics, no license validation that transmits data. Your seed phrase, your passphrase, and your recovery results exist only on your hardware.
Cryptographic accuracy is non-negotiable. The BIP-39 and secp256k1 implementations are validated against the official test vectors from the Bitcoin Improvement Proposals. A faster-but-wrong derivation is worse than useless — it makes genuine recovery impossible.
The design philosophy favours intelligent search over blind enumeration. The configuration-file system, mask patterns, substitution rules, and work-splitting architecture all exist to let users apply domain knowledge to reduce the search space by orders of magnitude before a single hash is computed.
The Free Version is genuinely free, with clearly documented limitations. The Pro Version is fully priced and provides full source code access for internal audit and customisation. There are no hidden modes, no crippled algorithms, and no obfuscated binaries. What you read in the documentation is exactly what the software does.
A partially legible paper backup with four unreadable word. eight words known, four missing. The CPU-only prototype could not solve it, so it became the foundation of wrecover.
Porting the full BIP-39 derivation to CUDA/OpenCL enabled recovery of three missing words in minutes and made three missing words feasible for the first time. The parallel architecture became the core of the Pro Version.
Real-world use cases demanded more than missing-word recovery: passphrases with leet-speak substitutions, masked patterns, dictionary attacks, file-based testing. Each scenario became a dedicated algorithm, growing the platform to 10 recovery modes.
Long recovery jobs needed resilience. The encrypted work-file system enables pause/resume of any job and distribution across multiple machines, turning what was a single-machine tool into a coordinated recovery platform.
wrecover 1.0 ships with all 10 recovery modes, full GPU acceleration (CUDA and OpenCL), an encrypted work-file system, multi-language BIP-39 support, and complete documentation for every algorithm. Free and Pro versions available.
Planned improvements include Solana, litcoin, dogcoin and others address verification, extended BIP-39 language packs, and optimised kernels for the latest GPU architectures.
wrecover is a dual-use technology. We take our responsibility seriously.
Every feature, every algorithm, every interface decision is optimised for the most common legitimate scenario: a person recovering access to a wallet they created and own. The tool's power is proportional to the knowledge you already have about your own wallet.
The local-only architecture is both a privacy feature and an ethical stance. We cannot access your recovery data. We have no server that could be subpoenaed. The cryptographic keys you're recovering never touch any infrastructure we control.
The license agreement is explicit: wrecover may only be used to recover wallets you legally own or have explicit written authorization to access. Any other use violates both the license and, in most jurisdictions, the law.
Have a question about your specific recovery scenario? Need technical support for the Pro Version? Want to report a bug or suggest a feature?