Everything you'd ask before plugging in.

When a wallet is recovered, Brute Brothers settles with the customer — they receive the funds minus our 15% recovery fee and on-chain transaction costs. The remaining 15% is split between Distribrute and the participant(s) whose compute produced the hit. Splits depend on workload type, case priority, and the wallet inventory the case was drawn from. Exact percentages are published in the participant agreement before you commit any compute, not after.

For password recoveries, you earn a participation credit proportional to the work done, redeemable against future hit-share bonuses. For seed-phrase recoveries on the confidential tier, the keyspace is bounded — someone is guaranteed to hit eventually, and credits accumulate until they do.

After the customer settles (typically 1–3 business days post-KYC re-verification) and funds clear the on-chain confirmations we require, the participant payout is queued. We pay in the same asset as the recovered wallet (BTC, ETH, etc.) or in USDC at the participant's choice.

Yes. The threshold is set per-asset to cover network fees plus a small buffer. Sub-threshold balances accumulate and are paid out on your next over-threshold settlement, or on request.

Not necessarily. Several supported wallet formats — notably Armory and Coinomi — use memory-hard KDFs (ROMix, scrypt) where high-bandwidth CPUs outperform consumer GPUs. The agent self-benchmarks on first run and only accepts jobs it can run efficiently. A modern Threadripper or EPYC with plenty of RAM is competitive for the CPU-bound workloads.

Compute tier: NVIDIA RTX 30/40/50-series with 8 GB VRAM or more. AMD RDNA3 is supported with reduced throughput. Confidential tier: NVIDIA H100 (SXM5 or PCIe), H200, B200 with Confidential Computing mode enabled and a recent driver/firmware. The attestation handshake will refuse any other hardware automatically.

The agent honors a configurable power cap and an idle-only mode that pauses work when your machine is interactive. Most participants run during off-peak hours or on dedicated rigs already optimized for power-cost-per-flop.

Yes. The agent treats each compute device as a separate worker and routes appropriate workloads to each. A box with two RTX 4090s and a 64-core Threadripper will see GPU jobs queued on the cards and CPU jobs queued on the cores in parallel.

On the compute tier: no. We send a ~40-byte verifier extract, not the wallet. A password match reveals only the password — the encrypted seed and private keys live in a file we never transmit, so the recovered password by itself cannot unlock funds. On the confidential tier: the wallet, candidates, and recovered keys exist only inside the GPU's encrypted memory, accessible only to the attested enclave, not to the host OS that runs the agent.

The extract format is publicly specified and open-source (BTCRecover's extract scripts). The agent verifies that each received payload matches the published structure — fixed length, parseable header, expected IV+block layout — before passing it to the hashing backend. Any payload that does not parse as a valid extract is rejected and reported.

Your agent triggers NVIDIA's remote attestation flow. The GPU produces a quote signed by a hardware-rooted key NVIDIA burned at manufacturing. Our server verifies the signature against NVIDIA's PKI, checks that Confidential Computing is enabled, checks the firmware version, and confirms the channel-encryption public key was generated inside that specific enclave. Only then do we send encrypted work payloads. The host that runs the agent is never trusted with cleartext.

Side channels (timing, cache, power) against confidential-compute hardware are an active research area. They typically require local access, specialized equipment, and known vulnerabilities. We track the published research and update firmware/driver minimums in attestation policy as new findings warrant. They are not in the threat model for opportunistic theft, which is what the participant agreement defends against.

Yes. The agent binary is signed and reproducibly built from a public repository. The enclave-side code (for the confidential tier) is published with its hash; the hash is part of the attestation measurement, so you can verify the code running in the enclave matches the source you read.

Two reasons. First, payouts cross jurisdictional and tax lines — we need to know who we're paying. Second, the participant agreement is legally enforceable only if the parties are identified. Anonymous participation would also remove our main defense against bad actors on the compute tier: real-name accountability backed by the agreement they signed.

Most jurisdictions where crypto compute services are not specifically prohibited. We maintain a current allow-list and reject signups from sanctioned jurisdictions or those where our compliance counsel has flagged blocking risk. The waitlist form captures jurisdiction so we can confirm eligibility before any compute is dispatched.

A standard institutional KYC vendor used by crypto exchanges. ID document + selfie + proof of address. Verification typically completes within 24 hours. Data is held by the vendor, not by us — we receive a pass/fail and a hashed reference.

A BIP39 seed phrase with N missing words has a finite candidate space. Once partitioned across the fleet, every chunk eventually gets evaluated, and exactly one chunk contains the hit. We commit to running the entire keyspace within a published window, and we publish the chunk size and total chunk count up front. Whoever's chunk contains the hit earns the share. There is no "we ran out of time" outcome — the case completes deterministically.

Compute tier: yes, with GPU/CPU passthrough configured. Confidential tier: the agent expects a CC-capable host configuration (confidential VM with GPU passthrough), and the attestation chain verifies the VM and the GPU together — substitution is detected automatically.

Yes. The agent's control plane lets you pause at any chunk boundary. In-progress work is checkpointed and resumed on restart. Partial work earns partial participation credit, even if you never restart.

Participation credit for work already completed is honored. Cases are rarely canceled mid-run, but the agreement is explicit about this scenario.