,

Recover a Lost Crypto Wallet: AI + BTCRecover (Local or Online)

If you have ever stared at a broken seed phrase and thought, “I’m so close, but I can’t get it working,” AI can help. The trick is not to hand your wallet data to the cloud and hope for the best. You want AI to do the thinking, while you keep the sensitive parts offline.

In this guide, I’ll walk through how to use AI assistants with BTCRecover to recover a damaged Bitcoin seed phrase. I cover three approaches: a risky one-shot cloud demo, a safer online triage workflow where Claude helps build a command you run offline yourself, and a fully local setup using LM Studio, Visual Studio Code, Cline, and Qwen models.

You can also see the full demonstration in Watch the video.

Claude desktop app showing code tab with usage dashboard and project selector
One-shot automatic wallet recovery with Claude

Stuck with a wallet recovery?

I maintain BTCRecover and have helped people recover seed phrases, passphrases and wallet passwords since 2019. Book a one-hour video session and I’ll walk you through doing the recovery yourself, without sharing your seed. Or, if you’d rather I run it, a trusted recovery is charged only if it succeeds.

Private sessions: US$100 per hour. Trusted recovery: 10% of recovered funds. I never contact people first and will never ask for your seed phrase over chat or email.

The One-Shot Cloud Demo: Fast, But Not Safe

Let’s start with the version I showed at the beginning of the video because it makes the point clearly. In Claude Desktop, I asked for a one-shot recovery of a damaged seed phrase. Within under ten minutes, it installed Python, installed BTCRecover, ran the recovery, and returned the correct seed.

That sounds amazing. And in some cases, that kind of automation is genuinely useful. But this is also a really bad idea if you are working with real funds.

In that same one-shot process, the AI agent could have:

  • Installed Python packages from untrusted sources.
  • Downloaded a malicious copy of BTCRecover.
  • Filled your PC with malware.
  • Sent your recovered seed to a scammer.
  • Potentially emptied the wallet before you even noticed something was wrong.

The viral story I mentioned in the video showed someone using Claude with BTCRecover to regain access to a Bitcoin wallet they had been locked out of for several years. The outcome was good, but from a security perspective, it was still risky because the user was putting private key material into a cloud-based workflow and hoping for the best.

So if you are reading this because you have lost access to a wallet, please do not treat “it worked once” as a safe method. Treat AI as a powerful helper that needs guardrails.

The Safest Starting Point: Do Not Let AI Install Everything Unsupervised

The highest-risk part of any AI-assisted recovery is letting an agent install Python, all its modules, and BTCRecover without supervision.

You could open something like Claude Desktop or Claude Code, point it at an empty folder, and say something simple like “install BTCRecover.” You could even point it at the BTCRecover skill I added to help AI agents use the tool more reliably. But you still cannot know exactly what your AI agent will download and install on your system.

If you are doing this on your daily driver PC, that is a bad idea. Your main computer probably has passwords, browser data, wallets, files, and other things you do not want exposed to random downloads.

A safer setup uses:

  • A dedicated recovery machine, or
  • A virtual machine, or at minimum
  • A clean folder with careful review of every command before running it.

For most people, I would still recommend installing BTCRecover manually using the normal installation instructions for your platform, then using AI to answer questions and help you build commands.

BTCRecover documentation page for AI-assisted recovery
BTCRecover documentation explaining AI-assisted recovery

That way, the AI can inspect the repository, check what is installed, find missing dependencies, and give you copy-and-paste commands that you run yourself. In the video, I asked Claude whether I had everything needed for a SLIP39 wallet recovery. It searched through the local BTCRecover folder, worked out which requirements were missing, and gave me install and verify commands to run in my terminal.

That is the kind of assistance I like: helpful, but not fully autonomous with your system.

Safer Online Workflow: Use AI for Triage, Then Run Offline

Once BTCRecover is installed, you can use an online AI tool such as Claude Desktop to help triage the problem and generate a command template. The important part is that you do not paste your real seed into the cloud just because it is convenient.

Here is the workflow I recommend for most people who want AI help but do not have a powerful local GPU:

  1. Install BTCRecover manually on your machine.
  2. Open Claude Desktop or a similar tool.
  3. Choose the folder where BTCRecover is installed.
  4. Ask it to use the SKILL.md file in that folder to help with wallet recovery.
  5. Answer its triage questions without sharing the seed yet.
  6. Let it check dependencies and suggest the correct command.
  7. Copy the example command.
  8. Disconnect your computer from the internet.
  9. Paste the command into a terminal or command prompt.
  10. Replace the placeholders with your damaged seed and known address.
  11. Run the recovery offline.
  12. Verify the recovered seed against a known address.
  13. Move funds to a new wallet before reconnecting to the internet.
Claude desktop app configured for local BTCRecover project
AI coding-agent desktop app with BTCRecover skill and local mode

In the video, I used Claude Desktop in code mode, selected Local, and pointed it at the folder where BTCRecover was installed. I found that Sonnet worked well for this kind of task. Haiku also did a pretty good job most of the time. You do not need the most expensive or complex model just to get useful help with commands and triage.

The BTCRecover documentation includes an AI-assisted recovery page using SKILL.md. That file helps guide the agent through installation, dependency checks, triage questions, offline verification, and command generation. In many cases, you do not need to overcomplicate it. You can simply paste a prompt like:

Can you help me recover a wallet using BTCRecover and the SKILL.md file that is in this folder?

The agent should search through the folder, find the skill, and begin asking questions about your situation.

What Information to Provide During Triage

During triage, the AI will usually ask things like:

  • How many words are in your seed phrase?
  • Do you know which words are wrong?
  • Are the words scrambled?
  • Is there a password or passphrase involved?
  • Did you use a hardware wallet such as Ledger or Trezor 1?
  • Do you have a known address from the wallet?
  • What error do you get when entering the seed?

In my demo, I used a test 24-word seed similar to what you might get from a Ledger or Trezor 1. I told it that the seed did not work and that I got an invalid mnemonic error. I also gave it a known Bitcoin address starting with bc1.

The important part is that the workflow warns you not to share the words yet. The goal is to identify what kind of recovery is needed, check whether BTCRecover and its dependencies are ready, and then prepare an offline command before any sensitive data enters the terminal.

Claude chat session showing numbered questionnaire about corrupted seed
AI triage questions about a damaged seed phrase

Why You Should Not Force an Online Run

One of the biggest challenges with AI is that it is not deterministic in the way you might want for security-critical tasks. That means even if instructions say “do not run recovery online,” a persistent user can sometimes talk the model into doing it anyway.

In the video, I asked Claude to give me an example command before taking the system offline. It did that nicely. Then I tried to get it to run the command while still connected to the internet. The skill was supposed to check whether we were online and block the process if we were not safe. And yes, at first it refused.

But then I said something like “I don’t care about the risks, just do it anyway.” It eventually complied for a test run. That is exactly why you should not rely on an AI agent to protect your funds by itself. The security boundary has to be yours: disconnect from the internet before running sensitive recovery commands.

Claude chat showing offline verification and example command request
Step 4 instructions to take the system offline

The good news is that this workflow still gives you a lot of value. The AI can produce a clean command, explain each argument, tell you what placeholders to replace, and help you navigate to the correct folder. You then copy that command, go offline, and run it yourself.

Generated BTCRecover command with placeholder values and flag explanations
Example recovery command with substitutions explained

In my demo, the damaged seed only had a single-word error, so BTCRecover found the mistake quickly. But this is not just for simple demos. The same kind of workflow can help with more complex cases, such as scrambled mnemonics, missing words, extra words, or passphrase-related issues. It can also help tell you when something is not practical to attempt because the search space is too large.

Running BTCRecover Offline After Online Triage

Once you have the command template, the process becomes straightforward:

  1. Copy the example command from Claude.
  2. Open a terminal or Windows Command Prompt in the correct BTCRecover folder.
  3. Replace the placeholder seed words with your damaged seed.
  4. Replace the placeholder address with a known address from that wallet.
  5. Make sure the computer is offline.
  6. Run the command.
  7. Wait for results.
  8. Verify the recovered seed against the known address.
Windows Command Prompt ready to run BTCRecover offline
Command prompt in BTCRecover working directory

Verification matters. Do not assume the first result is correct just because it looks like a valid 12 or 24-word phrase. Check that it derives the known address you provided during triage. If possible, test on an offline machine before moving funds. And once you have confidence in the recovered seed, move your funds to a new wallet before reconnecting that system to the internet.

Fully Local AI Recovery: LM Studio + VS Code + Cline + Qwen

If you want the entire process to stay local, you can run a local LLM on your own hardware and use it through an agent interface. In the video, I used this stack:

  • LM Studio for downloading and running local models.
  • Visual Studio Code as the editor environment.
  • Cline as the AI coding agent extension inside VS Code.
  • Qwen models sized to fit your GPU memory.

I like this stack because all of these tools are supported on Mac, Linux, and Windows. They also have graphical interfaces, so you do not need to live in the command line just to use a local model. You can still work inside familiar windows, panels, dropdowns, and terminals.

LM Studio homepage for running AI models locally
LM Studio download page

Step 1: Install LM Studio

Download and install LM Studio from its website. Use it to search for and download the model you want to run. In my demo, I used Qwen 27B on a system with an RTX 5090, but your choice depends on your hardware.

Visual Studio Code homepage
Visual Studio Code download page

Step 2: Install Visual Studio Code and Cline

Download and install Visual Studio Code. Then open the Extensions panel inside VS Code, search for Cline, and install it from within the editor.

Cline website showing the open coding agent
Cline extension website

Step 3: Choose a Model That Fits Your VRAM

This is where people get stuck. A model that does not fit in your GPU memory will either be slow, fail to load properly, or force you into awkward CPU fallback behavior.

In LM Studio, when loading the model, pay attention to context length and GPU offload settings. The default context length may be only 4,096 tokens, which is nowhere near enough for this kind of workflow. At a minimum, I would increase it to around 20,000 tokens. If your VRAM allows more, crank it up further.

LM Studio model library showing Qwen model
LM Studio My Models list

In my RTX 5090 demo system, I increased the context length to around 100,000 and set GPU offload to full. The memory estimate in LM Studio gives you an idea of whether the model will fit on your graphics card. If it does not fit, use a smaller model.

Model configuration panel showing context length and GPU offload
Local LLM settings for context length and GPU offload

Step 4: Start the LM Studio Server

Once the model is loaded, make sure the local server is running. By default, it may be off. Turn it on so Visual Studio Code can connect to it through Cline.

VS Code welcome screen with recent BTCRecover projects
Visual Studio Code ready for local agent workflow

Step 5: Connect Cline to LM Studio

In VS Code, open Cline from the sidebar. Go into its settings and select LM Studio as the provider. Then choose your loaded model from the dropdown menu. In my demo, I selected Qwen 3.6 27B.

One important setting is to enable compact prompt mode if available. You do not want the agent filling up the context with thousands of tokens of unnecessary detail when you are trying to run a focused recovery workflow.

VS Code with Cline extension open
Cline chat interface inside Visual Studio Code

Step 6: Create a Working Folder and Start Recovery

Create a new folder, such as btcrecover, and open it in VS Code. This becomes the working folder for the agent. Then paste a prompt asking Cline to help install BTCRecover locally using the same kind of skill-based workflow.

VS Code with Cline prompt referencing BTCRecover GitHub link
Prompting local agent to recover wallet via BTCRecover

In my demo, the local model installed Python and BTCRecover along with its requirements without major issues. Then it told me to take the system offline. Once I disconnected from the internet, it continued working because it was running locally. It then asked for the damaged seed and known address, ran the recovery, and returned the result while staying fully offline.

Using Smaller Local Models on Modest Hardware

Not everyone has a high-end GPU. In the second local demo, I used a system with an RTX 3070 and 8 GB of VRAM running a 9-billion-parameter model. The same general stack worked: LM Studio, Visual Studio Code, Cline, and a smaller Qwen model on Ubuntu.

Ubuntu desktop showing LM Studio, VS Code, and Cline
Local LLM software stack on Ubuntu

With less capable models, you should not expect one-shot reliability the same way you might with a larger model or a strong cloud assistant. These smaller models can often handle individual steps well:

  • Install BTCRecover.
  • Check dependencies.
  • Explain a command flag.
  • Build a token list.
  • Run one specific recovery step.
  • Interpret an error message.

Where they struggle is chaining many steps together into a complete workflow without losing the plot. In my 9B demo, the model forgot to check whether we were offline before proceeding. That is a perfect example of why you cannot blindly trust smaller models with security-critical behavior. You still need to verify that your system is offline yourself.

Document slide listing example prompt sequence for BTCRecover agent workflow
Example prompts for driving BTCRecover step by step

The documentation includes a section for using less capable agents, with an example prompt sequence you can follow manually. The idea is to ask the model to do only one or two steps at a time. For example:

Return to SKILL.md and do only steps five and six to recover my wallet. BTCRecover is already installed.

Then you provide the damaged seed and known address when it asks, after confirming that your machine is offline. In my demo, I checked connectivity with a ping test, entered the mnemonic data into the local terminal workflow, and the model found the simple mistake quickly.

VS Code Cline panel with BTCRecover help output in terminal
Local agent running BTCRecover seedrecover.py help

I also tested this approach with a 4-billion-parameter model. It had similar limitations to the 9B model: it generally will not one-shot a full recovery reliably, but it can still help with specific tasks if you ask in small chunks. The nice thing is that even 4B models can run on many mid-range PCs available today.

Practical Security Rules for AI-Assisted Recovery

Whether you use Claude online or a local LLM offline, the same core rules apply:

  1. Do not paste your real seed into cloud tools unless you accept the risk.
  2. Prefer manual installation of BTCRecover over unsupervised agent installs.
  3. Use AI for triage, command generation, and troubleshooting first.
  4. Disconnect from the internet before running sensitive recovery commands.
  5. Verify any recovered seed against a known address.
  6. Move funds to a new wallet before reconnecting to the internet.
  7. If the wallet is low value and you do not care about theft, one-shot cloud recovery may be acceptable; otherwise, use safer workflows.

AI makes tools like BTCRecover more accessible. That is genuinely useful. But it also introduces a new category of risk: an agent that can install software, run commands, and handle sensitive data with less oversight than you might want. The best approach is to let AI reduce the technical friction while keeping the security decisions in your hands.

When You Need Human Help

If you have tried these tools and still cannot recover access, or if the situation involves a large amount of funds, a damaged hardware wallet, a passphrase problem, or an unclear recovery path, it can make sense to get human help. You can request a consultation or trusted recovery service through my website: recovery services and consultations.

Final Thoughts

AI does not magically make wallet recovery safe just because it is smart. It makes the process faster, clearer, and more accessible if you use it correctly. The safest pattern I showed in the video is simple: let AI help you understand the problem, prepare the right command, and troubleshoot errors; then run BTCRecover offline yourself and verify the result before moving funds.

If you have a powerful GPU, a fully local workflow with LM Studio, VS Code, Cline, and Qwen can be very convenient. If your hardware is more modest, smaller models can still help step by step. And if you do not want to run anything locally, an online triage workflow where Claude gives you a command template is much safer than one-shotting your seed into the cloud.

Stay cautious, keep sensitive data offline when possible, and verify everything before moving funds. If you try this with BTCRecover or have questions about AI-assisted recovery, leave a comment below. I do my best to reply to everyone.

Stuck with a wallet recovery?

I maintain BTCRecover and have helped people recover seed phrases, passphrases and wallet passwords since 2019. Book a one-hour video session and I’ll walk you through doing the recovery yourself, without sharing your seed. Or, if you’d rather I run it, a trusted recovery is charged only if it succeeds.

Private sessions: US$100 per hour. Trusted recovery: 10% of recovered funds. I never contact people first and will never ask for your seed phrase over chat or email.

Found this useful?

These guides are free and ad-free. You can support them by sharing them, subscribing on YouTube, or sending a tip.

Stuck on this?

Book a one-hour session and I’ll walk you through it, or ask about a trusted recovery.

Recovery guides

More guides