Verlox is the AI terminal ledger for Windows that records, explains, and reverses every action across eleven hosted models plus a built-in offline one. See the plan before it runs, the reason on every step, and undo any change with one click. AI moves fast. Verlox keeps the record.
Free to start, 15 credits a day. Windows today, macOS and Linux soon.
clean up the build and reinstall dependencies
Read the project files
ls · cat package.json
Delete the build cache · saved to Vault
remove .cache · dist
Reinstall dependencies
npm install
Verlox is an AI terminal ledger for Windows. You describe what you want in plain English; Verlox turns it into a numbered plan with the exact commands, a reason on every step, and the files each step will touch. Nothing runs until you say so.
After any AI session, Verlox can tell you what changed, when, why, which AI did it, and how to undo it. One ledger across every model. One permission system across every action. One click to reverse any change.
Warp runs your commands fast. Claude Code helps you write code, and only guards Claude. Verlox is one ledger that keeps every AI's actions visible, explained, and reversible. Cross-vendor by design, because none of the labs will ever build this for you.
Anthropic, OpenAI, Google, xAI, and the best open models, all wired in. A local model you can run offline. Your own API key if you prefer that. Switch between them in the picker without leaving the chat.
Billed in credits. Free tier covers daily use of the smaller ones.
A Llama 3.2 3B model downloads once and then runs entirely on your machine. No network, no credits, nothing leaves your device. Verlox also detects Ollama and surfaces any model you have pulled there.
Free and unlimited. About 2 GB on first use.
Paste an OpenAI or Anthropic API key (or any OpenAI-compatible endpoint) and call your own account directly. Verlox never sees your key beyond your machine.
You pay your provider. No Verlox credits used.
Each request becomes a numbered plan with the real commands and the files they will touch, logged to one ledger no matter which model you used.
Every command shows its reason and a risk level in plain English before it runs. So when you look back, you remember why you said yes.
Anything the AI deletes or overwrites is kept restorable. One click to put it back. Keep changes for 24 hours, 7 days, or forever.
Verlox turns your request into a clear, numbered plan: every command, every file it will touch, and a risk score on each step. Low for reads and searches, medium for installs and edits, high for deletes, force-pushes, and production changes. The risky steps always stop and ask. You approve the whole plan, edit it, or decline.
Check the current branch
git status
Force-push to main
git push --force origin main
.cache/
folder · deleted 10:36 · expires in 7d
vite.config.ts
overwritten 10:33 · expires in 7d
old-seed.sql
deleted 10:31 · expires in 7d
Before any change lands, Verlox keeps a copy. Files, folders, configs, scripts: anything the AI deletes or overwrites stays in the Recovery Vault and reverses with one click. Choose how long copies stick around: 24 hours, 7 days, or forever. It catches your own mistakes too.
Set a rule for each kind of action. Let it read files and search the project freely. Make it ask every time before installing packages or hitting the network. Block production access entirely. Verlox enforces your rules on every plan, so you're not re-approving the same safe steps forever, or worrying about the dangerous ones.
3
files edited
2
packages added
0
deletes
Simulate a plan and Verlox predicts the outcome without touching anything: how many files change, what gets installed, whether anything is deleted, plus a before-and-after diff of every edit. When it looks right, run it for real. When it doesn't, you never had to find out the hard way.
Type your intent in plain English, like "reinstall dependencies", "commit my changes", or "what's taking up my disk?".
It lays out the exact steps, scores the risk of each, and shows the files it will touch, all before running a thing.
Approve, decline, or simulate. It runs only what you allow, and anything it changes stays restorable in the Vault.
What the AI can see, what's still running, and what happened while you were away. The quieter parts of keeping a terminal accountable.
A chronological record of every action the AI took: installed a package, edited a config, restarted a server, each with its risk and result.
A plain list of what the AI can and can't see: your working folder and open tabs, but never unopened files, your secrets, or other apps.
Dev servers and watchers get their own pane, tagged by who started them (you or the AI). See what's live, open its URL, or stop it in one click.
No flags to memorize. Say what you want; Verlox finds the right commands for your shell and explains the result in plain words.
Stuck on an error you can't put into words? Drop in a screenshot and work out the fix together.
Verlox keeps itself current in the background. New features and fixes arrive without a reinstall.
No card to start. The whole control layer (plans, risk scoring, permissions, Recovery Vault) and the offline local model are in Free. Move to Pro for the flagship models, dry-run simulation, and longer recovery.
The full control layer, no card needed.
For people who live in their terminal.
Cancel anytime. Free credits refill every day; Pro credits refill every week.
AI agents are powerful, but every tool tracks only its own. I wanted one ledger that worked across all of them: a record of what changed, when, why, which AI did it, and how to undo it. So I built it. It's early, it's improving fast, and I read every piece of feedback. If something's broken or missing, tell me.
Free to start. Download it, ask it to do something real, and watch it show you the plan before it moves.
15 credits a day, free. macOS & Linux soon.