Put a boundary around autonomy.
VeriCommand checks what Claude Code is about to run, and keeps one hash-chained record of every agent you connect.
A pause before the damage.
In Claude Code, VeriCommand stops a risky command before it runs, such as wiping your changes or reading your secrets. It asks you first, or refuses the worst outright, and writes the command (secrets it recognizes removed, up to 4,000 characters), the rule and that stop to your record when your board can be written (otherwise the prompt says it wasn’t recorded). Your answer to the prompt isn’t recorded.
When it can’t read a command (a command held in a variable, an encoded body it can’t decode, a line over 32 KB) or can’t finish checking it within 8 seconds, it asks you instead of letting it through. If the guard hook can’t start or doesn’t answer in time, Claude Code handles the command under its normal permission rules.
Claude Code today. Reads commands, not intent. Free. What it covers →
One record for every connected agent.
Every connected agent’s work lands on one hash-chained record on your machine. It shows what was recorded, and whether that record was changed afterwards; it doesn’t make the work right.
Any MCP-capable agent can join the same board and record, free. The guard covers Claude Code sessions today.
Two minutes, all real.
No mockups. Recorded on an earlier preview, so some screens have changed since.
AI writes code that looks right.
Looking right and being right are not the same thing — and from inside the tool that wrote it, you can't tell them apart. The decision you made three steps ago gets quietly ignored, everything still compiles, and you find out later, the expensive way.
So VeriCommand brings in a second opinion.
A model from a different vendor checks the work against the evidence attached to it. Findings show their evidence and expose gaps. You keep the final decision. Free includes 3 checks a month; Pro includes 200, and up to 100 of them can run automatically after you say yes once.
“Supported” requires the referee's recorded verdict. A hash-chained record establishes integrity; it does not make an answer correct.
Before you install.
Do I have to switch models?
No. Keep Claude Code and your own plan; VeriCommand works alongside it.
Will I pay for more tokens?
We don’t bill for AI. Your agents run on your own plan, and Pro’s checks are included in the $39.
Is my code sent anywhere?
Nothing leaves your machine unless you use a hosted feature. The guard and the record are local. A check sends the selected decisions, the result being checked (which can quote code), which AI company your agent runs on so a different one checks it, and whether it ran by hand or automatically. It runs only when you click (on Pro, automatically after you say yes once). Three buttons can send source code, each only when you are signed in and click it: Ask VeriCommand sends a sample you attached to the task (up to about 180,000 characters), Reconcile sends the full diff of an agent’s changes (up to 500 KB) to Anthropic, and Check against recorded decisions sends the agent’s result (up to about 60,000 characters), which can quote code. A two-agent conflict check sends both agents’ reports and the paths of files both changed. The MCP server’s drift-compare tool sends a check when a connected agent calls it and a sign-in credential or Pro licence is on the machine. Usage stats are off unless you turn them on. Every network path
Cheap to check. Expensive to miss.
Free: the guard, your record on your machine, and 3 independent checks a month. Pro includes 200 checks a month, and up to 100 of them can run automatically after you say yes once.
$39 / month · subscribe inside the app · cancel anytime, keep Free
- 200 checks a month included, no second AI bill
- Up to 100 of them can run automatically, after you say yes once
- The verdict on conflicts, with both sides quoted
- Cancel anytime and keep Free