What is free, exactly?
The record. Clawmont Free writes every tool call your agent makes to a hash-chained log on your own disk, hands you a receipt at the end of each run saying what it touched, and puts an audit.html beside the trail so you can read it in a browser. There is no card and no expiry, the same detectors run as on the paid plan, and nothing leaves your machine. It is a complete product, not a trial.
What do I actually get for paying?
Not having to read the log. Clawmont Security reads the record for you and sends a digest across your sessions and projects, which arrives on the quiet days too; it keeps a searchable copy of the trail off the machine that wrote it; it sends alerts to Slack, Discord, Telegram or email; and it can deny a tool call before it runs. The detectors are identical on both plans — we do not hold back detection to sell an upgrade.
What does a persona include, and is it a subscription?
A persona ($30 one-time) is a pre-configured OpenClaw setup — curated tools, skills, and MCP servers for your role. Apex ($40 one-time) merges every persona into one package. Personas are setup convenience, not the security product, and carry no recurring charge.
What is Clawmont Security (Guardrails) and what does it cost?
Clawmont Security — plan name "Guardrails" — is the paid rung. It reads your record for you, keeps a copy of it off the machine, alerts you off-machine, and can refuse a tool call before it runs. On Claude Code it inspects three boundaries locally (your prompt, the tool call, and the tool result) and denies at the tool call, which is the one boundary that can deny. It costs $9.99/month bundled with a persona or Apex, or $19.99/month standalone. Bundled and standalone are the same product; bundled is the setup-buyer price.
Does the paid plan let me undo what the agent did?
Not yet, and we would rather say so on the pricing page than in a changelog. Checkpoint and undo for a whole run is not built. What exists today is the record of everything that happened and a deny on part of the irreversible list — a recursive delete outside a build directory, a private key read, a .env going out over the wire. A force-push, a reset --hard and a secret going into a commit are still allowed even with enforcement on; we probed our own build on 2026-08-12 and that is what it did. Fixing that is the next thing we are building.
Is there a free trial?
Two ways to try it. Clawmont Free is not a trial at all: the record is free with no card and no expiry, so you can read your own traffic for as long as you like before deciding to pay for anything. Separately, the paid subscription includes a 7-day free trial on both the $9.99/mo bundled and $19.99/mo standalone rates, so you can run it against real work before you are charged. Personas and Apex are one-time purchases with no trial. After the trial, the subscription can be cancelled at any time with a pro-rata refund for the unused period.
Which coding agents does this support?
Claude Code, and only Claude Code. Adapters exist for Cursor, Codex CLI, Cline, Devin CLI and Aider, and none of them has been driven against a live attack yet, so we do not list them as supported — a written adapter is not a proven one. Continue.dev cannot work at all until Continue changes something upstream. The full state of every integration is published on the home page rather than summarised as a logo wall.
What is the refund policy?
Persona and Apex purchases are one-time licenses and final sale. Guardrails subscriptions are refunded pro-rata for the unused period when cancelled. Statutory EU/EEA consumer rights still apply.
Do my API keys ever leave my machine?
No. API keys are collected only in the terminal installer, stored in your OS keychain, and never sent to api.clawmont.com. The only thing that can leave your machine is opt-in, redacted, HMAC-signed alert metadata — and only if you enable the cloud alerts add-on.
Will Clawmont block my normal work?
Not unless you ask it to. Clawmont starts in monitor mode: everything is recorded to your local trail and nothing is denied, so you can see exactly what the brakes would have stopped on your own traffic before you turn them on. Denying happens at the tool-call boundary, and you can switch modes at any time. False-positive behaviour is measured and published on the security page rather than promised away.
What does Clawmont Security not do?
It is a detection layer, not a sandbox: it inspects agent activity and can deny tool calls, but it does not containerize your agent or filter syscalls, and a detection that misses means the call runs. We publish our detection numbers and known limitations on the security page instead of overstating what a detection layer can do. Pair it with OS-level isolation if you need containment.