Crawdad Dashboard runs locally, on the wire between your agents and the models they call. It shows you every action an agent takes, enforces the rules you set, and keeps the raw content on your machine. Install it in one command and watch your agents in real time.
Crawdad Dashboard, Overview. Demonstration data.
Modern AI agents don't answer questions anymore. They run commands, read files, call tools, reach the network, and move data — on your behalf, at machine speed. That's the point of them. It's also the risk. The moment an agent can act for you, a poisoned document or a crafted instruction can make it act against you. Crawdad Dashboard is where you finally see what your agents are doing, and decide what they're allowed to do.
Every agent, every action, in the open. Crawdad discovers every agent on your machine talking to a model — and tells you the truth about each one: the ones it's actively protecting, the ones ready but not yet routed, and the ones it honestly can't reach. Then it shows you what they're doing, live, in plain language — not raw event codes, but "this agent tried to read your AWS credentials," with what the attempt was and why it mattered.
Enforcement on the wire, on your terms. Crawdad reads every request an agent makes as it makes it, and blocks the dangerous ones — credential theft, data exfiltration, prompt injection, destructive commands. You choose how strict it is, from watch-only to full enforcement. Strengthening protection is one tap; reducing it takes a deliberate, confirmed step — so you never lower your guard by accident.
The watcher that can't leak you. To protect an agent, Crawdad has to see everything it does. So by design, none of that content leaves your machine. Inspection happens locally. What can leave — if you turn it on — is a verdict: a category, a severity, a decision. Never the prompt, never your data. The tool that keeps you safe can't quietly become the thing that exposes you.
A record that holds up. Every decision Crawdad makes is written into a tamper-evident chain, each entry cryptographically linked to the last. You can verify the whole thing yourself, offline, with standard math — and if a record has been altered, it tells you rather than waving it through. When you need to show what happened, the evidence is already there.
Crawdad Dashboard is organized around the questions you actually ask.
At a glance: which agents are protected, what's being enforced, what happened in the last day, and the proof trail behind it. It never shows green when the truth is anything less.
Overview. Demonstration data.
Every agent on the machine, what it is, and its true protection state. Honest about what can and can't be protected.
Your live protection posture, shown two ways: a plain-language view for "am I protected," and a rich view with every signal and control for when you want the detail.
The actual rules, in plain terms: allow, ask, deny, or kill — evaluated on every action. Read them, edit them, and test exactly what would happen before anything runs.
Every protection event, explained: what the agent tried, how the attack works, and what it would have cost you. Turn any one into a permanent rule in a click.
The tamper-evident record of every decision, filterable and exportable, ready for whoever needs to see it.
Erase data on request, permanently, with a logged, verifiable trail.
Sessions & Activity. Demonstration data.
Not raw codes — plain language. What the agent tried, how the attack works, and what it would have cost you. Turn any one into a permanent rule in a click.
Crawdad Dashboard is a local, hardened gateway that sits between your agents and the models they call. It bolts on in one command — one environment variable routes the traffic, and nothing else about your setup changes. It runs alongside your existing security tools, not instead of them. Start in watch-only mode, see what it would do, and turn on enforcement when you're ready.
Managing more than one machine? Fleet Console brings every device under one console, with policy you push everywhere and evidence for every client.
Free to install, every feature included. On your machine, in one command.