How many AI agents are running on your machine right now?
Most people can't answer that. Crawdad Sonar can. It finds every process talking to an LLM, tells you what each one can reach, and flags the ones running with no oversight. One command. Nothing leaves your machine.
No account. No API key. No signup. It runs and it's done.
The agents you know about, and the ones you don't
Live LLM connections
Every process with an open connection to an LLM provider — and exactly which process it is, down to its parent and the user running it.
Configured agents
Agents wired up to reach a model even when they aren't running yet, spotted by their provider config and endpoint overrides.
Exposed credentials
LLM API keys sitting in your config files and your shell. It tells you where a key is — never what it is.
Installed agent tooling
The agent CLIs and apps already on the box, from Claude Code and Cursor to Aider and Ollama.
Nothing leaves your machine, and we can prove it
Sonar has no way to phone home. There's no networking code in it that could send your results anywhere, and a test in the build fails if anyone ever adds one. It's a few hundred lines of Rust with zero dependencies — you can read all of it in an afternoon. Some discovery tools upload your inventory to their cloud to work at all. Sonar can't, by design. That's the principle the whole product is built on: what's on your machine stays on your machine.
Zero dependencies
The entire binary is Rust's standard library. There's nothing to audit but the source itself.
No way to transmit
No HTTP client exists in the binary. A build-time test fails if any networking code is ever added.
Keys are never read
The scanner reports that a credential exists and where — a key's value is never stored, printed, or sent.
No account, ever
No signup, no API key, no cloud. It runs entirely on your machine and reports only to your terminal.
What Sonar can't see, because we'd rather tell you
Sonar spots agents by their live connections, their config, and their credentials. It can miss an agent that's sleeping, one routing through a corporate proxy, or one using an endpoint we don't know about yet. Run it with elevated permissions to include agents owned by other users on the machine. We'd rather you know the edges than trust a number that's quietly wrong.
That's the point.
Every agent Sonar found is an open line to a model, with nothing watching what it does. Crawdad closes that gap — it sits between your agents and the LLM, on your machine, and stops prompt injection and data exfiltration before they land. Same install, same 60 seconds.