fawdyHow It Works
Something breaks. If you’re lucky, your observability stack tells you exactly what happened. Most teams aren’t that lucky. The telemetry covers half the services, the engineer who configured it left, and the real debugging tool is grep.
So someone starts digging. Logs, timestamps, config files, trying to piece together what happened. It takes hours, and it happens every time. Then you get to do it all over again in a post-mortem writeup.
What if you never had to read logs again?
Fawdy reads them for you. Point it at an incident and it observes, forms a hypothesis, runs read-only diagnostics to test it, and loops until it has an answer.
No runbooks. No tribal knowledge. Just a structured investigation that runs in minutes instead of hours.
Is it safe?
Diagnosing incidents means running commands on your infrastructure. Every one passes through ShellGuard, a deterministic parser that checks each command against an allowlist. Pass or block. No LLM, no judgment calls, no probability. Try it for yourself.
Parse
awaiting command
Validate
awaiting parse
Goals
0/3- Run a safe command and watch it pass
- Try a dangerous command and see it get blocked
- Find a command blocked by a specific flag(hint: tail -f)
ShellGuard is Open Source
You don’t have to trust us. Read the code.
Why bash?
Most AI agents talk to servers through custom tool APIs. Check disk usage? Build a tool. Read logs? Build another. Restart a service? Another one. Before long you are maintaining dozens of brittle abstractions that still cannot handle the task the LLM has not seen before.
LLMs already know bash. Trained on millions of shell sessions and man pages, they can compose commands for nearly any sysadmin task out of the box. One interface instead of fifty.
Custom Tools
Fawdy
Supports any stack.
General-purpose LLMs hallucinate commands, reference deprecated flags, and mix up syntax between similar tools. Fawdy ships with a library of skills. Curated reference documents for specific technologies, each containing verified commands, correct flags, and known failure modes. When an investigation starts, Fawdy scans the server, detects what’s running, and loads the relevant skills. Documentation, not guesswork.
Fawdy ships with skills for the most common server technologies out of the box, with more being added regularly. Need coverage for a legacy system or internal tooling? Custom skills can be written for any stack.