Live
Auto-Healer
A watchdog that diagnoses failures with an LLM
Tails logs across an entire fleet, works out what actually broke using an LLM, pushes an alert to Telegram and shows the whole thing on a live health dashboard.
The problem
Unattended systems fail quietly. By the time somebody opens the logs, the run has been dead for six hours and the log is 400MB of noise with the real error buried in the middle.
What we built
A standalone agent — deliberately separate from the systems it watches, so it survives when they don't. It follows every log stream, detects the failure signature, sends the surrounding context to an LLM for a plain-English diagnosis, and pushes it to Telegram with a suggested fix.
Capabilities
- Runs standalone — never edits or depends on the code it monitors
- LLM diagnosis turns a stack trace into a plain-English cause
- Telegram alerts the moment a process dies, not the next morning
- Live health dashboard across every watched service
- Crash-proof by design: the watcher outlives what it watches