How Medic works
A rules-first diagnostic engine that shows evidence, not guesswork. Here is exactly what happens when you paste your logs.
What inputs we analyze
Medic accepts any text-based log output. Paste directly from your terminal, or upload files. We support common config formats too.
Terminal Output
Copy-paste raw terminal output, including ANSI colors and escape codes โ we strip those automatically.
Error Logs
Application logs, stack traces, and error messages from any AI agent runtime โ structured or unstructured.
Config Files
Upload JSON, YAML, TOML, or .env config files. We cross-reference config values with log errors for more accurate diagnosis.
File Uploads
Drop multiple files at once. We support .txt, .log, .md, .json, .yaml, .yml, .toml, .env, and more.
How rules and parsing work together
We do not feed your logs to an LLM and hope for the best. Instead, we use a multi-step deterministic pipeline.
Parse & Normalize
We clean the input, strip artifacts, identify line types, and extract structured entities โ timestamps, error codes, providers, model names, file paths, and API key patterns.
Pattern Match
25+ diagnostic rules fire against the parsed data. Each rule has specific trigger patterns, exclusion conditions, and severity levels. Only rules that match are reported.
Score Confidence
Confidence is based on how many trigger patterns matched within each rule. More evidence = higher confidence. We never claim certainty without evidence.
Generate Report
Results are formatted into a structured diagnosis: what happened, why, evidence excerpts, root cause chain, fix recommendations, and a copyable support summary.
Why this is different from generic AI chat
โ Generic AI Chat
- โข Guesses from context
- โข No evidence cited
- โข Gives 5-10 "maybe try this" suggestions
- โข No confidence level
- โข Your logs are sent to a third party
- โข Different answer each time
โ Medic
- โข Matches known patterns deterministically
- โข Shows exact evidence from your logs
- โข One best fix + alternatives + "do not"
- โข Shows confidence level (high/medium/low)
- โข Logs never leave your browser
- โข Same input = same output
Your data stays with you
Medic runs entirely in your web browser. There is no backend server processing your logs. The analysis engine is a client-side JavaScript module that runs locally on your device.
Zero Transmission
No HTTP requests are made with your log data. Everything runs in the browser.
Nothing Stored
We do not save your logs anywhere. Close the tab and the data is gone.
Fully Auditable
The analysis engine is open-source. You can inspect exactly what it does.