tocsin masks numbers, IPs, UUIDs, hex values and emails, then groups the remaining lines into templates with a streaming Drain engine, so a pattern that repeats a million times only needs judging once.
Each new template goes to Jev in a single request carrying your plain-English paging policy and four typed questions, and the answer is cached against that template so nothing gets asked twice.
In the builder's own benchmark, judging 22.8 million log lines this way came down to 13,080 Jev calls for 64 cents, against $45.82 through a pricier flash model or $1,120 asking about every line instead of every template.
What you get
- Full Rust source code under the MIT license
- The Drain-based templating pipeline and the four-question Jev paging policy
Jev
Rust
TypeScript
Node.js
Claude Code
Codex
CodeRabbit Top10
Python
