Group log lines into patterns and ask which ones matter
Collapse millions of log lines into templates, then ask Jev one question per template to surface the ones worth an engineer's look.
Author's figure: Author's figure: 22.8M lines in 6 minutes for $0.64, against an estimated $1,120 with an LLM (Ateeq). Reported by the builder, not measured by us.
The use case
On-call engineers and agents both drown in logs. Running a language model over every line is far too slow and expensive. This tool first groups lines into templates in code, so millions of lines become a few hundred patterns, then asks Jev one typed question per pattern, such as whether it signals an error a person should look at. The flagged templates come back with example lines. An engineer still investigates and decides on any fix or rollback, and should sample unflagged templates now and then to confirm the question is not too narrow.
Example workflow
- Ship logs to a file or a store such as Grafana Loki
- Group lines into templates with a log parser
- Ask Jev a Noul question per template about whether it needs attention
- Post flagged templates with sample lines to Slack
- An on-call engineer investigates and decides next steps
Tools
- Jev (TypeSafe AI)
- Grafana Loki
- Slack
Risks and guardrails
- Rare but critical lines can hide inside a common template
- Log text may contain secrets, so scrub before sending
- Alert fatigue if the question is too broad
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Tocsin by Ateeq on Made with Jev: madewithjev.com/builds/tocsin
- Tocsin: original post or repo by Ateeq: github.com/TPAteeq/tocsin
- Slack MCP server (official page): docs.slack.dev/ai/slack-mcp-server (checked 2026-11-09)