Leviathan
Asking an agent a question about a million support tickets should not mean pasting a million rows into context. Leviathan is a single Rust binary that turns your records into a ranked full text index first, then lets the agent ask in plain words and get back short, cited result cards. It takes JSONL, JSON, CSV, TSV, SQLite, or anything a database CLI can export, and stores the index in one SQLite file. In the author's own test on one million records, the median answer cost 436 tokens, against more than 100,000 with grep. It plugs into agents by CLI, skill, or MCP. Version 0.3.0 shipped on 29 October 2026, and the repo now has 879 stars under Apache-2.0.
What you get
- A ranked full text index over your records, stored in a single SQLite file
- Short, cited result cards instead of whole rows pasted into context
- Plain-word questions from the agent, no SQL to write
- CLI, skill, or MCP integration with the agents you already run
The win is not the index. It is the token bill: 436 tokens for a cited answer instead of 100,000 rows in context. Try it on a copy of your own data first, and check the citations before you trust the cards.
Repo: https://github.com/elstongun/leviathan
Quick poll
What would you use Leviathan for first?
- Searching a million support tickets
- Cutting context costs on big log files
- Giving an agent cited answers over my own data
- Watching it for now
Vote in the comments.
If you found Leviathan interesting, I write about tools like this every week in The AI-Native Engineering Leader, what actually works for engineering leaders building with AI agents, and what to skip.
