jgrep (kyu1204)

# jgrep

**grep for what code *does*, not what it's called.**

```
jgrep "catches an error and silently ignores it" src/
```

[](https://www.npmjs.com/package/jevgrep)
[](LICENSE)
[](package.json)
[](https://docs.typesafe.ai)

*No index. No embeddings. No LLM round-trips. A whole `src/` tree in ~2 s for about a cent.*

<img src="docs/demo.gif" alt="jgrep demo: semantic search over src/ and a git diff" width="900">

</div>

---

## Why

| you want to find…                          | `grep` / `rg` | embeddings | an LLM | **jgrep** |
| ------------------------------------------ | :-----------: | :--------: | :----: | :-------: |
| an exact name or string                    | ✅ instant     | meh        | 🐢 $$  | use grep  |
| "code that swallows errors"                | ❌             | ❌ fuzzy    | ✅ slow | ✅ **2 s** |
| "endpoint with no auth check" *in my diff* | ❌             | ❌          | ✅ $$   | ✅ **¢**   |
| needs an index / vector DB                 | no            | yes        | no     | **no**    |

jgrep runs on [Jev](https://docs.typesafe.ai), a *System One* model: it never
generates text, it answers typed yes/no questions with calibrated
probabilities, in parallel, at $0.042 per million input tokens with output free.
jgrep packs 16 code chunks and 16 questions into one request and turns the
probabilities into `file:line` hits.

## Install

```bash
npm i -g jevgrep     # installs the `jgrep` command
jgrep init           # paste your TypeSafe key, pick where to keep it, done
```

`jgrep init` verifies the key against the API, stores it with `chmod 600`,
and optionally teaches Claude Code / Codex to use jgrep. Get a key at
[console.typesafe.ai](https://console.typesafe.ai).

<details>
<summary>Prefer not to run init?</summary>

```bash
export TYPESAFE_API_KEY=..