# CLI

```sh
ev-grep 'Performs database operations' src/
```

The first argument is one query. Remaining arguments are paths; without paths, ev-grep searches the current directory.

## Select files

```sh
ev-grep 'Performs database operations' src/ --glob '*.rs'
ev-grep 'Performs database operations' src/ --glob '!**/generated/**'
```

Directory searches respect ignore files and skip hidden files. Explicit file paths bypass ignore files, but glob filters still apply. Globs use gitignore syntax; later matches take precedence. They cannot re-include files inside an ignored directory. Overlapping paths are evaluated once.

## Search part of a file

```sh
ev-grep 'Catches a database failure and returns an empty result' src/users.py:20-56
ev-grep 'Writes to the database' src/users.py:42
```

Add `:N` or `:N-M` to a file path to assess only those lines. Lines count from 1, and the range includes both ends. Jev still receives the whole file as context, such as imports and helpers defined elsewhere in it, and judges what the selected lines do.

Ranges apply to regular files. A range that starts at 0, ends before it starts, or goes past the last line is an error; ev-grep does not shorten it. Each file and range is assessed once, and different ranges of one file are assessed separately. If a file's real name ends in something like `:12`, that file is searched whole.

Results print the range after the path, as in `src/users.py:20-56`. That is the same form ev-grep accepts, so results can feed another command.

## Read a query from a file

```sh
ev-grep -f query.txt src/
ev-grep -f - src/ < query.txt
```

The whole query file is one query, even if it spans several lines. With `-f`, all positional arguments are paths. Stdin is reserved for the query; source files are read from disk.

## Preview a search

```sh
ev-grep 'Performs database operations' src/ --dry-run
```

A dry run lists selected text files and line ranges with the size of each file. It makes no API requests and needs no key.

## Limit concurrency

```sh
ev-grep 'Catches a database failure and returns an empty result' src/ --jobs 2
```

Up to four requests run at once by default. `--jobs` accepts 1 to 256. Lower it when sharing a provider account with other tools. More jobs do not reduce the number of requests.

## Options

```text
ev-grep [OPTIONS] <QUERY> [PATHS]...
ev-grep [OPTIONS] --query-file <FILE> [PATHS]...
```

| Option | Meaning |
| --- | --- |
| `-f, --query-file FILE` | Read one query from a file; `-` reads stdin |
| `-g, --glob GLOB` | Filter paths; repeat to add filters, prefix with `!` to exclude |
| `--provider PROVIDER` | `openrouter` (default) or `typesafe` |
| `--model MODEL` | A supported pinned model for the selected provider |
| `-j, --jobs N` | Maximum concurrent requests (default 4) |
| `--min-confidence N` | Route answers below N to uncertain; 0 to 1 (default 0.8) |
| `--json` | Write versioned JSON Lines to stdout |
| `--sarif` | Write one SARIF 2.1.0 log to stdout for code scanning tools; cannot be combined with `--json` |
| `--dry-run` | List files without credentials or API requests |
| `-h, --help` | Show usage |
| `-V, --version` | Show the executable version |

Empty queries fail. Directory symlinks are not followed; explicit symlinks and special files are errors.
Paths and text must be valid UTF-8. Files containing NUL bytes are skipped as binary. Other invalid UTF-8 inputs are errors.

See [Output](./output.md) for exit codes, JSONL, and SARIF, and [Request limits](./limits.md) for size limits.