Tools

Grepticon gives your agent four read-only tools, named after their command-line counterparts: ls, find, cat, and grep. Every result is plain text. You don't drive them yourself; your agent decides which to call, and understanding how they behave helps you reason about what it will do.

ls: see the layout

Lists what's in a directory, starting from the workspace root. It's how an agent orients itself: list the root to see the shape, then list into the area most related to the question.

find: locate files by name

Finds files by name or path pattern, like **/*.md. An agent reaches for find when it already knows something about a file's name or location.

cat: read a file

Reads a file's contents. An agent reads a whole file, or jumps to the lines a grep match points at.

grep: search contents

Searches file contents for a pattern. An agent searches stems and synonyms rather than the user's exact words, since the documents rarely use them, and starts broad to find the right file before reading it. A search that matches nothing is a signal to reword and try again.

How an agent navigates

The tools compose into a simple retrieval loop:

1. ls the root to learn the workspace's shape.
2. grep a term, or find a path, to locate the right file.
3. cat that file to read the answer.

Every step is scoped to what the agent's token is allowed to see. See Access control.

No semantic search

Grepticon has no semantic search and no embedding index. Retrieval is literal search over whole files. That's deliberate: there's no index to build, keep fresh, or drift out of date, and the agent reasons over complete documents instead of scattered chunks.

In an evaluation against a naive RAG baseline, this approach scored higher on answer accuracy. See the benchmark write-up.

Reference

For each tool's exact parameters, see the Read API reference.