# What it can't see

> The languages LLMGRAPH reads, the plan's limits, and the blind spots an answer names instead of hiding.

## Languages

Python and TypeScript / JavaScript. Other languages are not read.

> Only Python and TypeScript/JavaScript files are read; in a repo with neither, answers will find nothing.

TypeScript interfaces and types are not tracked yet.

## Plan limits

| Limit | Value |
| --- | --- |
| Repositories | 3 |
| Code lines per repository | 50,000 |
| Devices | 2 |

Over a limit, your agent gets a fixed message that says which one. Remove a repository with `npx -y llmgraph-mcp repos --remove`.

## Blind spots

what_is_unknown lists these, with the reason for each:

- Computed and dynamic calls.
- Calls through values of unknown type.
- Code that registers itself, such as decorators and plugin registries.
- Files outside the project.
- Files that could not be read.
- Files with syntax errors (counted).

In TypeScript any value with a matching shape fits a type, so calls through those types stay maybe_more.