Search
TatsuCode has its own search engine. It finds text inside files, or files and folders by name, size, or date, in your project, in any folder, or across every drive at once, usually in a fraction of a second. You don't run it yourself: describe what you're looking for and the agent searches.
Searches that take longer keep running in the background while the agent carries on, and every search has a live card in the chat with a Stop button.
What You Can Ask
Find every TODO in the C# files
Where is the database connection string set?
Find DriveTrain.cs anywhere on my drives
List the PDFs in my Downloads folder changed this week
Which files on D: are bigger than 1 GB?
Find the folder called "Builds" under D:\Projects
The agent picks the right kind of search: text looks inside files, files finds files and folders by name, size, or date without opening them.
Where It Looks
| Where | What's searched |
|---|---|
| Your project (the default) | The whole project, following your .gitignore files, including nested ones and .git/info/exclude |
| Any folder | Folders or single files you name, anywhere on your machine |
| Every drive | All fixed drives at once (network and removable drives are left out) |
Searching every drive by name is quick once Windows has the folder lists cached. The first search after a restart can take a minute or more. A text search across every drive reads every text file, which takes minutes unless the file type is narrowed (for example, only *.cs files).
What It Skips, and Why
Search skips places that are almost never what you want, so results stay useful and fast:
- Build and dependency folders:
bin,obj,node_modules,.git,.svn,.hg,.vs, and.idea. To look inside one, ask for that folder by name. - Ignored files: files your
.gitignoreignores, when the project is searched. Outside the project,.gitignorefiles aren't followed unless you ask. - Unity's generated folders:
Library,Temp,Logs, and the others Unity rebuilds, in Unity projects anywhere in the tree. - Private and system folders outside your project: saved credentials, browser data, app sessions, and Windows' own folders, unless you ask to include them.
- Folder links: junctions and symbolic links aren't followed.
- For text searches only: binary files, files over 32 MB, and OneDrive "online-only" files (searching them would download your whole OneDrive). Name searches still list online-only files.
"Nothing found" means nothing found
Folders and files that couldn't be read (access denied, in use by another app) are counted, and the first one is named with the reason. A search that finds nothing and reports zero unreadable places really found nothing. When one file holds more than 200 matches, it's reported as "200+" instead of a total that looks exact.
Long Searches Keep Going
A search waits a little while for its results (15 seconds by default). If it isn't finished by then, the agent gets what was found so far and the search keeps running in the background. The agent is told when more results arrive, and when it has nothing else to do it waits for them instead of checking over and over.
- Up to three searches run at once.
- A finished search is kept for 30 minutes (the five newest), so the agent can page through its results.
- Searches end with the session: starting a new session or switching sessions stops them.
- The agent can stop a search as soon as it has what it needs, or ask for just the first few hits when one is enough (finding a single file by name, for example).
The Search Card
Every search shows a live card in the chat:
- While it runs: an animated scan, the hits and files found so far with the first few files, how many files were scanned, the elapsed time, and a Stop button.
- When it ends: how it ended, such as Done, Stopped at the limit (it found the number of hits it was asked for), Stopped by you, or Failed with the reason.
Press Escape while the agent is waiting on a search and that search stops too. A reopened session shows each card as it last was; a search that was still running then reads Continued in the background.
Options, for the Curious
You never have to name these, but they explain what the agent can do with a search:
| Option | What it does |
|---|---|
| File types and names | *.cs, .cs, src/**/*.ts, or an exact name such as Program.cs |
| Exclude | Skip names or patterns: Generated, *.min.js, docs/** |
| Whole word, match case | Exact-word and case-sensitive matching for text |
| Regular expressions | .NET regular expressions, including text that spans lines |
| Comments only | Only matches inside code comments |
| Ignore indentation | A multi-line search that matches however each line is indented |
| Context lines | Up to five lines before and after each match |
| Size and date | Files modified before or after a date, or within a size range |
| Include private folders | Search the private and system folders that are skipped outside your project |
With Claude Code
Claude Code uses its own Grep and Glob inside the project and TatsuCode's Search for everything beyond it: other folders, every drive, and searches by size or date. News about a background search reaches it between tool calls. See Claude Code.
Instead of Shell Scripts
When an agent reaches for a slow or fragile shell search (grep -r, findstr /s, Get-ChildItem -Recurse | Select-String, find, dir /s), TatsuCode points it to Search with a ready-made equivalent. Claude Code gets a short reminder instead. When the shell really is the right tool, for example for an option Search doesn't have, the agent can still run the command.
Next Steps
- Find & Replace — change text across many files, with a preview
- Tools Reference — everything else the agent can do
- File History — undo and recovery