Compare · A codebase index
For the team
What happens after the search finds the file?
An index points at the right lines, fast. The question after that — what the code does now, and what happens when it runs — takes the checkout itself. PRCHD’s agent reads it on a computer the team owns, and can run it.
A codebase index
PRCHD
Where an index is better
Speed, breadth, and a machine nobody has to run.
Each one is real, and each is a reason plenty of teams should keep the search box exactly where it is.
- It answers in milliseconds
For “where is this string used”, a search box has the answer while an agent is still opening files.
- It spans every repository at once
One query across the whole organisation. PRCHD reads one project per question, in that project’s own root.
- It is light
Thousands of queries an hour on modest hardware. An agent session is a whole process, on one computer, for one question.
- Nothing has to stay awake
A hosted index has no machine to keep on. A team host does: it answers while that computer is on, and picks up questions asked while it was off when it reconnects.
- Finding is not reading
“Show me every call site” is a search problem, and search solves it better than reading files one at a time.
A snapshot you search, or a checkout you can run.
| A codebase index | PRCHD | |
|---|---|---|
| What it holds | A copy of the code, as of the last crawl | No copy — it reads the checkout at the moment you ask |
| How current it is | As current as the last index run | The answer names the commit it was read from |
| Can it run the code | No | Yes — install, test, build, start the dev server, read the output |
| Can it open the app | No | Yes — a built-in server drives a browser already installed on that computer |
| Can it reach the tracker or the warehouse | Only what was indexed | Whatever the agent CLI on that computer is configured to reach |
| What it returns | Passages, ranked by similarity | An answer — and optionally a filed ticket or a branch to review |
| Where the code sits | Uploaded somewhere to be indexed | On the computer the team owns, where it already was |
What the agent can actually reach
Not a copy of the code. The code.
- The real working tree — files, imports, git log, and the state you asked about, checked out
- Commands and processes — “is this actually broken” settled by running it, rather than by argument
- A browser — open the page, click through it, read the console, screenshot what it sees. Off by default; enabled per project
- Whatever the team plugs in — the tracker, the crash reporter, the warehouse, through the agent’s own configuration
- An answer stamped with the project, commit, branch, and the mode it was read under
- Environment values sealed in the Keychain and write-only, so a run can reach a service without any surface reading the value back
Where PRCHD sits
Finding the file is the easy half.
An index is a good way to find code, and when the search box is faster for the question in front of you, use it. PRCHD is for the question after that one: what does it do now, is that still true, and what happens when it runs. Answering that means reading the checkout on a computer the team owns — and, when it matters, running it.
The repository is already on a computer you own.
Install the host on that computer and bind a channel to the project. The checkout is the index.