Not Just Servers: SentinelX for Working on Repositories
The same toolkit that lets Claude or ChatGPT run your servers makes it a careful way to work on code: orienting, searching, editing, and git.
SentinelX is usually introduced as a way to connect your servers to Claude or ChatGPT, so they can check a service, read a log, or restart something. That's a big part of it.
But the same toolkit turns out to be a very good way to work on code. The repositories you care about usually live on a machine somewhere, and SentinelX gives Claude or ChatGPT careful, structured access to them.
Getting oriented in one step
The first thing anyone needs in an unfamiliar repository is a sense of its shape: what's here, how it's organized, where things live.
SentinelX has a dedicated operation for that. sentinel_project_snapshot gives Claude or ChatGPT a compact, deterministic overview of a project in a single call, instead of making it wander the tree one directory at a time. It starts from a map rather than from a guess.
Reading and searching
From there, your assistant can read the files it needs and search across the codebase for a function, a config key or an error message. These are proper operations, not shell commands stitched together, so the results come back clean and in a shape the model can actually use.
Changing code carefully
Editing is where a structured approach pays off most.
sentinel_edit makes precise, targeted changes to a file rather than rewriting it wholesale, which means a small fix stays a small fix. For larger changes, your assistant can apply a unified diff through sentinel_git, and it's checked before anything is written: a patch that doesn't apply cleanly changes nothing.
Git, built in
sentinel_git covers the everyday work around a repository:
- diff to see exactly what changed
- apply_patch to apply a set of changes in one go
- clone and fetch to bring code down
- push to publish a branch
Pushing gets extra care. If your assistant needs to force a push, it has to say which commit it expects the remote branch to be on. If someone else pushed in the meantime, the push is refused and nothing is overwritten. It's the safe version of a force push, and it's the only one on offer.
Credentials never pass through SentinelX. Git runs on your machine and uses whatever access that machine already has.
Picking up where you left off
Real work on code rarely fits in one conversation. sentinel_context lets your assistant save where things stand, what was done, what's left, and what it learned, and resume it later, even in a brand-new chat. The thread of a piece of work survives the end of the conversation it started in.
The same guardrails apply
Everything that makes SentinelX safe for servers applies to code too. The allowlist on each machine decides which paths your assistant can read and which it can change. Every read, edit and git operation is recorded, so you can always see what changed and why.
That combination, structured tools on top of strict control, is what makes it comfortable to hand real work on a real codebase to an AI.
Try it on a repo you know
Connect SentinelX to Claude or ChatGPT, point it at a project you already understand, and ask it to get oriented, then to find something and make a small change. Watching it work through structured tools rather than improvised commands is the clearest way to see what's different.
Details and plans at sentinelx.app.