Atlarix vs Cursor: We Had to Cut Our Own Best Argument
This post used to be easy to write. Cursor indexed your codebase into embeddings, Atlarix did not, and the comparison wrote itself. When we went back to Cursor's own documentation to source that claim properly, it was not there any more — and the honest version of this comparison turned out to be more interesting than the one we had drafted. Everything below is quoted from Cursor's docs, verified 18 September 2026.
We were about to publish something stale
Cursor's codebase-indexing documentation now redirects to a page titled Search, and that page says: "Cursor ships with Instant Grep, a custom search engine that outperforms ripgrep on large codebases. It runs automatically; no configuration needed."
The words "embedding" and "embeddings" do not appear anywhere on it. We checked the full page source rather than skimming. So the contrast we had built a draft around — vector search versus lexical search — is not something Cursor's current documentation supports, and we cut it.
What that leaves, which is narrower and truer
Both tools now describe fast lexical search as the primary way the agent finds code. That is convergence, and pretending otherwise to preserve a talking point would be exactly the behaviour this blog exists to correct.
One stated difference remains, from Cursor's own page: "File paths are encrypted before being sent to Cursor's servers. Code content is never stored in plaintext." Atlarix sends no file paths anywhere — search runs entirely on your machine, shelling out to bundled ripgrep, with no server involved in finding code.
Pull-request review
Cursor's reviewer is Bugbot: "Bugbot reviews pull requests and identifies bugs, security issues, and code quality problems." Billing is "usage-based", and the docs point to the pricing page "for current rates" rather than publishing a number — we looked, and there is no dollar figure on the Bugbot page to quote. The page also flags that Bugbot pricing changed in a May 2026 update, so older comparisons citing seat pricing are stale.
Atlarix Reviewer is included in Pro on the same balance as everything else, with no separate rate to track.
Whether either reviewer runs your code
We can only answer half of this. Cursor's Bugbot page contains no sentence about executing code, running tests, a sandbox, a container or a runner — we grepped the full source for all of them. That is an absence of documentation, and we will not convert it into a claim that Bugbot does not execute code.
What we can state is our side: Atlarix Reviewer's entire tool surface is three read-only operations, and the hosted path has no clone, no checkout and no runner. There is no build step to poison because there is no build step.
Browser work
Cursor's browser is documented as running "as a secure web view" controlled through an MCP server, opening "as a pane within Cursor." That is an in-editor capability rather than a browser extension, and for testing the app you are building it is a sensible shape.
Atlarix ships an actual Chrome extension on the Web Store, which works in the browser you are already signed into — different sessions, different cookies, different use case. If your task is debugging your own dev server, Cursor's pane is closer to the point.
The shape difference that actually matters
Cursor is an editor. Atlarix is a workstation that sits beside whatever editor you already use, which means it does not ask you to move and it does not compete with your existing setup.
That cuts both ways honestly: if you want the agent in the file you are editing, in the window you are editing it in, an editor is the right architecture and a separate app is friction. We chose the separate app because it lets us put a terminal, a browser and an approval queue around the agent without renegotiating an editor's extension API.
We set out to publish a contrast and found the contrast had expired. The remaining differences are real but smaller than our draft claimed: where search runs, how review is billed, and whether the agent lives inside your editor. Checking before publishing cost us a talking point, which is roughly the point of checking.