QuorX News
Articles
Research and how QuorX works. The same News is inside the QuorX app.Your own browser, for the agent
With the QuorX extension, the agent works in your everyday Chrome, Brave or Edge, signed in as you, only in the tabs you allow, and asks before it sends, deletes or buys. From 0.2.12 you can stop it from the toolbar, keep sites off limits and teach it a routine.
A smaller, cache-friendly agent prompt
0.2.7 reorders the agent's prompt so its stable parts are cached on every step, offers helpers only where they help, and counts how long a reply reads before it writes. The honest headline: the conversation is the cost, not the system prompt.
Agentic integrity in 0.2.7
QuorX is built so an agent's work is checked, not trusted. What 0.2.7 measured across models, what broke one release earlier, and what changed, in one page with links to the full articles.
Building a release in parallel
0.2.7 was built as five feature bundles, a stack of four workflow features and two early fixes, with a merge coordinator. What made it fast, where the pull requests collided, and what we'd do the same way again.
Deciders vs creativity
Strict deciders see only the user's words, so they turned down what a request only implies, like the voice-over in a 30-second movie. How a prompt enhancer, a 15-second wait, production plans and model cards fixed that without loosening the judge.
Four workflow features, for any model
Diff review, hooks, scheduled tasks and memory arrived in 0.2.7. Each had to work with a frontier model, a cheap one and one without tools, with the rules enforced by QuorX rather than by the prompt. What each does and the choices behind it.
Shipping safer
In 0.2.5 a bad version reached installed apps through auto-update. 0.2.7 puts tests on every pull request, stages each release as a pre-release with Promote and Pull, checks its checksums, and makes QuorX answer only the computer it runs on.
The screenshot that was always blank
A user shared a chat where the agent spent six minutes reloading a report that looked fine to them. Its screenshots after every scroll were bare background. How the chat's own log showed it, how the bug was reproduced in minutes, and the one-line cause fixed in 0.2.11.
When a security hardening broke the data
QuorX 0.2.5 locked down its data folder so tightly that QuorX itself could no longer read it. Projects and chats seemed to vanish; nothing was lost. What happened, how 0.2.6 repaired it an hour later, and what every release does now.
Agents route around guardrails
An agent once wrapped a whole-computer install inside a small Node script, and QuorX's approval card showed only "node". Here is how that happened, how QuorX now reads a script before it runs, and what that costs honest build scripts.
Installed is not working
A PC can have Rust and still be unable to build a Rust program. Here is what an agent did when it found that out halfway through a task, and what QuorX checks now before the agent starts.
Stop means stop
When Stop is pressed, everything the agent's command started should end, including programs that slipped away from their parent. How QuorX does that with Windows Job Objects and process groups, what the tripwire adds, and where the Running tab fits.
Testing an agent through its own UI
QuorX's UI test starts a separate copy of the desktop app, drives it the way a person would, answers the agent's approval cards, and checks the program the agent built from outside. What it covers, a real run step by step, and what it can't show.
The quorum experiment
QuorX began as one question. Can a cheap model do better by asking a quorum of fast decider models? How QuorX measures that on every question, what the first benchmarks found, what isn't known yet, and where any computer running QuorX shows the same numbers for itself.
When to give an agent a debugger
QuorX's agent can now set breakpoints and read variables in nine languages on Windows. It only gets that debugger after plain debugging has failed. Here is why, how the debug tool works, and how much of it is proven on a real PC.