Context, permissions and boundaries

What uses AI credits#

A longer answer uses more credits. When your AI allowance is used up, Nelix Agent is not called until it renews, and the lab tells you so. The stages keep working.

What Nelix Agent can see#

When you ask something, the lab assembles a request for that one question: the relevant files from that one project, and the relevant results. Nelix Agent sees that request and nothing else.

• In Design Chat it can also look things up in the project you have open: list, read and search its files, and read its runs and reports. It cannot open anything outside that project.
• It cannot see another project of yours unless you are asking about it.
• It cannot see anyone else's work.

Your files are yours#

Nelix Agent proposes. It does not write. A file reaches your project only when you Push it from Design Chat, Merge a repair you accepted, or accept a generated test.

Text in your project is treated as data#

A comment in your code that looks like an instruction to the AI is not followed. This protects you when you bring in code from elsewhere.

Checking its work#

Nelix Agent can be wrong. The lab is designed around that:

• Its numbers must come from measured results.
• A repair is run against your test, on a copy, before you can accept it. If you see it earlier, it is marked as not tested yet.
• Every design still goes through simulation, synthesis and the build.

Treat what it writes as a capable colleague's draft: usually right, always worth reading.