โ† Runtime Guard
Runtime Guard Trace
๐Ÿ” Early exploration

AI wrote the code.
Did it invent the packages?

Trace scans AI-generated code for hallucinated dependencies โ€” package names that don't exist, can't be installed, or were simply made up โ€” before they become a security or reliability problem.

trace ยท dependency scan ยท requirements.txt
# scanning 12 dependencies...

โœ“ requests==2.31.0 โ€” verified on PyPI
โœ“ fastapi==0.104.1 โ€” verified on PyPI
โœ“ pydantic==2.5.0 โ€” verified on PyPI
โš  openai-helpers==1.2.0 โ€” not found on PyPI
โœ— anthropic-utils==0.3.1 โ€” does not exist
โœ— llm-pipeline-tools==2.1.0 โ€” does not exist
โœ“ httpx==0.25.2 โ€” verified on PyPI

# scan complete
2 hallucinated packages detected
1 unverified package โ€” review recommended
The problem

AI agents are prolific. They're not always accurate.

When an AI agent writes code, it generates package names from training data โ€” and sometimes those names are plausible-sounding but fictional. The agent doesn't know the package doesn't exist. It just writes what seems right.

The risk is twofold. A hallucinated package name that gets registered by a malicious actor becomes a supply chain attack waiting to happen. A package that simply doesn't exist breaks your build in a way that's hard to diagnose.

Trace catches both before pip install or npm install runs.

requirements.txt โ€” agent generated
requests==2.31.0 Real
fastapi==0.104.1 Real
openai-helpers==1.2.0 Unverified
anthropic-utils==0.3.1 Hallucinated
llm-pipeline-tools==2.1.0 Hallucinated
What Trace checks

Focused on what AI gets wrong

Existence verification

Every package name is checked against PyPI, npm, and other registries. Packages that don't exist are flagged immediately, regardless of version.

Suspicious naming patterns

Packages with plausible but unverifiable names โ€” common in AI hallucinations โ€” are flagged for review rather than silently passed.

Multi-ecosystem support

Scans Python requirements files, package.json, and other common dependency formats. Where the agent writes code, Trace follows.

Early exploration

Is this a problem you're facing?

We're in early exploration for Trace. If AI-generated dependencies are causing friction in your workflow, we'd genuinely like to understand the problem better. Leave your email and we'll be in touch.

No commitment. We're listening, not selling.