The problem
What is an LLM hallucination in software engineering?
In engineering work, a hallucination is an AI answer that looks authoritative but is wrong or unverifiable: an invented API, an unchecked assumption, a missing edge case, or a confident conclusion with no evidence trail.
The approach
How Undes checks an answer
Undes does not simply ask another model to agree. It runs a structured review process and records each step in the final artifact:
What you get
A reviewable artifact, not a single answer
Instead of a polished paragraph, your team gets a structured result showing what was checked, what evidence was used, which hypotheses were rejected, and what remains uncertain or unsafe to rely on.
npm install -g @undes.ai/cli
cd your-project
undes run "Check this AI answer about the auth flow for hallucinations and missing evidence."FAQ
Common questions
Can Undes guarantee that an AI answer is correct?
No. Undes does not guarantee correctness. It makes the trust boundary explicit by showing evidence, assumptions, critique, open checks, and a final verdict, so a human can judge the answer instead of trusting it blindly.
Does Undes replace code review?
No. It gives your reviewers a structured artifact to review faster, with the evidence and open checks already surfaced.
Does Undes send my source code to its own servers?
Undes is BYOK-first: it uses your own model provider keys. It does not sell bundled model tokens. You control which providers are used in your workflow.
Is Undes a chatbot?
No. Undes is a multi-agent review workflow around AI-generated engineering work, not a single-shot chat assistant.