An AI that studies public UAP evidence and answers with provenance, trust tier and uncertainty — it does not host the archive.
Every answer the AI gives is the end of a repeatable chain. Nothing skips a step, and every item carries a record of where it came from and how it was handled.
We pull from publicly available material — government releases, official archives, witness and whistleblower accounts, imagery and video. For each item we record where it came from before anything else happens.
We OCR documents and transcribe video, scoring the quality of every extraction. Low-quality items are quarantined — never fed to the model — so garbled text can't leak into what the AI learns.
Each accepted item is assigned an evidence type and a T1–T4 trust tier by source reliability, so the model always knows how much weight a source deserves.
Clean, tiered evidence is organised into a searchable knowledge base — kept per tier so reliability stays attached to every piece of knowledge.
The AI model studies the clean corpus — only the material that passed extraction, review and tiering. It learns from the evidence, not from the noise around it.
You ask a question in plain language. It responds with citations, trust-tier labels and stated uncertainty — and links you to the official originals rather than re-hosting them.
UAP evidence is a mix of hard fact, contested testimony and open questions. Instead of blending it all together, UAP Intel keeps the categories separate and says which is which. We separate what is known, what is claimed, what is inferred, and what is unknown:
We state uncertainty rather than hide it, and we draw no conclusions for you — we surface the evidence and its reliability so you decide. Every claim carries provenance back to where it came from.
Three disciplines protect the integrity of what the AI learns and what it tells you.
Bad OCR and low-quality transcriptions are quarantined and never trained on. The model only ever learns from material that scored clean — noise stays out of the knowledge base.
Items pass through human review before acceptance. Nothing enters the corpus purely on an automated pass — a person gates what the AI is allowed to learn from.
Every answer is tied to a specific knowledge and model version, so results are reproducible. If a version turns out to be flawed, it can be frozen and rebuilt without contaminating past or future answers.