A Diakryon product
SourceProof
Verify first, then speak.
Standing instructions that make your AI assistant check before it speaks. SourceProof interviews you for ten to fifteen minutes and builds a guardrail you install in Claude, ChatGPT or Gemini.
What changes
Your assistant checks before it speaks
- It says “I don’t know” instead of making up an answer that sounds right, the habit people call AI hallucination.
- It checks a quotation against its source, or marks it unverified.
- It labels each factual claim by how it knows it.
- It says “no verified data found” instead of inventing a statistic.
- When a document or web page carries instructions aimed at it, it tells you and does not follow them.
- It skips the flattery, the habit called sycophancy, and looks for the weak spot in your idea before it agrees with you.
Ten protections are built into every guardrail and can’t be removed. The rest comes from your answers.
Individual
$39
Your own guardrail for your own assistant. 2 builds.
Team
$149
One shared guardrail for an organization, plus a one-page Staff Quick Card and Rollout Notes. 5 builds.
Already have a key?
Sign in with your email and we send you a one-time link. There is no password.