Spark helps identify whether an AI answer contains a factual, safety, medical, legal, technical, or time-sensitive knowledge claim.
Spark is moving to an English product experience. New development and assisted reviews focus on English technical content. Existing accounts, quotas, and privacy choices are unchanged by this interface update. Language transition details
Independent knowledge trust layer for AI
Review English technical facts before publishing.
Select a technical claim in a draft, AI answer, or webpage. Spark looks for applicable source support and shows what has support, what needs review, and what remains unchecked.
Chrome and Edge update independently; each store shows its available version. Browser verification allows up to 10 uses per 5 hours without signing in.
Product
Review sources, support scope, and evidence gaps.
Compare individual factual claims with public sources before using them in technical content. Spark keeps supporting references, conflicts, and evidence gaps visible. Coverage is limited, and original references still need your review.
Spark uses KNE source candidates, evidence rules, applicability gates, and trust policy rules to return states such as source supported, source missing, needs review, conflicted, outdated, high risk, or not knowledge.
Users can submit sources, corrections, and challenges. These contributions enter review instead of directly changing the KNE core data.
Available clients
Review selected claims in Chrome or Edge.
Chrome and Edge share the same selected-text verification flow. New product development focuses on English technical content.
Chrome
Status: PublicSelect text and review the returned claims, sources, and evidence gaps in the side panel. Check each source before relying on a result.
Install for ChromeMicrosoft Edge
Status: PublicUse selected-text verification in Edge. Availability and updates follow the Edge Add-ons listing.
Install for EdgeWhy it matters
As AI becomes the interface to knowledge, trust needs its own infrastructure.
Agents, skills, and tools give AI hands and feet. Spark focuses on the knowledge layer beneath those actions: what should be trusted, what needs a source, and what must be reviewed before it affects human decisions.
Turn AI answers into reviewable claims with source and risk signals.
Flag safety, health, legal, and other sensitive claims before users rely on them.
Use reviewed human corrections and source submissions to grow a claim-evidence knowledge trust graph.
Spark is designed to work beside mainstream AI platforms instead of replacing them.
Workflow
A lightweight verification flow inside daily AI use.
The first product stage focuses on active selected-text verification. Users choose what to verify; Spark does not automatically scrape pages by default. Results are cautious trust states, not final truth certification.
SPARKVERIFY
AI output
Spark treats model output as claims.
3
ClaimIllustrative workflow, not a live result
A mebibyte (MiB) equals 1,048,576 bytes.
Basis: Quantity and unit to compare with the original referenceVerifyChallengeSubmit source
RiskIllustrative workflow, not a live result
Do not substitute MB for MiB without checking the notation.
Basis: Unit notation and applicable product documentationVerifyChallengeSubmit source
Source hintIllustrative workflow, not a live result
Use original standards and product documentation where applicable.
Basis: Original-source reviewVerifyChallengeSubmit source
KNE source layer
Verification is separated from generation.
AI answer -> Claim -> Source evidence -> Trust state -> Review queuesource_supportedtrust state
A claim has relevant, acceptable source support for its risk level.
Output: cautious verification result, not final truth certificationsource_missingtrust state
No candidate source in the current KNE reference corpus supports the claim.
Output: cautious verification result, not final truth certificationneeds_reviewtrust state
High-risk, conflicted, outdated, or insufficiently sourced claims require human review.
Output: cautious verification result, not final truth certificationTechnical content teams
Review key facts, versions, quantities, and source scope in English technical articles and documentation.
Verify answersEvidence gapsRisk warning
AI product builders
Inspect individual factual outputs and collect evidence gaps before deciding whether a scoped review pilot fits your workflow.
Source layerReview queueClaim history
Source contributors
Suggest public sources, corrections, and conflicts for review. Contributions do not automatically change KNE.
Source registryTrust policyHuman review
Start with a review task
Try selected-text review, then discuss a scoped deliverable.
- The Chrome and Edge extensions provide the existing self-service review flow.
- New assisted review enquiries focus on English public technical content. Describe the task before sharing source material.
- The browser extension is in free public beta. Assisted review pilots accept enquiries; scope, pricing, and delivery are agreed separately. Subscriptions and team products remain future options.
Company
KNE Spark is building an independent knowledge trust layer for the AI era.
The current focus is source-based review for English technical content. The browser beta remains available, while assisted review pilots validate complete deliverables. The old Spark Web Chat has been retired.
The long-term KNE vision is a claim-evidence knowledge trust system that protects real-world human knowledge from hallucination, pollution, weak sourcing, and commercial manipulation. Spark is the first user-facing wedge into that system.
Team
Founder-led product and engineering.
SZ
Seth Zhai
Founder and builder of Spark. Product, engineering, AI memory, KNE source systems, and knowledge verification.
seth@knespark.com