Example · Hurricane Marlow · every answer carries its labelExample · Hurricane Marlow
Three example answers about Hurricane Marlow, a fictional storm that hit the Southeast on August 6. What have we paid out on Hurricane Marlow claims: 4.3 million dollars, Certified by Maria, Claims, reused by everyone who asks. How many Marlow claims are still open: 1,175, Governed, built from the approved open_claims metric. How did Marlow change Southeast claims week by week: claims peaked at 905 in the week of August 10, 7.5 times a normal week of 121, Needs review, new SQL written by AI, with one click to make it a certified answer. The card then turns Certified, by Maria.
What have we paid out on Hurricane Marlow claims?Certified · by Maria, Claims
$4.3MAn expert verified it. Reused by everyone who asks.
How many Marlow claims are still open?Governed
1,175Built from your approved open_claims metric.
How did Marlow change Southeast claims week by week?Needs reviewCertified · by Maria
7.5× at the peakNew SQL written by AI. Make this a certified answer →Certified by Maria, Claims. Everyone gets it next time.
Connects to
Snowflake
Databricks
BigQuery
Redshift
PostgreSQL
DuckDB
and more
Runs in
AWS
Google Cloudcoming
Open-source core
The problem in every company today
One question. Three different answers.
Ask a simple business question: “What was the impact of Hurricane Marlow on our claims operation?” The dashboard, the analyst and the AI chatbot each give a different number. Nobody can say which is right.
What was the impact of Hurricane Marlow on our claims operation?
Hurricane Marlow cost you $4.6M in claims.
No label. No sources.
AI sounds sure, even when it’s wrong.
Chatbots answer everything with the same confidence. Nothing shows what was checked.
Every question becomes a ticket.
The data team rebuilds the same numbers instead of verifying them once.
Security blocks AI on real data.
Sensitive data can’t go to an outside AI, permissions aren’t respected, and nothing is audited.
One platform, from question to decision
From a question to a decision, with every answer labelled.
01 · Ask
Open source
Ask in plain language.
In the browser, Slack, Claude or Cursor. The story first, with its label.
A ZLens chat. Priya asks what the impact of Hurricane Marlow was on the claims operation. ZLens looks for certified answers, then uses the approved claims metrics, and answers with the story first: since it hit on August 6, Marlow has added 2,140 claims and 4.3 million dollars in payouts; settling is slower and old claims are piling up. A short table follows, with this week's change: Marlow claims open over 30 days, 405, up 43, or 12 percent; days to settle, 19, up 1 day; Marlow claims still open, 1,175, down 62, or 5 percent. The answer is labelled Governed: approved metrics plus the certified payouts. Priya starts typing a follow-up: which teams are behind on Marlow? That one is new, so AI will draft it and label it Needs review.
ZLensASKMarlow impactClaims
What was the impact of Hurricane Marlow on our claims operation?PS
Marlow added 2,140 claims and $4.3M in payouts. Settling is slower and old claims are piling up.MarlowNowThis weekWeekOpen 30+ daysOpen 30d+405+43 +12%Days to settleTo settle19 days+1 +6%Still openAll open1,175−62 −5%Governedapproved metrics + certified payouts
Which teams are behind on Marlow?Browser · Slack · Claude
02 · Apps
Open source
Dashboards where every number is labelled.
Every tile shows its label. ClickTap any number to ask why it moved.
The Claims Weekly App. One tile, paid claims in August by product and region, is labelled Certified. Southeast Home is the darkest cell. It shows 512 and is selected, with its menu open on Explain this: Southeast Home rose by 298 versus July, mostly Hurricane Marlow claims.
Research breaks a change down. Promote it to be certified.
A ZLens notebook, Days to settle. A SQL cell averages the days to settle a Southeast claim, week by week; its line chart holds near 11 days until Hurricane Marlow, then climbs to 19 in the week of September 7. A Research cell asks why settlement times slowed after Marlow: adjuster capacity added about 5 days, 4 of them in the Gulf team; reopened claims added about 2. It offers these as likely drivers, not proof, and offers to promote the analysis to a block and certify it.
ZLensNOTEBOOKDays to settle▶ Run all
SQLdf ▾Days to settle by week● 2ms
1SELECT date_trunc('week', settled_at), AVG(days)2FROM se_settlements GROUP BY 1
1SELECT date_trunc('week', settled_at),2 AVG(days) FROM se_settlements3GROUP BY 1
12 rows · 1.7msTableLine ▾
Sep 7 · 19 days
RESEARCHWhy did settlement times slow after Marlow?Why did settling slow after Marlow?
Adjusters+5d
↳ Gulf+4d
Reopened+2d
Likely drivers, not proof.Promote to a block → certify
04 · Share
Enterprise · coming soon
Labelled numbers, delivered.
Every Monday in Slack or email, each number with its label, on a schedule you set once.
The claims-leadership channel in Slack on Monday, September 14, at 8:00. The ZLens app posts Claims Weekly: Marlow claims still open, 1,175, down 62, Governed; Marlow claims open over 30 days, 405, up 43, Governed, which crossed the alert line of 400; Marlow claims paid so far, 4.3 million dollars, Certified by Maria. It was scheduled once in ZLens Share for the Claims Leadership group.
How ZLens worksone question, start to finishOpen source
From question to certified answer.
ZLens starts from your models: your dbt project and semantic layer, or your database's own tables and keys. It reuses what your experts already verified. When nothing fits, AI writes new SQL labelled Starts from your dbt project, or straight from your database, and reuses what experts verified. When nothing fits, AI writes SQL labelled Needs review.
ZLensStep 1 of 5 · Ask in plain language
Priya types a question in plain language, what have we paid out on Hurricane Marlow claims, from the browser, Slack or an AI assistant, and asks ZLens.
PSWhat have we paid out on Hurricane Marlow claims?Ask ZLens
Ask fromBrowserSlackClaude or Cursor
ZLens doesn’t start by writing SQL. First it looks for an answer your experts already certified.
PSWhat have we paid out on Hurricane Marlow claims?Ask ZLens
Animation: the question passes the Certified gate (none yet) and the Governed gate (no fit yet), and is answered at the Needs review gate with new SQL written by AI.
1 · CertifiedNone yet
Has an expert already verified this answer?Has an expert already verified this?
2 · GovernedNo fit yet
Can your approved metric definitions answer it?Can approved metrics answer it?
3 · Needs reviewAnswered here
New SQL, written by AI, and labelled.
The answer arrives labelled Needs review: 4.3 million dollars paid on Hurricane Marlow claims so far, written by AI and not yet checked, with its query, data and lineage shown, and a button to make it a certified answer.
Answer · ClaimsNeeds review
Claims paid · Hurricane Marlow · so far
$4.3M
Written by AI. Nobody has checked it yet.
Make this a certified answer
SELECT SUM(p.amount) AS claims_paidFROM claim_payments p JOIN claims cWHERE c.cat_event = 'MARLOW'
SELECT SUM(p.amount) AS claims_paidFROM claim_payments pJOIN claims c USING (claim_id)WHERE c.cat_event = 'MARLOW'AND p.paid_at >= DATE'2026-08-06'AND p.paid_at < DATE'2026-09-14'
Written by AI · SQL fingerprint 9f3a c21e
Tables read
harbor_claims.claim_paymentsharbor_claims.claims
In your warehouse, as models in your dbt project. Nothing is copied out.
Claims systemsource
claim_paymentsdbt model
claimsdbt model
This answerneeds review
Steward reviewEnterprise · coming soon· in open source you certify on your own machine
The review flow. Request R-142: Priya asks to make the Hurricane Marlow payouts answer certified. Draft: Dan Kim, the author, adds a test to the block claims_paid_marlow proving only claims tagged to Hurricane Marlow are counted, and the tests pass; the number stays 4.3 million dollars. Review: Maria Lopez, the Claims steward, a different person, approves Dan's change, and it is merged as a1f93c2. Published: the answer is certified by Maria for the Claims area, and everyone gets it next time.
Request · R-142PS
Make this a certified answerMake this certifiedPriya · Claims paid, MarlowPriya · ClaimsAssigned to Dan Kim
Draft · Dan KimDK
claims_paid_marlow+ Marlow test
Tests pass
Review · MariaML
Dan's changeSteward. Not the author.Not the author
ApproveChanges
Published
Certified · MariaCertified$4.3MEveryone gets it next time
$4.3M
Needs reviewCertified · by Maria, ClaimsWritten by AI. Nobody has checked it yet.Next time anyone asks, gate 1 answers.
Versioned and traceable. A certified answer is code: a DQL block with a solid Certified outline, block Claims paid, Marlow, domain claims, status certified, owner maria.lopez@harbor.example, a SQL query, and a test that the row count is above zero. Owned, tested, reviewed and versioned in git — then reused by everyone who asks. Beside it, the git history: Dan writes it, Maria approves it, never the same person. a1f93c2, merge claims_paid_marlow, approved by Maria Lopez, the steward; 7d41e0b, add test, only Marlow claims, by Dan Kim, the author. Lineage runs from the Claims source system through the dbt models to the certified block, and on to the App, Slack and Claude.
Certified · Maria Lopez, Claims
A certified answer is code: a DQL blockclaims_paid_marlow.dql
ZLens’ Domain health page for the Claims domain over the last 30 days: 64% of answers certified; 12 answers waiting for review; 2 answers flagged This looks wrong; 3 certified items due for review.
ZLensHEALTHDomain health · ClaimsLast 30 days
Answers certified64%
Waiting for review12
Flagged “This looks wrong”2
Due for review3
Counted from each answer’s label and fingerprints in the audit trail; questions and numbers are not kept.
Stewards see what’s asked most, certify it once, and everyone reuses it.
Alwaysinside your cloudEnterprise · coming soon
Runs in your cloud. Nothing calls home.
Your sign-in, your policies, your model and your warehouse, all inside your AWS account (Google Cloud coming). Follow one question through it: every step is checked, and every step is logged.Your sign-in, your policies, your model and your warehouse, all inside your AWS account (Google Cloud coming). Every step is checked and logged.
Architecture diagram. Your people ask from the browser, Slack or AI assistants. Everything else sits inside your AWS account (Google Cloud coming): 1, sign-in with your company’s single sign-on, or ZLens’ own sign-in with MFA, identifies Priya Shah in Claims; 2, the policy engine applies row and column rules, Southeast only with SSNs masked; 3, ZLens Enterprise, running as your service, finds or builds the answer; 4, the query runs in your warehouse, Snowflake, Databricks, BigQuery or Redshift, and nothing is copied out; 5, the AI model runs in your account in a region you approve, with no retention. Every step is written to a tamper-evident audit trail with no result values. DuckCode, the vendor, is outside, and nothing calls home.
Your cloud · AWS · Google Cloud coming
Your people
Browser
Slack
AI assistants
Sign-in
Your company sign-in, or ZLens’ own with MFA.Priya Shah · Claims
Policy
Row and column rules: Southeast only.SSN 521-38-6789***-**-6789
ZLens Enterprise
Answers, labels, reviews.Runs as your service
Your warehouse
Snowflake, Databricks, BigQuery or Redshift.Nothing copied out
AI model
In your account, a region you approve. No retention.
Audit trail
Every step, tamper-evident. No result values.7c2ea91f3b0d
12354
1Priya signs in with your SSOYour company’s single sign-on (SAML or OIDC), or ZLens’ own sign-in with MFA.2Policy limits what she can see3ZLens finds or builds the answer4The model drafts SQL, in your account5The query runs in your warehouse
DuckCodethe vendorNothing calls home.
PeopleBrowserSlackAI assistants
1
Your cloud · AWS · Google Cloud coming
Sign-in
Your company sign-in, or ZLens’ own with MFA.
Your company’s single sign-on (SAML or OIDC), or ZLens’ own sign-in with MFA.
Priya Shah · Claims
2
Policy
Row and column rules: Southeast only.
SSN ***-**-6789
3
ZLens Enterprise
Answers, labels, reviews.
Runs as your service
4
AI model
In your account, a region you approve. No retention.
5
Your warehouse
Snowflake, Databricks, BigQuery or Redshift.
Nothing copied out
Audit trail
Every step, tamper-evident. No result values.
7c2e — a91f — 3b0d
DuckCode, the vendor
Nothing calls home.
You never pay us for AI or compute — your model and servers run in your account, billed by your cloud.
Proofthe receipt behind a certified answer
Every certified answer comes with its receipt.
Who certified it, when, what was checked, and where the number comes from. Here is the one behind Maria’s $4.3M.
CertifiedReceipt · claims_paid_marlow
Claims paid · Hurricane Marlow · so far
$4.3Mas of Mon, Sep 14
Certified by
Maria Lopez, Claims steward
On
Fri, Sep 11, 2026 · merge a1f93c2
Author
Dan Kim, who added the test “only Marlow claims” 7d41e0b
Checks
Tests passed · row count, only Marlow claims
Metadata complete · owner, domain, description
Validated on the warehouse · ran on harbor_claims
Lineage
Claims systemsource→stg_payments + claimsdbt models→claims_paid_marlowcertified block→App · Slack · Claudewhere it is used
Reused by everyone who asks. Nobody rebuilds it.
Plansopen-source core · Apache-2.0
Start free on your own machine. Enterprise is coming for your team.
The open-source workbench is free for one person. ZLens Enterprise runs in your own cloud account, for everyone who builds, reviews and reads the numbers.
Open source
Free · Apache-2.0
For one person, on your own machine
Certified answers
Trust labels
Apps
Notebooks
Lineage from your tables to every answer, and through your dbt models if you use dbt
Git-native
AI assistants via MCP
Using dbt? ZLens reads your models, tests and lineage. Not using dbt? Start from your database.