Use Cases

Why don't these two numbers match?

Connect ClariLayer to Claude Code, Cursor, or Codex. It bootstraps your real working context from the SQL you already have, then your AI data agent stops making the same data mistakes — wrong table, wrong join, refunds counted in revenue — session after session.

These are the moments an individual analyst hits in their own agent, against their own — often messy, ungoverned, cross-source — warehouse data. No data team, no platform account, no procurement. Just you, your data, and the agent you already work in.

The acute one is the first reconcile: “why don’t these two numbers match?” ClariLayer reconciles a saved definition against your real warehouse result and flags mismatches as caveats — checked against source, not blindly asserted. The team / governance scenarios are still here too, demoted to where this is going.

A worked example

“Net revenue” — two numbers, one caveat

Your dashboard reports $1.42M in net revenue for the quarter. The query your agent just wrote returns $1.51M. A $90K gap, and the QBR is in an hour. Which one is right?

You ask your agent to reconcile the saved net_revenue definition against the warehouse. It runs the stored SQL with its own access and reports the result shape back to ClariLayer, which compares the saved definition's declared signals — the columns, grouping, and aggregates its SQL implies — against the shape your agent actually got back. They don't line up: the live result is missing a column the saved definition expects, so reconcile records a caveat. That's the signal to look closer — not a ruling on which dollar figure is right.

The caveat is what sends your agent digging. With its own warehouse access it traces the $90K back to the cause: the ad-hoc query never joined the refunds table, so it reported gross, not net. ClariLayer didn't adjudicate the two numbers or diagnose the missing join — it flagged that the live result drifted from the saved shape and pointed you at it. You ship the refund-adjusted $1.42M, then remember the correction once — “net revenue must net refunds” — so next quarter the agent starts from it instead of rediscovering it. ClariLayer never held your warehouse credentials and never ran the SQL itself — your agent was the connector.

reconcile

Why don't these two numbers match?

The moment

Two reports that should agree don't. Your dashboard says one revenue figure, the ad-hoc query you just ran says another, and you have no idea which one to trust — or where the gap crept in.

With ClariLayer

Ask your agent to reconcile the saved definition against your real warehouse result. Your agent runs the stored SQL with its own access and reports back the result shape; ClariLayer compares what the definition declares against what actually came out. A mismatch surfaces as a caveat — so you and your agent know exactly which number to treat with care, and why.

What actually happens

  1. 1Your agent runs the stored SQL itself — ClariLayer never holds your warehouse credentials and never executes SQL server-side.
  2. 2Declared-vs-actual mismatch is flagged as a caveat; a clean pass leaves the entry asserted (the honest baseline).
  3. 3The caveat travels with the definition, so the next session starts from the discrepancy instead of rediscovering it.
See how reconcile works

recall + remember

New agent session, zero re-explaining

The moment

Every fresh chat starts from nothing. You re-type which table is canonical, which join is the right one, that refunds shouldn't count in revenue — the same context, session after session, before you can ask the real question.

With ClariLayer

ClariLayer installs as an MCP server into Claude Code, Cursor, or Codex. From then on your agent has an in-flow recall tool it can call to pull the right context mid-task, and a remember tool that saves each correction so it sticks. You stop re-explaining; the agent carries your context across sessions.

What actually happens

  1. 1Recall (the get_analysis_context verb) returns the most relevant saved context for the task, each entry tagged with its provenance and status.
  2. 2Remember saves a single definition, schema note, join path, or gotcha so it persists — default status asserted, provenance you.
  3. 3It rides inside the agent you already use — no destination app to open, no context to paste.
See recall and remember

bootstrap

Start from the SQL you already have

The moment

A blank context store is its own kind of friction — and hand-typing a CLAUDE.md of every definition is the documentation graveyard all over again. You want day-1 value, not a cold empty store you have to fill by hand.

With ClariLayer

Point ClariLayer at the work you already have. Bootstrap accepts your SQL files, your dbt models, and an existing CLAUDE.md. Your SQL is validated and deterministically structured — tables, joins, group-bys, time grain — while your dbt models and CLAUDE.md are imported and stored as schema-notes and notes. Day one, your agent already knows your world.

What actually happens

  1. 1Bootstrap accepts five sources today: SQL, dbt models, CLAUDE.md / freeform notes, a codebook / data dictionary, and a semantic-layer artifact (a Databricks Metric View or dbt semantic models).
  2. 2SQL, the dictionary, and the semantic model are structured (SQL is server-parsed; a codebook maps into one schema-note per variable; a semantic model maps into one canonical metric definition each); dbt and CLAUDE.md are stored as raw schema-notes and notes you can reconcile later.
  3. 3Your agent supplies the content — ClariLayer doesn't connect to your warehouse or filesystem directly.
See bootstrapping your context

compounds

Your agent stops making your data mistakes

The moment

It keeps reaching for the wrong table. It joins on the column that looks right but isn't. It counts refunds in revenue — the exact mistake you corrected last week, and the week before that.

With ClariLayer

Every correction you make, every reconcile, every note quietly persists. Your agent grounds on more of your context over time, so it gets progressively more right about your data — the wrong-table, wrong-join, refunds-in-revenue mistakes stop repeating. The longer you use it, the more it feels like Claude Code finally understands your project.

What actually happens

  1. 1Corrections accumulate through remember and recall — the context you build is the moat.
  2. 2Caveats from earlier reconciles keep steering the agent away from the definitions you've already flagged.
  3. 3More grounded every week: the value compounds the longer the context layer rides along.
See how context compounds

Reconciled, not blindly asserted

A hand-written CLAUDE.md of claimed definitions has the same trust problem as the original numbers — it’s just asserted text. ClariLayer is different: it reconciles a definition against your real warehouse result, and surfaces mismatches as caveats so you know what to trust. Today reconcile records a caveat on mismatch or leaves the entry asserted; a stronger verified stamp is the documented trajectory — it ships when our SQL-parser fast-follow lands.

For teams — where this is going

Yours today, your team’s tomorrow.

ClariLayer starts as your personal context layer. As teams adopt it, the personal context analysts build merges into shared, owned, governed team context — ownership, approval, the one right metric. These team scenarios are the Expansion path: still here, just not the headline.

Two VPs, two numbers, same metric

The problem

The board meeting stalls while Finance and Marketing debate whose revenue number is correct. Both pulled from the warehouse. Both are technically right. Neither is governed.

As personal context merges into shared team context, ClariLayer keeps canonical definitions with managed variants — both versions stay transparent, but only the approved, owned definition is marked the board-ready source of truth.

Explore For teams

Audit and approval trails

The problem

The auditor asks who approved the revenue-recognition metric and when. The answer is scattered across Slack threads, email chains, and someone's memory.

Once context is team-owned, every definition carries version history, an approval chain, and the conversation behind it — the 'why', not just the 'what' — so a change stays inspectable when it's deployed, deprecated, or rolled back.

Explore For teams

Recognize any of these moments?

If you already live in Claude Code, Cursor, or Codex against your own warehouse data, connect ClariLayer and let your agent stop re-explaining — and stop repeating your data mistakes.

Connect your AIRead the quickstart