Use Cases

Pain

Claire wants to know whether the new onboarding flow truly improves activation, but with chat-based AI, conclusions are scattered across conversations while SQL, assumptions, filters, and analysis logic are hard to trace. When leadership asks where a conclusion came from, the analyst has to dig through chat history and SQL tabs instead of presenting a reviewable, versioned analysis process.

Solution (User Story)

1. 1

   Analyst and Agent co-author the Plan

   The analyst and Agent first turn the objective, data scope, success metrics, and methodology into a reviewable Plan so PMs, leaders, and analysts align before execution.

2. 2

   Review with annotations

   Claire can comment directly on the Plan, such as excluding internal test accounts or splitting new and returning users, with every revision preserved in document context instead of disappearing in chat bubbles.

3. 3

   Execute to produce Report

   Only after the Plan is confirmed does the team execute the analysis; Lantide locks the Executed Plan and produces a Report so conclusions, methods, and versions remain traceable.

Pain

The company already has a secure Azure endpoint and wants to adopt AI to improve analysis efficiency, but many cloud AI assistants still require sensitive customer data or transaction CSVs to be uploaded into an external environment. Data egress, control boundaries, and auditability remain major security and compliance concerns.

Solution (User Story)

1. 1

   Load data in local DuckDB

   Analysts assign the local folder containing sensitive data as the Workspace, and the data is immediately mapped into DuckDB tables with full physical isolation.

2. 2

   Disable or approve knowledge injection

   Any new business rules summarized by the model from conversations stay locked in the Queued Knowledge approval queue and are never injected into AI context without review.

3. 3

   Results stay in workspace

   With a SELECT-only read guard, AI can only read data and create temporary computation caches. It cannot write back to original data, reducing tampering, accidental deletion, and unauthorized modification risks.

Pain

Mia’s analysis is rarely a single SQL query. In the AI era, SQL is generated faster, longer, and in greater volume. Analysis moves faster, but cache freshness, query dependencies, rerun order, and lineage become harder to manage and trace.

Solution (User Story)

1. 1

   Build multi-tab SQL with cache

   Mia can split each query step into clear SQL tabs and cache intermediate results as reusable tables instead of recalculating from scratch every time.

2. 2

   Source Run recalculates via DAG

   When upstream data or a query changes, Source Run uses the DAG to determine dependency order, especially for recurring updates that continue to branch into deeper exploratory queries.

3. 3

   View lineage and source SQL

   Every result can be traced back to source SQL and upstream cache, helping Mia answer where a number came from and deliver with confidence.

Pain

Kevin wants to analyze whether ad campaigns are causing stockouts in specific regions, but the data lives across ad platform CSV exports, ERP databases, support systems, and local Excel files. Before any analysis begins, he spends too much time moving data, converting formats, and aligning fields.

Solution (User Story)

1. 1

   Load local files

   Kevin can load CSV, Excel, Parquet, and other local files directly into the workspace and quickly turn them into queryable tables.

2. 2

   ATTACH external databases

   Operational data in PostgreSQL or MySQL can be connected through DuckDB ATTACH without manual export and re-import.

3. 3

   Query with shared logical table names

   AI agents and analysts use the same logical table names and unified query layer, making cross-source JOINs natural while reducing data movement and schema confusion.

Pain

Sophie can get analysis results using SQL and notebooks, but client executives do not read query processes or chat history. They need a report with context, conclusions, charts, and recommendations. Previously, Sophie still had to manually move analysis into slides or documents, wasting time and risking copy-paste errors.

Solution (User Story)

1. 1

   Execute to produce Report

   After analysis is complete, Lantide produces a structured Report from the reviewed Plan, preserving methods, findings, evidence, and recommendations.

2. 2

   Generate HTML report

   Sophie can turn the Markdown Report into a standalone HTML report and use Quick Edit in the browser for frequent refinements, tuning the tone and layout to match executive or client preferences.

3. 3

   Share in browser or export

   Reports can be displayed, shared, or exported from the browser, turning analysis deliverables from engineering artifacts into business communication material.

Pain

Angela’s team previously discussed analysis requests in Slack or document comments, but feedback was scattered across tools. PMs added business context, analysts changed SQL, and leaders refined decision questions, yet nobody could tell which comments had been handled or which version became the formal execution basis.

Solution (User Story)

1. 1

   PM annotates Plan

   PMs and leaders can comment directly on specific Plan sections, such as segmentation logic, excluding special customers, or refining KPI definitions.

2. 2

   Analyst resolves and revises

   Analysts revise the Plan comment by comment and use Resolve Comments to turn feedback into the next analysis input, preventing discussion from being lost.

3. 3

   Execute after consensus

   The team executes formal analysis only after the Plan is confirmed, making every Execute based on a jointly reviewed version with traceable collaboration.