Welcome to Lantide Data

Version: 0.3.0

Query CSV, Excel, Parquet, and other files on your machine with SQL—no database installation required.

Design note: This guide explains how to use the product. For a first try, start with Documentation hub (pick a role: Analyst / PM / Business & Operations / Platform Enablement (Data Engineering) introductions). To understand why it is designed this way, read Series guide and product positioning.

1. Welcome to Lantide Data

1.1 What Is Lantide Data

Lantide Data is a desktop application that lets you query data files on your computer with SQL—without installing database software or uploading data to a remote server. Everything runs locally on your machine.

Point Lantide Data at a folder of data files and you can write SQL, view results, and export reports as you would with a traditional database. The product includes AI Agent collaborative analysis that produces reviewable, annotatable Plan and Report documents—not just one-off chat answers.

1.2 Three Ways to Work

Lantide Data supports three complementary paths. Choose by task complexity and your preferred tools, or mix them:

Path A: Manual SQL

Load a local folder or database connection → write queries in the SQL editor → view results → export. Best when you already know the schema and need a quick check or one-off pull. Full steps: §2.8 and §5–§6.

Path B: Agent collaborative analysis

Describe the analysis goal in natural language → Agent explores data and drafts a Plan → you review and annotate revisions → press Execute → Agent runs SQL step by step → delivers a Report with numbers and limitations. Best for multi-step analysis, metric alignment, and formal delivery. First full walkthrough: §2.

For a simple SQL syntax question, the Agent answers directly without creating a project. If analysis grows complex mid-flight, the Agent may suggest upgrading from quick analysis to project mode.

Path C: External AI Agent collaboration (MCP Server)

If you already use Cursor, Codex, Claude, or another MCP-capable AI Agent, you can let it read your data environment, create analysis artifacts, and run permission-guarded operations through Lantide's local MCP Server—with results captured back in Lantide. You can still review the Plan, Report, query steps, approvals, Save History, audit trail, and the external Agent's actual activity inside Lantide. This is not another database or cloud service; the host serves only on your machine's loopback address. For full setup and usage, see §13.

1.3 Five Stages of Agent Analysis

Whether funnel, retention, or KPI monitoring, the Agent in project analysis mode roughly follows five stages. The table shows what you see and what you control—operational detail in §2; UI and buttons in §11, §12.

Stage	What you see	Your control
1 Intake & mode	Agent decides whether this is a simple SQL question or a data analysis; if no project is focused, it may ask quick vs. project	Choose mode; add goals, denominator, time range, or success metrics
2 Explore & plan	Agent inspects schema and columns first, then drafts a Plan: goals, denominator, grain, join keys, and scenario-specific validation checkpoints	Review the Plan draft; no SQL executed yet
3 Review & revise plan	Annotate the Plan or revise in chat; the Agent patches the cited sections incrementally rather than rewriting the whole document	Revise metrics and steps locally; the Agent will not start fetching data until you are satisfied and press Approve & Execute
4 Execute SQL	Plan enters Executing; the Agent reads the Plan, then asks the execution mode (foreground / hybrid); it builds a Todo and runs SQL step by step; Pipeline stages in the Plan open SQL tabs	Choose execution mode (foreground / hybrid); Continue to resume after an interrupt; overly long or complex SQL may be split into smaller steps
5 Report & wrap-up	Agent produces a Report: concrete numbers, scenario essentials (e.g. funnel stage definitions and max drop-off), and limitations	Annotate and revise the Report; export HTML in one click; open a new Plan for follow-up if needed

1.4 Capability Map (by Topic)

Main capabilities grouped by theme below—each is a summary; follow the links for detail.

Data & queries — DuckDB out of the box; supports CSV, Parquet, Excel, and other formats, plus ATTACH to mount PostgreSQL / MySQL / SQLite for cross-source JOIN; built-in SQL editor with syntax highlighting; export results as CSV, Parquet, etc. See §3–§6, §9.
Cache & lineage — Query results can be cached as intermediate tables for later reuse; multi-step analysis can keep SQL tabs and rerun upstream chains with Source Run, with a lineage graph visualizing dependencies. See §7, §8.
Projects & delivery — Edit Plan / Report (Visual or Markdown), Reference docs (large reference corpus, see §11.3.1), Plan state machine, immersive annotations, project focus mode; Report to HTML in one click, tables copy to Excel; Compare view opens Plan / Report / persistent SQL in a standalone side window for read-only comparison. .lantide portability (project / workspace / Application / Full backup) and which export to use when: see §11.1.0; operational detail in §11.1. Also §10, §11.10.
AI copilot — Quick and project analysis in two tracks; the Agent can create SQL tabs, run queries, and open Plan / Report; status indicator, conversation fork, file change bar with per-message undo, background execution, and in-generation guards (see §12.2, §12.14). See §12.
Skills — Reusable analysis methodology and delivery rules (Skill Library); load explicitly with / in chat; the same definition can be shared with the built-in Agent and MCP external Agents. See §14.
External AI Agent integration — Use Lantide as a local MCP Server for external Agents such as Cursor, Codex, and Claude; choose a one-off Quick connection or a Persistent connection that survives restarts, while retaining Plan / Report / evidence, Activity, Save History, and approval records. See §13.
Analysis quality — Built-in scenario methodology (funnel, A/B, retention, etc.); Plan includes validation checkpoints, Report includes concrete numbers and limitations; plus 11 statistical / ML tools (t-test, regression, clustering, etc.). See §12.9, §12.10.
Governance & preferences — Agent Memory (injected after Queued Knowledge approval), dark / light theme, memory monitoring dashboard. See §15–§17.

1.5 Supported Local Data Formats

Format	Extension	Description
CSV	`.csv`	Comma-separated values
TSV	`.tsv`	Tab-separated values
Plain text	`.txt`	Text file (read with delimiter)
JSON	`.json`	JSON format
JSON Lines	`.jsonl`	One JSON object per line
Parquet	`.parquet`	Columnar storage
Excel	`.xls` / `.xlsx`	Excel workbook (sheet required)

1.6 Main Screen Tour

[Image] Main application screenshot labeling: left sidebar (Workspace, Data, Projects, Cached), center editor (SQL / Markdown tabs), results panel below, right AI panel (Astroid icon).