FAQ
Product Positioning
What is an analysis asset?
An analysis asset is the complete context needed to inspect, rerun, revise, and hand off an analysis. It includes not only the final Report, but also data sources, metric logic, the Plan, SQL evidence, validation steps, lineage, and Activity.
→ What is Lantide Data?How does another Agent continue the work?
The Agent reads the Workspace analysis-context manifest, follows artifact and history navigation, and then uses a scoped cursor for incremental changes. A private owner key lets the same connection profile resume work across client sessions and Workspace switches while preserving valid Plan approvals. Treat the key as a private credential; sharing it does not stop the original Agent.
→ Read the guideWhat if a local data file moves or is renamed?
Refresh Data to check source locations. Supported local tracking can recover identifiable same-volume moves; otherwise use Locate folder or Locate file. Unsaved SQL edits are preserved. Historical SQL stays unchanged and readers show current source mappings, which do not prove unchanged file contents. Cross-volume, removable, network, and cloud-synced locations are not guaranteed to recover automatically.
→ Read the guideWhat is new in Workspace exports?
Unencrypted Workspace exports use .lantide.zip with a readable README, artifact index, and available analysis-context snapshot. Imports restore data into main/ and additional/<alias> and rebuild context. Stable source IDs travel with the bundle; native tracking tokens do not. Missing physical files block direct Workspace export unless you explicitly allow an incomplete bundle with warnings.
→ Read the guideIs Lantide an analytics tool, an IDE, or an analysis asset platform?
The most accurate description today is a local-first AI analysis workspace. Its core value is preserving one-off analysis as reusable analysis assets. A broader platform is the long-term direction as these capabilities extend into team deployment and integration.
→ Product definitionCan I use Lantide Data if I don't know SQL or Python?
Yes. You can describe the analysis you need in plain language, and Lantide Data helps generate a Plan and SQL before execution. You review the direction and results, without writing code from scratch.
→ Quick StartIf general AI chat tools can analyze CSV files, why do I still need Lantide Data?
General chat usually produces a one-off answer. Lantide Data preserves the analysis as a reviewable Plan, verifiable SQL evidence, a Report, Source Run, and data lineage. Data and queries stay in the local workspace, so you can revisit definitions, understand dependencies, and continue the work later.
→ Why Analysis Conclusions Can't Stay in ChatI already use Claude Code or Codex. When do I need Lantide?
You do not have to replace your preferred Agent interface. When an analysis needs a formal Plan and evidence review, metric changes, historical investigation, lineage-based reruns, or a handoff, Claude Code or Codex can use Lantide through MCP Server while leaving the Plan, Report, SQL evidence, and Activity in a traceable workspace.
→ External AI Agents and MCP ServerSome cloud AI tools can connect to Google Drive or external databases. How is Lantide Data different?
The core difference is the analysis workspace and execution boundary. DuckDB reads, queries, and caches data on your device, while the Plan, SQL evidence, Report, and lineage stay together in the same workspace. Content sent to a model endpoint depends on the AI feature and settings you choose.
→ Where Lantide Sits in Your Data ArchitectureWhat mechanisms are included in Lantide Data’s Harness?
Lantide Data’s Harness includes state machines, domain tools, context engineering, IDE integration, and the local data plane. Together, these mechanisms control when AI can draft, query, execute, and deliver, turning analysis from a one-off answer into a reviewable, repeatable workflow.
→ AI AssistantExternal Agents and MCP
What is the difference between MCP Server and MCP Sources?
They run in opposite directions. MCP Server lets external Agents such as Claude Code and Codex connect to Lantide, use its local analysis environment, and leave reviewable artifacts. MCP Sources lets Lantide connect to external data or APIs and materialize the results as queryable tables.
→ Connect External AI Agents to Lantide SafelyCan Claude Code or Codex operate Lantide without an LLM API Key?
Yes, through the external Agent + MCP Server path, which does not require an additional LLM API Key in Lantide. The external Agent still operates under its own subscription and MCP support terms. Lantide's built-in Agent still requires your own LLM API Key or another configured model endpoint.
→ External AI Agents and MCP ServerSkills
Does Lantide support Skills, and can a Skill grant more Agent permissions?
Yes. You can import methods into the Skill Library, explicitly load them with “/” in built-in chat, and optionally share them with external Agents. A Skill provides guidance, not tool authorization: it cannot raise the MCP access mode or bypass Plan approval or audit.
→ Skills (Skill Library)Workflow
How is Plan → Approve → Execute different from asking AI to write a plan first?
In general chat tools, that is mostly a prompt-level agreement. Lantide Data makes it a product-level engineering boundary: during Plan, AI can draft but cannot start formal analysis; only humans can press Execute. The approved Plan is locked and recorded for review and traceability.
→ Plan → Annotate → Execute → ReportWhat is the difference between Quick and Project Analysis?
Quick Analysis is for ad hoc queries and lightweight SQL Q&A. Project Analysis is for formal work that needs a Plan, annotations, Execute, and Report. You can choose based on complexity instead of starting every question as a full project.
→ Quick Analysis vs Project AnalysisCan I rerun an analysis in the correct order after the data changes?
Yes. Lantide Data records data lineage and DAG dependencies between analysis steps. After data or a query changes, Source Run can rerun intermediate caches and downstream results in dependency order. The system provides a repeatable execution path, while people still review the final definitions and results.
→ Cache and Source Run: What Analysts Need to KnowWhen a departing analyst leaves behind a pile of Skills, Markdown, HTML, and SQL that even an AI Agent struggles to untangle, how is Lantide different?
In a general AI analysis environment, project context is often scattered across chat history, Skills, Markdown, HTML, notebooks, and multiple SQL versions. The next person has difficulty taking over, while Agents such as Claude Code and Codex must read many files and infer their relationships—consuming context tokens and increasing the risk of missed definitions or references to the wrong version. Lantide preserves the Plan, Report, SQL evidence, Source Run lineage, and Activity in a project structure that people can inspect directly and Agents can retrieve selectively through Lantide-specific resources and tools. This reduces unnecessary context loading and unsupported guesses. If upstream data was deleted or overwritten, reproducing the exact result still depends on whether the original data was retained.
→ .lantide Export, Import, and RebindingHow can I share analysis results with managers or clients?
Formal analysis produces a Report with methods, findings, evidence, and recommendations. You can generate standalone HTML or PDF, and use External Delivery to record where a Report was delivered. External Delivery stores Report-owned pointer metadata only; it does not upload, verify, synchronize, or delete external content.
→ Report to HTML: Weekly Meetings and TweaksWhat does a Workspace bundle include by default?
A Workspace export defaults to Balanced compression and includes local files from the initial data folder and Additional folders, MCP cache, and the Agent query ledger. Chat sessions remain optional. Remote database and cloud-folder content, recoverable passwords, and authorization caches are not included. Full backup is a separate export for one or more Workspaces plus an Application profile, with its own selections.
→ .lantide Export, Import, and RebindingWhen does a Workspace export require an extra size confirmation?
The extra confirmation appears only when deduplicated physical data is strictly greater than 200 MiB. The 10 MiB threshold controls progress display, not approval. All-files scope can include hidden files, source code, or credentials, so review the scope before sharing.
→ .lantide Export, Import, and RebindingWhat do Plan execution snapshots preserve?
Plan-local execution snapshots preserve the formal query and Source Run steps in full, even after the conversation ledger's latest 50 steps are trimmed from Agent context. They preserve execution history, not the source data itself, so exact reproduction still depends on retaining or reconnecting the required data.
→ Projects, Plans, and ReportsCan I go back to a Workspace bundle I already imported?
Yes, when the bundle has the same snapshot as the local workspace. Choose Restore …from this bundle and the shared content and analysis history return to what the bundle carries, without raising the version. The pre-restore state becomes the single recovery copy, so Restore Previous Version brings your edits back. Restore is never selected automatically, only bundled physical files are restored, and external files that were not packaged are left alone.
→ Workspace import and restoreMemory
Will knowledge be recorded randomly? How do we keep definitions consistent?
No. New suggestions first enter the current Workspace's Queued Knowledge. A user must explicitly apply each item to User or Project Knowledge before it can be injected into later AI context; pending items are not injected. Approval is not restricted to a manager role, so Workspace access and team policy determine who should review it.
→ Agent Memory vs Plan: Don't Mix Them UpCan I disable memory?
Yes. Agent Memory has a master switch and can be fully disabled for specific projects or compliance needs. When disabled, AI will not write new knowledge or inject it into context, which is useful for sensitive or tightly controlled analysis.
→ Agent Memory (Knowledge Base)Technical
What data formats are supported?
Lantide Data supports local CSV, Excel, and Parquet files, using DuckDB as the unified query layer. Common data files can be placed directly into the workspace and quickly mapped into queryable tables.
→ Welcome to Lantide DataMust I download data as files, or can I use company databases and external data services?
Files are not required. In addition to local CSV, Excel, and Parquet files, Lantide Data can query PostgreSQL, MySQL, SQLite, and other databases through DuckDB ATTACH, or use MCP Sources to materialize external data and API results as queryable, refreshable tables.
→ Data Browser and Data SourcesHow does AI work in a local-first product?
Data queries and computation run in the local workspace through DuckDB. When you use AI features, prompts, conversation content, data structure, field information, or other necessary context may be sent directly to the model endpoint you configured; the exact scope depends on the feature and settings you choose.
→ AI AssistantCan the AI assistant accidentally modify or delete my Excel, CSV, or database files?
Lantide keeps original data sources read-only: query tools do not modify, delete, or overwrite source Excel, CSV, or database data. Agents can still create caches and, when the selected mode and approval boundary allow it, create or update Workspace artifacts such as SQL tabs, Plans, Reports, references, knowledge proposals, and delivery pointers.
→ Backup, Security Boundaries, and TroubleshootingCan an Agent read or write across Workspaces?
An external Agent can read supported snapshot-safe artifacts in another Workspace on a per-call basis without switching its selected writer Workspace. This path does not allow live queries, schema or source access, GUI handoff, drafts, pending operations, or writes, and separate calls are not one transactional snapshot.
→ External AI Agents and MCP ServerCan SQL Server or Azure SQL join local files in one query?
Not in one transparent cross-source statement. SQL Server and Azure SQL use a staged ODBC path: run a single-source remote query first, materialize its result as a local cache, then join that cache to local files or other sources. PostgreSQL, MySQL, and SQLite use DuckDB ATTACH and follow a different path.
→ Data Browser and Data SourcesDo I need internet?
Local queries do not require internet, and data processing runs inside your workspace. AI features can call your configured LLM API; enterprise teams can also use self-hosted endpoints or Ollama to keep usage fully inside their own network.
→ AI AssistantWhat is BYOK?
BYOK means configuring your own LLM API Key for Lantide's built-in Agent, giving you control over model choice and usage costs. If you use an external Agent such as Claude Code or Codex through MCP Server, you do not need to configure an additional LLM API Key in Lantide; the external service's own plan still applies.
→ AI AssistantDownload
Is the full version free?
Yes. Lantide Data's full desktop version is available at no cost. Contact us if your organization needs a tailored enterprise deployment or integration.
→ DownloadWhich operating systems are supported now?
Lantide Data currently supports macOS 12+ and Windows 10+. You can choose the installer for your platform on the download page; future platform support will also be updated there first.
→ Installation & SetupHow do I get started after downloading?
After installing the app, create or select a local data workspace, add CSV, Excel, or Parquet files, and follow Quick Start to run your first analysis. We recommend starting with a small dataset to learn the Plan → Execute flow.
→ Quick StartWhy might an older Windows installation still show an update wizard?
When moving from 0.3.1 to 0.3.2, Windows may still show the last installer flow once during the updater transition. Follow the prompts for this update; later updates use the new path. UAC or SmartScreen prompts may still appear according to Windows policy.
→ UpdatesHow do automatic updates work?
The app checks about every 24 hours and offers Update and restart, Skip this version, or Later. Skip suppresses only that version and Later asks again after 24 hours; Help → Check for Updates still works at any time and can also show clearly labelled Preview builds. Before restarting, the app checks the main window and compare views for unsaved work and returns to the app by default. Preview builds are a public manual channel, not an internal-account program.
→ Updates