Question
Mobile conversion · last week
What is Lantide Data?
Lantide Data is a local-first workspace where people and agents complete analysis together. The agent queries data and advances the work while keeping the Plan, Reference, SQL evidence, and review record visible for you to confirm, revise, and deliver. Through that collaboration, you and the agent create experiential context together.
Not a chat archive. Plans, decisions, SQL evidence, reviews, and Reports become the project.
You / analysis team
Ask, review, confirm, and decide.
Your agent
Converse, reason, and orchestrate tools.
Lantide Data
PlanReferenceStepsSQL snapshotReviewReport
Data sources
CSV, Excel, Parquet, database
Deliverables
Report, HTML, reusable analysis assets
- You / analysis team: Discuss, confirm, and iterate — Your agent
- Your agent: Advance and retrieve — Lantide Data
- Lantide Data: Query and validate — Data sources
- Data sources: Query and validate — Lantide Data
- Your agent: Discuss, confirm, and iterate — You / analysis team
- You / analysis team: Review and approve — Lantide Data
- Lantide Data: Produce and connect — Deliverables
What does an agent need to work like your team?
Data Context finds the data. Domain Knowledge supplies the definition. Experiential Context helps decide how your team would work through the problem.
AI Agent
Q
Why did retention drop?
Where experience comes from
Experience is created while the work happens.
Both sides do the same analysis. What changes is what happens to each step after the next one begins.
Same analysisAsk
Why did mobile conversion fall?
Without Lantide
Chat history
Earlier work fades into scattered traces
No earlier work yet
With Lantide
Question
Mobile conversion · last week
Lantide Workspace · preserved as artifacts
No earlier work yet
1 / 9
Ask
Why did mobile conversion fall?
Set the goal, time range, and metric.Discuss
Should guest checkout be included?
Keep it to match the existing KPI.Explore
iOS events changed after v8.2.
Flag a possible measurement break.Plan
Validate denominator, platform funnel, failures.
Make the method reviewable before execution.Prepare
Map paid_success to paid.
Record the legacy schema decision.Analyze
Run funnel, device, and failure queries.
Keep SQL, checks, and intermediate results.Review
Could the decline be a logging change?
Separate measurement impact from behavior.Refine
Stop Plan v1 and rerun the affected branch.
The new Plan keeps why the method changed.Deliver
Mobile Checkout Analysis
Both workflows reach the same conclusion.What survives?
The output survives. Most of the experience does not.
ReportSQLChat historyHuman memory
The work becomes reusable context.
Analysis ProjectGoalPlanDefinitionsReferenceExecutionSteps / SQL snapshotsReviewReport
Lantide preserves how the analysis evolved, not just how it ended.
What changes when experience survives?
Continue the reasoning, not just the files.
A teammate or another agent can understand what happened before continuing the work.
Why did we exclude Android v7?
I can infer that…
Checkout_complete is incomplete before v8.Validated against server payment logs
How long analysis keeps its context
Let your choices decide what deserves to stay in context.
In chat, context compression is a general solution from a general agent. Human + AI analysis already has review, execution, verification, and iteration checkpoints. At those moments, Lantide and the agent extract and save context according to your choices—so long-running analysis stays leaner, steadier, and easier to continue.
The same analysis
One conversation drives both context models
Confirm the metric and last week’s time range.
Without Lantide
+6k loadedAgent active context 6kInput 6k · Output 1k
Illustrative model—not a benchmarkStored in the active context as
Conversation turn
With Lantide
+5k loadedAgent active context 5kInput 5k · Output 2k
Illustrative model—not a benchmarkSaved outside the chat as
Goal
General compression guesses what mattered after the fact. Lantide preserves your choices while the work happens.
General compression guesses what mattered after the fact. Lantide preserves your choices while the work happens.
In Lantide, the context becomes an analysis asset.
Data ContextSources · Schema · Reference Docs
Analysis ContractGoal · Plan · Grain · Metrics · Validation
Execution EvidenceSQL · Checks · Results · Activity · Dependencies
Experiential ContextDecisions · Revisions · Stopped Plans · Review
DeliverablesReport · HTML · Next analysis
ProjectAnalysis Asset
InspectableContinuableReusable
Agents can change. Analysts can continue. The asset stays.
Product reality
What an analysis asset contains
Codex, Claude Code, and other MCP-compatible agents can use Lantide as the local analysis and review surface. The external agent handles reasoning and orchestration. Lantide keeps execution boundaries and reusable analysis artifacts visible.
Data context
Sources, schema, Reference Docs, and approved business definitions used by people and agents.
Analysis contract
The Plan records the question, grain, metric definitions, joins, and validation requirements.
Execution evidence
Each Plan keeps a complete local execution snapshot with SQL, intermediate checks, Activity, dependencies, and analysis lineage.
Portable workspace
A workspace bundle carries the project, while snapshot-safe reads can reuse evidence across workspaces.
Deliverables
Report and HTML outputs keep findings and limitations; External Delivery records a pointer to delivery outside Lantide.
Evaluation detail
Ways of working
The difference is not whether an Agent can produce an answer
The difference is how the work can be reviewed, retained, and continued after the answer is produced.
Primary unit of work
- Traditional analysis workflow
- SQL, notebooks, reports, and communication records
- General Agent without a dedicated analysis workflow
- Conversations, task files, and a final answer
- Lantide analysis workspace
- A project that keeps the Plan, SQL evidence, Report, and Activity together
Review before execution
- Traditional analysis workflow
- Depends on team process; review may begin with SQL or results
- General Agent without a dedicated analysis workflow
- Depends on prompts, files, and tool permissions
- Lantide analysis workspace
- Review and annotate the Plan before a person starts formal Execute
Evidence behind results
- Traditional analysis workflow
- A reviewer may need to reconstruct context from several tools
- General Agent without a dedicated analysis workflow
- Depends on which files, logs, and tool results the session retains
- Lantide analysis workspace
- Executed Plan, query evidence, checks, and Report remain in one workspace
Traceability
- Traditional analysis workflow
- Depends on Git, notebooks, BI history, and documentation practices
- General Agent without a dedicated analysis workflow
- Context may be split across sessions or project files
- Lantide analysis workspace
- Trace from a Report to formal evidence and stopped or replaced Plan history
Handoff and reuse
- Traditional analysis workflow
- Depends on documentation quality and the original analyst
- General Agent without a dedicated analysis workflow
- Depends on retained context and file structure
- Lantide analysis workspace
- People and Agents can reopen project artifacts and Reference Docs across sessions
Data and execution boundary
- Traditional analysis workflow
- Depends on the existing analytics stack
- General Agent without a dedicated analysis workflow
- Depends on the Agent environment and granted access
- Lantide analysis workspace
- Portable workspace bundles and Plan-local snapshots; cross-workspace access is snapshot-safe read only, not live query or write, with one writer
Best fit
- Traditional analysis workflow
- Highly customized, human-led analytical work
- General Agent without a dedicated analysis workflow
- Fast exploration and low-risk, one-off questions
- Lantide analysis workspace
- Analysis likely to be reviewed, questioned, rerun, or handed off
This compares typical working patterns, not every product or team. Lantide creates review and traceability checkpoints; it does not guarantee that AI analysis is correct.
01 / The method
From a question to a rerunnable analysis asset
Explore quickly, then preserve work worth keeping through Plan → review → Execute → Report.
Start with quick exploration, then move into a focused project when you need a reviewable Plan → Report workflow.
1
Quick Analysis
Data exploration and problem framing
2
Project Analysis
Focus a project for deep-dive analysis
01
Plan
Write an analysis plan: what to query, which tables to use, and how to validate
02
Review
Review comments stay attached to the source; the Agent updates the analysis plan
03
Execute
After human approval, the Agent may perform analysis while keeping an ongoing analysis log
04
Report generation and continued analysis
Put conclusions and evidence into a Report, link its external delivery location, export an HTML presentation, or continue into the next analysis cycle
When Lantide is useful
- Repeating a weekly CSV or Excel analysis with a new dated export.
- Moving a workspace bundle or safely reading a completed snapshot from another workspace.
- Reviewing AI-generated SQL and validation evidence before using a result.
- Recording an External Delivery pointer alongside the Plan's complete local execution snapshot.
When another tool may be a better fit
Lantide is not intended to replace Excel for workbook formatting, Power Query automation, or established BI dashboards. External Delivery records a pointer; Lantide does not upload or verify the delivery. Cross-workspace access is not a live query or write path and permits one writer. Complete reproduction still depends on retaining the original source data. Local-first also does not mean model requests are always offline: AI features may send necessary context to the model endpoint configured by the user.
Choose the next level of detail
Start with real scenarios, or inspect the workflow and engineering capabilities behind the product.