Architecture Click any block to open it Nekt Academy EN
How Nekt works

From raw data to action, in a single architecture.

01Sources
02Storage
03Context
04Consumption
→start the tourEscback to the mapFfull screen
01Sources

Connect the tools the company already uses.

Nekt copies the data from each tool into one place and keeps it up to date, with no spreadsheet in between.

  • 300
    More than 300 connectorsCRM, ads, payments, databases, spreadsheets and files.
  • You choose what to copyEach source becomes tables. Select only the ones you will use.
  • ↻
    Updates on its ownIncremental sync brings only what changed, on the schedule you set.
In the appSources → Add source
LakehouseTABLES0 tables
1Pick the sourcefrom the connector catalog
2Authenticateor send the setup link to whoever has access
3Select the tablesonly what you will use
4Automatic syncincremental, on the schedule you set
02Lakehouse and Catalog

Where the data lives, ready to query.

The Lakehouse stores the tables and files from every source in one place, in the same format. The Catalog shows what exists and what each thing means.

    Tip
    02Transformations

    Refine when the question becomes routine.

    SQL queries and Python notebooks turn raw data into tables ready for the team and for AI. You write them, or ask AI to write them.

    • </>
      QueryJoins, filters and aggregates. The result becomes a new table.
    • Py
      NotebookFor more complex logic: heavy cleanup, models, external calls.
    • AI
      AI builds it for youAsk in plain language through MCP, review and save.
    TipValue first, optimization later
    Yourequest via MCP
    query → sales_by_channel
    Result
    ✓ Saved to sales_by_channel · runs every day at 6 AM
    03Context

    What a new analyst needs to know.

    The semantic layer explains the business to AI: what each metric means, which rules apply and where each piece of data lives.

    • Descriptions in the CatalogWhat each table and column is.
    • Context documentsRules that span tables: active customer, funnel, revenue.
    • ✳
      All connectedConcepts point to tables, which point to sources.
    TipStart with the "About the company" document
    Semantic layer · workspaceread by AI via MCP
    concepttablesource
    Click any point on the map
    04MCP · AI and agents

    Your AI asks your data directly.

    MCP is the plug that connects ChatGPT, Claude, Gemini or Cursor to the workspace. AI reads the context, finds the right table and answers.

    • ↯
      In the AI you already useClaude and ChatGPT by login. Cursor and others by token.
    • Reads the context firstThat is how it gets customer, sale and revenue right.
    • ✦
      Not just readingIt also connects sources, creates queries, writes context and schedules routines.
    In the appMCP button → pick your AI
    mcp.nekt.com
    workspace
    What AI can also do through MCP
    Every action uses the permissions of whoever connected it. AI only sees what that person sees.
    04Data API

    Your systems pull the number, the same every time.

    The Data API serves any table through an HTTP request. Same input, same output, with low latency.

    • =
      DeterministicThe number does not depend on interpretation.
    • From any tableFrom raw data to fully modeled tables.
    • +
      Alongside MCPThe team asks through MCP, the product pulls through the API. Same table.
    ExampleUsage panel inside your product
    Request · your system
    GET /v1/tables/usage_by_account?account_id=c_5521Authorization: Bearer ••••••••
    No requests yet
    Response · Nekt
    {
      "account_id": "c_5521",
      "plan": "Pro",
      "usage_month": 1842,
      "limit": 5000
    }
    ✓ same result, every time
    Your product · customer screen
    Account usage this month
    0 / 5.000
    { }Data API
    • A system asks
    • Same input, same output
    • Low latency, predictable
    Apps, portals, automations.
    MCPMCP
    • A person asks
    • The agent picks the path
    • Flexible, for new questions
    Exploration and business questions.
    04Destinations

    Data goes back to where the team works.

    Destinations send Nekt tables back to CRM, ads, spreadsheets and messaging. It is the reverse path, from the Lakehouse to operations.

    • ↺
      Reverse ETLThe finished table syncs with the destination tool.
    • ◎
      In the right toolA task in the CRM, an audience in ads, a heads-up in Slack.
    • ↻
      ScheduledRuns with the pipeline. Nobody exports a spreadsheet.
    ExampleAt-risk accounts become CS tasks
    accounts_at_risk
    updated today at 06:10
    04Apps and BI

    Dashboards and apps on the same base.

    Dashboards the team checks every week, data apps for day-to-day work and reports that arrive on their own. All reading the same tables.

    • ▤
      DashboardsThe view you check every week, always up to date.
    • ▣
      Data appsInternal apps: sales playbook, CS hub, customer portal.
    • ✉
      Reports and alertsBy email or Slack, on schedule.
    TipOne table, many uses
    reading sales_by_channel
    Revenue this month
    R$ 0
    +12% vs. last month
    Orders
    0
    +8%
    Average order
    R$ 0
    +4%
    Revenue by week
    By channel
    0%

    Sales playbook · November

    Monthly target, pace per business day and every opportunity with owner and stage. Updates on its own.

    Weekly sales summaryFrom: Nekt · To: sales team · Monday, 08:00
    Good morning! Last week we sold R$ 318K, 9% above the previous week.
    • Instagram is still the top-selling channel: 40% of the total.
    • Average order went up to R$ 373.
    • 3 large opportunities close this week.
    EmailSlack⏰ every Monday, 08:00
    ↻Orchestration

    Runs on its own, and tells you when it breaks.

    Triggers start each step on schedule or when the previous one finishes. Every run leaves a log and an AI-written summary.

    • ⏰
      By time or by eventEvery day at 6 AM, or as soon as the source finishes.
    • ↻
      Incremental by defaultBrings only what changed and saves credits.
    • !
      Alerts with a diagnosisWhen something fails, you know why and what to do.
    In the appSettings → Alerts
    ⏰ by time⚡ by event05:55
    Following today’s runs…
    Configured alerts
    #data on Slackwhen a pipeline fails
    Email to the ownerwith the AI summary
    ✓Governance

    Every person, and every AI, sees only what it should.

    Permissions by layer and by table apply to the team and to agents. What a person cannot see, their AI cannot see either.

    • 3
      Viewer, Editor and ManagerView, edit or manage each resource.
    • ⚇
      By person or by groupIf there is more than one, the highest permission wins.
    • ◉
      Everything loggedEvery AI access goes into the log, with date and table.
    PlanGranular permissions: Growth and Custom
    Who accesses What they access
    Access log
    /
    click to jump