This AI Turns Spreadsheets Into Apps

The Spreadsheet Apocalypse is Over

Spreadsheets were supposed to be where decisions happen. Instead, they’ve become graveyards for data—thousands of rows, dozens of tabs, cryptic columns like “Status,” “Owner,” and “Created_at,” all quietly swallowing lead notes, call outcomes, and support tickets that nobody will ever scroll far enough to see. CRMs don’t help much; they just wrap the same tables in heavier UI.

Most teams live in this state: data-rich, insight-poor. Sales logs every call, marketing tracks every click, support tags every ticket, yet when someone asks, “What actually changed last week?” the room goes silent while someone fights a VLOOKUP. Valuable context hides in free-text notes, Timestamps, and status fields that no one has time to filter, group, and chart by hand.

Manual analysis demands serious cognitive overhead. To answer “Which campaigns drove the most qualified leads in the last 30 days?” you mentally juggle date ranges, segments, and definitions of “qualified,” then stitch together pivot tables, filters, and half-broken charts. Every follow-up question—“Okay, but what about by region?”—means starting that mental gymnastics routine again.

That overhead turns basic questions into mini-projects. People screenshot raw tables into slides instead of interrogating the data itself. Dashboards, when they exist, freeze a few pre-baked views in time, while everything else rots in CSV exports and forgotten Google Sheets.

Now imagine skipping all of that and just talking to your data. Ask, “Compare last week’s call volume to the previous two weeks,” and an AI agent pulls from your Google Sheet or CRM, then returns an interactive chart plus summary metrics: 4,800 calls, 4,200 appointments, trend lines, and a short narrative explaining what changed.

This is the shift from data entry to data conversation. Instead of dumping rows into a system and hoping someone cares enough to analyze them, business users become active interrogators: “Show me top objections from prospects,” “Highlight accounts at churn risk,” “Surface outliers in response time.” The interface stops being a static table and becomes a responsive counterpart.

Tools like n8n paired with Thesis.dev push this even further. Your existing automations keep writing to spreadsheets, but an AI-powered, generative UI layer turns those dead grids into living dashboards that answer back.

Meet Your New AI Data Analyst

Meet your new AI data analyst: a three-part stack that behaves less like software and more like a colleague who actually reads your spreadsheets. n8n acts as the automation brain, Google Sheets holds the memory, and Thesis.dev provides the voice and face that turns answers into dashboards you can click, hover, and explore.

Inside n8n, an AI Agent for Analytics sits at the center of this stack. It receives raw rows from Sheets, understands natural-language prompts like “Compare last week’s call volume to the previous two weeks,” and decides what analysis to run. Then it returns a structured response describing charts, KPIs, and narrative insights that Thesis.dev can render instantly.

Instead of hand-building a schema or designing a chart library, you wire your existing data—voice agent logs, web-scraped leads, call stats—into the agent’s workflow. The agent can filter by date ranges, group by outcomes, and surface patterns such as “4,800 calls produced 4,200 appointments” without you touching SQL or pivot tables. You talk; it translates that into queries and visuals.

Thesis.dev acts as a generative UI layer that consumes the agent’s output and turns it into interactive components. One response can spawn: - A time-series chart of call volume - KPI cards for total calls, appointments, and conversion rate - A table of top objections - A short text summary with recommended next steps

Traditional BI tools like Tableau or Power BI demand upfront modeling, data warehouses, and specialists who babysit dashboards. You import CSVs, define dimensions, tweak joins, and then wait days or weeks for a new report. Every new stakeholder question becomes another ticket in the backlog.

This AI-driven stack flips that dynamic. Once wired, the same workflow answers dozens of new questions with zero extra modeling: “Show me performance by agent,” “Highlight days with drop-offs,” “Rank campaigns by booked calls.” You get a new dashboard in seconds, not a sprint later.

Speed becomes the primary feature. Non-technical teams can interrogate live operational data as fast as they can type, while n8n and Thesis.dev quietly handle the plumbing, layout, and Timestamps behind the scenes.

The 'Generative UI' Revolution

Generative UI is what happens when AI stops answering with paragraphs and starts answering with interfaces. Instead of a wall of text, you get charts, cards, tables, and full dashboards that materialize from a single prompt like “Compare last week’s call volume to the previous two weeks.”

Under the hood, the AI never draws pixels. It generates a structured description of what the screen should look like: “a bar chart of daily calls, x-axis is dates, y-axis is call count, highlight last 7 days, add two KPI cards for total calls and appointments, plus a short narrative summary.” That JSON-like schema goes to Thesis.dev, which renders the UI instantly in the browser.

Think of Thesis.dev as a real-time front-end engine wired to an LLM. n8n feeds it live data from Google Sheets or a CRM, the AI agent decides which visuals make sense, and Thesis.dev turns that into an interactive dashboard you can hover, filter, and restyle without touching code. No React components, no CSS variables, no build step.

The pitch is brutally simple: zero front-end work. You describe what you want in natural language, and the system spits out: - A layout - Visual components - Bound data and interactions

If you can say “show me top objections from prospects this month and trend them week over week,” you can ship a working analytics UI.

Analogy-wise, imagine describing a painting to a speedrunner artist who paints in real time while you talk. “Make the background dark, add a bright red line across the center, annotate the peaks.” Generative UI does that, except the “painting” is a live data cockpit driven by Timestamps, filters, and metrics.

For anyone wiring this up themselves, n8n handles the orchestration and AI prompts; Thesis.dev handles rendering. The official Tutorial: Build an AI workflow in n8n (Official n8n Docs) walks through how to get an AI agent producing the structured responses that generative UI layers like Thesis.dev expect.

Laying the Foundation in n8n

Before dashboards and charts, you need a canvas. In this build, that canvas is n8n, a visual automation platform where every piece of logic lives as a node on a flowchart. You start by signing up for a free n8n cloud account, then hitting “New workflow” to land in an empty editor with a single, lonely entry point waiting to be wired up.

Instead of a cron job or webhook, this system begins with conversation. n8n’s Chat Trigger node acts as the front door, catching every message that comes from the Thesis.dev interface and turning it into structured input for the workflow. Drop in “Chat Trigger” as your first node and you’ve effectively told n8n, “Whenever a user types a question, start this flow.”

From there, you add the brain. n8n ships with an AI Agent node that bundles a language model, memory, and tools into a single orchestrator. Search for “AI Agent,” add it as the next step, and n8n automatically links it to your Chat Trigger so every user query routes straight into the agent.

Initial setup focuses on picking a chat model and wiring credentials. In the AI Agent node, you choose a chat model provider (for example OpenAI-compatible) and create new credentials, pasting in your API key from Thesis.dev’s C1 middleware instead of a standard OpenAI key. You also update the base URL from api.openai.com/v1 to the Thesis.dev endpoint so all model calls pass through the generative UI layer.

Once those two nodes are in place, the skeleton of the app exists: user message → Chat Trigger → AI Agent → response. Every additional behavior—querying Google Sheets, filtering by date ranges, aggregating call stats, formatting outputs for dashboards—becomes a new node or branch off this spine. You’re not writing functions; you’re drawing logic.

Understanding that visual, node-based model is crucial. Each node represents a specific action, and connections define data flow, so you can literally see how a question like “Compare last week’s calls to the previous two weeks” travels through triggers, agents, data fetches, and back out to a rendered dashboard.

Wiring the AI's Brain to See

Think of this step as giving your AI glasses. Until you change where it sends its requests, your n8n AI Agent is just another chatbot wired to OpenAI’s default API, blind to the idea of layouts, charts, or dashboards.

Instead of pointing the agent at api.openai.com, you repoint it to Thesis.dev. In the AI Agent node, add a new chat model credential, pick OpenAI, and paste in your Thesis.dev API key. Then overwrite the Base URL field so requests hit Thesis.dev’s middleware endpoint, not OpenAI’s servers directly.

That single field swap is what unlocks generative UI. Thesis.dev sits in the middle as an API proxy: it intercepts the agent’s prompt, injects instructions about components, styling, and layout, then forwards a refined prompt to the underlying LLM. When the LLM replies, Thesis.dev reformats the response into a UI‑ready schema that its frontend can render as charts, cards, tables, or full dashboards.

Instead of a blob of text, the agent now returns structured JSON describing visuals. A query like “Compare last week’s call volume with the previous two weeks” turns into: - A line chart spec with dates on the x‑axis and call counts on the y‑axis - KPI cards for total calls and appointments (e.g., 4,800 calls, 4,200 appointments) - Narrative insights and recommended next steps

To wire this up, head to Thesis.dev and create a free account. After logging in, open your account menu, click API keys, and generate a new key (name it something like “n8n‑test” so you remember what it’s for). Copy that key and paste it into the API key field in your n8n OpenAI credentials.

Now fix the Base URL. By default, n8n fills it with `https://api.openai.com/v1`. Replace that value with the Thesis.dev endpoint provided in their docs or dashboard. From this point on, every AI Agent call routes through Thesis.dev’s middleware instead of talking to OpenAI directly.

One more critical piece: add a Simple Memory node after the AI Agent. Memory lets the agent remember earlier questions, filters, and entities so follow‑ups like “Now break that down by rep” or “Show only last month” still reference the same dataset and Timestamps. Without it, every question becomes a cold start, and your “conversation with the dashboard” turns into a series of isolated, amnesiac queries.

Giving Your Agent Access to Data

AI agents do not magically “see” your spreadsheet. They need Tools—explicit capabilities that let them read, write, and search external systems on demand. In n8n’s AI Agent, Tools function like a loadout: you decide which services the model can call, and n8n exposes them as safe, parameterized operations.

Start by opening your AI Agent node in n8n and jumping to the Tools tab. Add a new Tool and choose the Google Sheets node as the underlying action. This turns the Sheets integration you already know into something the agent can invoke mid-conversation, for example when a user asks, “Show me last week’s call volume.”

To authorize access, create or select Google Sheets credentials directly from that Tool configuration. n8n will prompt you to sign in with your Google account and approve scopes so it can read the spreadsheet that stores your raw call logs, lead data, or performance metrics. Once approved, the agent now has a secure, revocable window into that document—nothing more.

Next, point the Tool at the exact dataset you want the agent to use. Configure: - Spreadsheet ID or URL for your voice-agent log - Specific sheet name (for example, “Calls_2024”) - Range or mode (such as “Read all rows”)

You can also constrain columns—dates, caller IDs, outcomes, Timestamps—to avoid feeding irrelevant noise into the model. Those guardrails help the AI answer questions precisely and keep your dashboards snappy.

From the agent’s perspective, every natural-language question that touches data now routes through this Tool. Ask for “appointments scheduled in the last 7 days,” and the AI will call the Google Sheets node, pull just those rows, then hand structured summaries to Thesis.dev, which turns them into charts, cards, or tables automatically.

The same Tool pattern scales beyond Sheets. Swap in an Airtable node to query a base, or a PostgreSQL / MySQL node to hit a production SQL database with parameterized queries. For a deeper dive into wiring up these kinds of automations, How to Build AI Workflows with n8n – freeCodeCamp breaks down more advanced patterns that slot neatly into this generative UI stack.

Flipping the 'On' Switch

Power finally arrives when the workflow stops being a diagram and starts behaving like a live product. That happens in n8n with a few deceptively small switches that control how your AI agent talks, listens, and stays online for real users.

Start with responsiveness. Both the Chat Trigger and the AI Agent nodes need Streaming turned on. Without streaming, users stare at a blank UI until the model finishes its entire response; with streaming, Thesis.dev renders partial output in real time, so charts, cards, and explanations appear progressively as the model thinks.

In the Chat Trigger node, enable streaming so n8n forwards tokens as they arrive instead of buffering the whole reply. Then open the AI Agent node and toggle streaming there too, which instructs the underlying Thesis.dev endpoint to emit data incrementally. Together, these two settings turn your dashboard from “submit and wait” into something that feels like a live analyst typing back.

Interactivity also depends on reliability. Configure the workflow to Always-on mode in n8n Cloud or ensure your self-hosted instance stays running, so your public endpoint does not go dark overnight. Add basic error handling nodes if you expect messy data or API hiccups, especially when your Google Sheets grow past a few thousand rows.

Once streaming and stability are set, hit Activate on the workflow. n8n publishes it as a live endpoint and exposes a public URL—typically a webhook-style link that accepts POST requests from external apps.

Copy that generated webhook URL immediately. This address becomes the bridge between your n8n backend and the Thesis.dev front-end: paste it into Thesis.dev’s configuration so every user query in the generative UI routes straight into your AI agent and returns structured UI components in real time.

Your First Data Conversation

First real contact with your data happens at thesis.dev/n8n. You get a minimal front-end: one URL field, one chat box. Paste your n8n webhook URL into the field at the top, hit save, and your browser tab now talks directly to the workflow you just built.

Now type a prompt that actually stresses the system: “Compare last week’s call volume with the previous two weeks.” That single sentence becomes a full request pipeline: Thesis.dev sends it to n8n, your AI agent hits Google Sheets, pulls the right rows, and returns a structured answer instead of a blob of text.

When the response comes back, Thesis.dev doesn’t just print numbers. It renders a comparative chart of call counts across the three weeks, with hoverable points and labeled axes so you can see daily swings and week-over-week deltas at a glance.

Above or beside the chart, you’ll see KPI cards that summarize the story. In the demo, those cards include totals like 4,200 appointments and 4,800 calls, computed from the same live sheet that used to be just rows and Timestamps. Now it feels more like a dashboard from a SaaS analytics tool than a hacked-together spreadsheet view.

Scroll down and the UI adds narrative context. The agent explains whether call volume is trending up or down, flags any spikes or dips, and may suggest next steps such as increasing agent capacity on specific days or double-checking lead quality if appointments lag behind call volume.

From here, experiment aggressively. Ask: - “Show top-performing agents by completed appointments over the last 30 days.” - “Surface the most common call outcomes and their conversion rates.” - “Highlight days with abnormal drop-off between calls and booked appointments.”

Each new question reuses the same stack: n8n orchestrates, Google Sheets supplies data, Thesis.dev autogenerates the interface. You can jump from time-series charts to segmented bar graphs, from KPI cards to detailed tables, without ever touching a chart builder or writing a single line of frontend code.

From Demo to Dollar Signs

Suddenly this stops being a neat AI demo and starts looking like a product you can sell. Once you have an AI agent wired into Sheets and Thesis.dev, you’ve effectively built a reusable template: connect a new data source, tweak the prompt, hand a client a login, and charge a monthly fee for “AI-powered analytics.”

Automation agencies and solo freelancers can package this as an “AI Analytics Cockpit” on retainer. You handle data plumbing and prompt design; clients get a private URL where they can ask questions like “Show MTD revenue by channel” or “Which reps are slipping this week?” and watch dashboards materialize in real time.

A startup might use it as a live sales cockpit. Hook n8n into HubSpot or Pipedrive, sync to Google Sheets every 5 minutes, and let the founder ask, “Which outbound sequences produced demos over $5k ARR last week?” The system responds with a conversion funnel, rep leaderboard, and a short narrative on where deals stall.

Marketing agencies can sell a lead performance tracker per client. Pipe ad platform data, landing page events, and CRM outcomes into a unified sheet, then expose questions like: - “Which campaigns drive the cheapest SQLs by channel?” - “Show creative fatigue for our top three ad sets.” - “Compare this week’s CPL to the 4-week average.”

E-commerce brands get an operational overview without hiring a BI team. n8n ingests orders, refunds, inventory, and support tickets; the AI cockpit answers, “What products are driving 80% of returns?” or “Alert me if today’s AOV drops 20% below the 30-day median.” You deliver a board-ready dashboard instead of a CSV.

Value comes from compressing BI timelines from months to hours. Traditional dashboards require schema design, ETL pipelines, and front-end work; this stack uses generative UI to go from raw rows to interactive charts, cards, and Timestamps-driven reports in a single workflow. You can even pair it with community tools like n8nDash – An Open Source Dashboard for n8n for clients who want hybrid AI + static monitoring.

Pricing then becomes straightforward: setup fee for the initial cockpit, plus a recurring charge for maintenance, new questions, and additional data sources. You’re not selling “an n8n workflow” — you’re selling always-on, conversational analytics.

The Future is Composable AI

Composable AI is the pattern hiding in plain sight behind this n8n + Thesis.dev setup. Instead of praying for one model or one platform to do everything, you wire together small, specialized services: n8n for orchestration and logic, Google Sheets for storage, Thesis.dev for generative UI. Each piece does one job extremely well, then hands off to the next.

Monolithic “AI platforms” promise end‑to‑end magic, but they age fast and lock you into their way of working. Composable stacks behave more like modern web apps: you swap in better components without burning everything down. Today it’s n8n + Thesis.dev; tomorrow you might replace Sheets with a Postgres warehouse or add a vector database, without touching the UI layer.

This pattern quietly shifts who gets to build software. A solo operator with a free n8n cloud account, a Google Sheet, and a Thesis.dev API key can now ship an AI dashboard that used to require a full‑stack engineer, a data viz specialist, and a frontend team. You drag nodes, paste keys, and describe behavior in natural language instead of writing React, SQL, and chart configs by hand.

Non‑developers suddenly get a menu of capabilities instead of a wall of code. Want to go beyond dashboards? You can bolt on: - Lead‑scoring agents that trigger follow‑ups - QA copilots that summarize call Timestamps by rep - Client‑facing portals that expose curated slices of your data

All reuse the same composable backbone.

You already have the ingredients: messy spreadsheets, half‑used CRMs, orphaned reports no one opens. Clone the workflow, point it at your own data, and connect your Thesis.dev front end at thesis.dev/n8n. Start asking questions your rows and columns could never answer on their own, and ship the kind of AI‑powered tools your future competitors are quietly building right now.

Frequently Asked Questions

What is Thesis.dev?

Thesis.dev is a 'generative UI' platform. It takes structured output from an AI model and automatically renders it as interactive user interface components like charts, tables, and dashboards, eliminating the need for manual front-end coding.

Do I need coding skills to follow this tutorial?

No. This workflow is built using n8n's visual, node-based interface. You will configure nodes and provide text prompts to the AI, but no traditional programming is required.

What kind of data can this AI dashboard work with?

You can connect any structured data source that n8n supports, including Google Sheets, Airtable, CRMs like HubSpot, or any SQL database. The core requirement is data in a row-and-column format.

Is n8n free for this project?

Yes, n8n offers a free cloud tier that provides enough credits to build and test this entire workflow. For higher volume or production use cases, they also have paid plans and a self-hosted option.