How to Get Started with Count for Data Analysis

Count is an agentic analytics platform that combines a collaborative canvas, a built-in AI agent, and full code access in a single interface. Unlike traditional BI tools that produce static dashboards, Count lets teams work interactively with data using natural language, SQL, and Python without switching tools. The agent runs queries, builds charts, and explores sub-hypotheses on request, while every output stays editable so analysts can inspect and adjust the logic.

What Makes Count Different

Most analytics tools fall into one of two camps: no-code dashboards for business users, or developer notebooks for engineers. Count occupies the space between them. A founder can type a question in plain English and receive a chart; a data engineer can open the same canvas and edit the underlying SQL directly. Both users see the same output, and changes from either side update the shared view in real time.

The platform runs queries in three places: directly against connected data sources such as warehouses and databases, on Count's own servers for in-memory work, and in the browser for lightweight client-side computation. This architecture means Count can federate data across sources without replicating it into a proprietary store.

Count's AI agent is powered by models from Anthropic, OpenAI, and Google. The agent operates with full context of your data model, prior queries, and canvas history, which means follow-up questions work correctly rather than starting from scratch each time. The agent asks permission before accessing external data, and Count does not train models on your data.

Step 1: Create an Account and Set Up Your Workspace

Go to count.co and sign up for a free account. During onboarding, Count prompts you to name your workspace and invite teammates. You can skip the invite step and work alone initially.

Once inside, Count presents a blank canvas. The canvas is Count's core interface: an infinite, zoomable surface where you place queries, charts, notes, and agent outputs. Think of it as a collaborative whiteboard that understands data.

For first-time users, the fastest way to start is to drag a CSV file directly onto the canvas. Count reads the file, infers column types, and makes the data available to the agent immediately.

Step 2: Connect a Data Source

For ongoing work, connect Count to a database or warehouse rather than uploading files manually. Count supports the major SQL data sources including PostgreSQL, MySQL, BigQuery, Snowflake, Redshift, and Databricks, among others. Connections are added from the workspace settings panel under Data Sources.

To add a connection, provide the host, port, database name, and credentials. For cloud warehouses like BigQuery or Snowflake, Count accepts service account credentials or OAuth. Once connected, Count reads the schema and makes tables available in the canvas sidebar.

Count also connects to Google Sheets and can reference existing metric layers if your team uses a semantic layer tool. This means Count can sit on top of a dbt project or a Cube deployment and query pre-defined metrics rather than raw tables.

Step 3: Navigate the Count Canvas

The canvas has three main areas. The left sidebar shows your connected data sources, saved canvases, and shared catalogs. The main canvas surface is where all analysis happens. The right panel shows the agent's reasoning steps, which you can expand to inspect what queries it ran and why.

To start an analysis, click anywhere on the canvas and either type a natural language question in the agent input box, or add a SQL cell manually. SQL cells display their output as a table directly on the canvas. You can then add a chart cell linked to that table, or ask the agent to build one automatically.

You can zoom and pan the canvas to see multiple analyses at once, which makes it useful for presentations: the canvas itself becomes the deliverable rather than a separate slide deck.

Step 4: Ask the AI Agent Questions

The agent input sits at the top of the canvas. Type a question in plain English, such as "Show me monthly revenue by product category for the last 12 months" or "Which customer segment had the highest churn rate in Q1 2026?"

The agent responds by writing SQL against your connected source, running it, and placing the result on the canvas as an editable table. It then asks whether you want a visualization. If you say yes, it builds a chart and places it alongside the table.

One pattern that works well is chaining questions. After the agent answers the first question, ask a follow-up: "Break the top category down by region." The agent uses the context of the prior query to write the next one correctly without requiring you to restate the table names or filters.

For complex analyses, the agent's reasoning panel on the right is worth reading. It shows each step the agent took, including intermediate queries that were discarded. This matters for auditing: you can see exactly what the agent did before the result was presented.

Step 5: Build and Share Reports

Once you have a collection of charts and tables on the canvas, you can share the canvas directly with a link. Teammates who receive the link can view the live canvas, leave comments, and edit cells if they have write access.

For stakeholders who should not see raw data or SQL, Count supports a presentation mode that hides the code and shows only the visual outputs. This mode is useful for weekly business reviews where non-technical stakeholders need to see charts without navigating the underlying logic.

Count also supports scheduled refreshes. Set a canvas to refresh on a daily or weekly schedule, and Count re-runs all queries against the live data source so the charts stay current without manual intervention.

For teams that want to standardize analysis across projects, Count's Catalog feature lets you save reusable query blocks and metric definitions. Any canvas can reference a Catalog entry, which ensures that revenue, for example, is calculated consistently across all analyses in the workspace.

When to Use Count Over Other Analytics Tools

Count is well-suited for teams that need analysts and non-technical stakeholders to work in the same environment without exporting data between tools. The collaborative canvas eliminates the common pattern of an analyst building a report in a notebook, exporting it to slides, and then answering follow-up questions in a separate thread.

For pure no-code BI, tools like Metabase or Tableau remain stronger for organizations that need pixel-perfect dashboards with fine-grained permissions. For heavy notebook workflows with complex machine learning pipelines, Jupyter or Hex are more mature.

Count's specific advantage is speed of collaborative exploration: when a question comes in that requires ad-hoc analysis, a small team can open a canvas, connect to the warehouse, and have a shareable answer in the same interface within minutes. If you find the setup process too technical for your team, VSLZ handles the same kind of ad-hoc analysis from a file upload with no configuration needed.

Practical Summary

Count works best as an exploration layer on top of an existing data warehouse. The fastest path to value is to connect a Postgres or BigQuery instance, open a canvas, and ask the agent five to ten business questions about your data. The agent's ability to chain questions and maintain context across a session is what separates it from a standard text-to-SQL tool. Every output is editable, every query is visible, and the whole canvas is shareable, which makes it easier to move from raw analysis to a decision without passing files between applications.