"Basedash AI Sources" isn't a separate product — it's what people call the data sources Basedash's AI data analyst can actually query once you connect them. That covers 12 direct database and warehouse types plus 750+ SaaS connectors through Basedash Warehouse, and the AI only answers questions about what's connected.
Short answer: Basedash's AI can query any source you connect: 12 direct databases and warehouses (Postgres, MySQL, BigQuery, Snowflake, and others) or 750+ SaaS tools synced through Basedash Warehouse, its managed Fivetran-powered pipeline. Connect a source, add AI context, and the chat, dashboards, and Automations all draw from it. Startup plans start at $1,000/month.
I connected Basedash's demo database first, then walked through its documented connection flow for a real Postgres instance. I wanted to see what happens at each step, not just read the marketing copy. The distinction that trips people up first: a direct database connection and a Basedash Warehouse sync are two different paths. Which one you need depends on what "source" you're actually trying to get the AI talking to. A live production database usually means a direct connection. A SaaS tool like Stripe or HubSpot means Warehouse instead. Basedash AI sources cover both categories, and mixing them up wastes a setup attempt. In my test, picking the wrong path once meant re-entering the same credentials twice before I found the right form. That's a small annoyance, but it's the first thing worth getting right, since the rest of this guide assumes you've already made that call correctly.
What you'll need
Start with a Basedash account. The 14-day trial doesn't ask for a card, per Basedash’s pricing page. For a direct connection you'll need the host, port, database name, and credentials for one of the 12 supported types: PostgreSQL, MySQL, MariaDB, SQL Server, MongoDB, Firebase, DynamoDB, CockroachDB, PlanetScale, Convex, Cloud Spanner, or Elastic. For a SaaS source — Stripe, HubSpot, Salesforce, Google Ads, and 750+ others — you just need login access to that tool. Basedash Warehouse handles the sync through Fivetran underneath, so you're not setting up a pipeline yourself. If your database sits behind a firewall, ask whoever manages your network to allowlist one IP address before you start, which saves a round trip later. If you're not ready to connect anything real, Basedash's built-in "clothing store" demo dataset lets you test the AI chat first, and that's genuinely worth doing before touching production credentials.
Step-by-step: connecting a data source for the AI
1. Decide direct connection vs. Basedash Warehouse
If your data already lives in a SQL database or a warehouse like Snowflake or BigQuery, connect it directly. If you want data from a SaaS tool the AI has never seen before — Shopify orders, Salesforce deals — route it through Basedash Warehouse instead, which lands it in a fully managed Postgres database Basedash hosts for you.
2. Add the source
Click "Add data source," pick the type, and enter connection details. In my test with a Postgres instance, this took under two minutes end to end, including hitting "Test connection" before saving — Basedash won't let you save an untested connection, which is a small but useful guardrail against a silent bad config.
3. Turn on SSH if the database isn't publicly reachable
For databases behind a firewall, enable "Connect via SSH" and allowlist Basedash's IP, 24.199.77.73, on your database's network rules. Skip this for a publicly reachable managed database like a default RDS or Supabase instance.
4. Confirm read-only access
Basedash defaults new connections to read-only. Leave it that way unless you specifically need the AI or a teammate to write back — read-only means a bad AI-generated query can't touch your production data, which is the main safety net worth keeping on.
5. Add AI context so answers use your business terms
Press Command+K, search "AI context," and add short descriptions of what your tables and columns actually mean. Basedash automatically pulls table and column names and any existing database comments, but it doesn't know that "status = 3" means "churned" until you tell it. I added three lines of context to my test source — what "active user" meant, and two odd column names — and a follow-up question about "active users last month" pulled the right filter on the first try where it had guessed wrong before.
6. Ask a question to confirm the AI can see it
Open the chat and ask something simple tied to the new source — "how many rows are in [table]" is enough to confirm the connection is live and the AI is actually pulling from it, not from a cached demo answer.
Example prompts you can copy
- Confirm a new source is live: "How many records are in [table name] and when was the most recent one added?"
- Cross-source question (after connecting Warehouse + a database): "Compare last month's Shopify orders to what's in our orders table — any mismatch?"
- Context-aware follow-up: "Show active users this month" (after you've defined "active" in AI context)
- Source-scoped request: "Only look at the Snowflake warehouse, not the production database, for this one"
Common mistakes to avoid
The mistake I made first: I assumed connecting a database was the same as making the AI "understand" it. It isn't. The AI can see schema and run queries against a fresh connection right away. But its answers stay generic, and sometimes wrong about business terms, until you add AI context. Second, don't expect a raw CSV upload. Basedash's documented sources are databases, warehouses, and Fivetran-backed SaaS connectors, not file drops. Route file-based data through a spreadsheet connector or a warehouse instead. Third, if you're behind a firewall and skip the SSH step, the connection just times out with a generic error. That reads like a wrong password, and it wastes time chasing the wrong fix. Fourth, teams jump straight to write access "to save a step." Leave it read-only until there's an actual reason not to. Fifth, don't connect every Basedash AI source you have access to on day one. I connected three sources in my test before adding any context, and the AI mixed up column names between two of them until I scoped and described each one — one clean source with good context beats five undocumented ones.
Tools that make this easier
If you're evaluating the rest of what Basedash bundles around these sources — Automations, Insights, the Autopilot agent — my Basedash AI Kit guide covers that layer and current pricing in more detail. For a lighter starting point that doesn't require a database at all, my starter kit for AI rounds up free chatbot accounts worth opening first. Teams sizing this against a smaller budget should see best AI tool for small business. If you'd rather write and review the SQL yourself with AI assistance instead of handing it to a chat layer, my best AI tool for code guide covers that alternative. Basedash's MCP server also lets outside AI clients query a connected source directly — anyone comparing that setup against general-purpose agents should read how to use ChatGPT agents. And for the method behind how I verify claims like the ones in this guide, see how we test AI tools.
Comparing the two ways to get a source into Basedash's AI
| Direct connection | Basedash Warehouse | |
|---|---|---|
| Best for | Databases you already run | SaaS tools (Stripe, Salesforce, Shopify, etc.) |
| Source count | 12 database/warehouse types | 750+ connectors via Fivetran |
| Setup | Host, port, credentials, test connection | Authenticate the SaaS tool; Basedash syncs it |
| Where data lands | Stays in your existing database | Managed Postgres warehouse Basedash hosts |
| Typical setup time | Under 5 minutes for a reachable database | Varies by connector; first sync can take longer |
My take
Once I understood the two connection paths, the setup itself was straightforward. It matched Basedash's own documentation closely enough that I trust it as a reference, not just marketing copy. The part worth taking seriously is AI context. Skip it, and the AI answers questions about your schema, not your business. That gap is exactly where wrong-but-confident answers come from. Add context early, keep new connections read-only, and treat every Basedash AI source the same way: connect it, describe it, then ask it something simple before trusting it with a real report. The whole process took me under fifteen minutes per source, context included, which is faster than I expected going in.
Frequently Asked Questions
Is connecting a data source to Basedash free?
The connection itself doesn't cost extra, but Basedash isn't free past the 14-day trial — Startup plans start at $1,000/month for up to 25 seats with $1,000 in AI usage credits included, per Basedash’s pricing page.
How long does it take to connect a source?
A direct connection to a reachable database took under 5 minutes in my test, including the required test-connection check. A SaaS source through Basedash Warehouse depends on the connector and how much historical data it's syncing.
What's the easiest way to start?
Connect the built-in demo database first, ask the AI a basic question, then connect one real source and add a few lines of AI context before inviting a team or connecting anything sensitive.
Can I upload a CSV directly as a source?
No. Basedash's documented sources are databases, warehouses, and Fivetran-backed connectors — there's no raw file upload option, so route spreadsheet data through a supported connector instead.
Does the AI only see sources I've connected?
Yes. Basedash's AI answers are scoped to whatever's connected and whatever permissions your account has — it can't see or query anything you haven't added.