Perplexity turns Snowflake into a plain-English analyst

Getting answers out of enterprise data usually involves the same painful ritual: submit a ticket to the data team, wait for someone to write SQL, and hope the dashboard matches what you actually asked.

Perplexity wants to skip that line. The company just released a new Snowflake App Connector for its Computer product, allowing employees to query enterprise data warehouses in plain English.

The feature is available now for Pro, Max, and Enterprise subscribers.

Turning company data into plain text

Instead of just throwing an LLM at raw database tables and hoping for accurate SQL, Perplexity relies on a shared semantic context layer.

Once connected, admins can click a button to generate a data map, which teaches the model which tables matter, what specific metrics mean, and how business concepts map to SQL logic. Depending on the size and complexity of the warehouse, building this map takes up to 90 minutes.

Once set up, anyone can ask tactical questions without code. You can ask for a demand forecast for the next eight weeks to adjust staffing, pull the top drivers of support tickets this quarter, or check which marketing campaigns had the highest conversions.

The tool actually started as an internal project at Perplexity. The team built a Slackbot wired to their own Snowflake warehouse to track product usage and support signals. Today, internal teams across product, growth, and security run up to 3,000 queries per week through it.

On the security side, it respects Snowflake role-based access controls and can be locked to read-only mode. Perplexity recommends User OAuth so employees only see data their individual accounts already have clearance for. The company stated enterprise data is never used to train models.

Why it matters

We think this is a smart, grounded move for Perplexity as it expands deeper into enterprise workflows. Most text-to-SQL features fail because business logic is messy and undocumented. Forcing the system to generate a data map upfront takes a bit of time, but it keeps the AI from hallucinating queries against deprecated tables.

For non-technical staff in marketing or operations, the payoff is immediate if their company's data hygiene is already decent. That said, it is not magic: if your data warehouse is an unorganized mess, the AI will simply surface wrong answers faster.

More broadly, Perplexity is showing that its ambition goes way beyond web search. By tying itself directly into Snowflake, Computer is turning into an interactive BI layer, putting Perplexity in direct competition with traditional enterprise software players.