Answers from the data warehouse

Connect Computer to Snowflake or Databricks while preserving existing roles, permissions, and security controls.
Build a Data Map that captures business definitions, table relationships, and query patterns for more reliable analysis.
Create reports, dashboards, and recurring monitoring workflows using warehouse data alongside other approved enterprise sources.

Query examples

Frequently asked questions

Connecting Computer to data warehouses

How do data teams work with Computer?

Computer can give business teams a natural-language interface to governed warehouse data and combine it with context from other approved systems. Customers have reduced repetitive data requests and let the data team spend more time on modeling, governance, and strategic analysis. It does not replace the warehouse, semantic layer, or data-quality efforts.

What is the best way to start?

Start with one well-understood business domain and a small group of users. Configure User OAuth, limit access to approved schemas and tools, document the canonical metrics, generate and review the Data Map, and test representative questions against trusted reports. Expand only after the data team is satisfied with SQL quality, permissions, cost, and user behavior.

Can Computer proactively monitor metrics and data pipelines?

Scheduled Tasks can run recurring checks against supported sources and deliver results through configured channels. Users should define the data source, cadence, thresholds, expected output, owner, and escalation path.

Architecture and governance

What is a Data Map, and who should own it?

The Data Map is an organization-level context layer that captures important tables, columns, relationships, query patterns, and business context. It helps Computer translate business questions into warehouse-sourced queries. A data platform, analytics engineering, or data governance owner should maintain it, define authoritative metrics, review proposed changes, and add durable supplementary context.

Does the warehouse remain the system of record?

Yes. The Data Map stores context about how to navigate the data. Current metrics still require an active connection, and SQL executes in the customer's Snowflake or Databricks environment. The warehouse remains the authoritative source for data, compute, permissions, query history, and access policies.

How are identity and permissions enforced?

Computer uses User OAuth with its connectors. Each person signs in with their individual account, and queries run within that account's configured role context. Warehouse-native grants, row filters, column policies, and role configuration remain the authoritative controls. A shared service account for certain connectors is also supported, but it applies the service account's access model rather than each user's individual permissions.

Can Computer reconcile data across warehouses, SaaS applications, files, and the web?

Computer can combine information from enabled sources in one workflow. It does not remove the need for reliable entity resolution. For material reporting, maintain canonical customer, product, and account identifiers upstream, document approved crosswalks, and define the source of truth for each metric.

Accuracy and auditability

How should a data team validate generated SQL and answers?

Computer generates reviewable SQL. Data teams should inspect the SQL, compare material outputs with trusted reports, test edge cases, and correct missing business context in the Data Map.

When systems disagree, how is the authoritative metric selected?

The data team should make the decision explicit. Administrators can edit the Data Map so an approved definition becomes the live ground truth, and durable business definitions can be placed in supplementary context. Conflicting user corrections are routed for human review rather than resolved automatically. For every material metric, document the canonical source, grain, required filters, owner, and known alternatives.

What audit trail is available?

Perplexity Audit Logs capture events across user input, agent actions, completion, errors, and administrative changes. Data warehouses like Snowflake and Databricks retain the authoritative warehouse-side query and identity records.

Can every answer be reproduced exactly for an auditor?

Exact deterministic replay is not publicly documented or guaranteed. Audit Logs, the thread, reviewable SQL, Data Map version history, and warehouse records can help reconstruct a run, but no documented identifier binds every answer to immutable snapshots of the data, permissions, Data Map, and model runtime. Organizations with strict reproducibility requirements should preserve their own versioned evidence manifest.

Security and compliance

Where does data persist, and is it used for model training?

Enterprise data is not used to train or fine-tune Perplexity models. Computer tasks run in isolated sandboxes, and credentials are destroyed with the sandbox. Session artifacts and warehouse query history are governed separately. Attached session files are deleted after seven days. Additional configurable retention settings are available to Enterprise organizations.