Tursio adds AI cost tracking & daily spending caps

Tursio has introduced per-query AI cost tracking and daily spending caps, while keeping existing database security rules in force for AI search.

The update lets customers see token usage, view the cost of each query in Tursio's portal and set daily budgets for total spending on input and output tokens. When a customer reaches the daily limit, the system stops processing new queries until the cap resets.

The changes are aimed at credit unions and other regulated organisations weighing the cost of broader AI use against tighter controls on data access. Tursio's platform lets users search structured data in operational databases and data warehouses through natural-language queries.

Alekh Jindal, Co-founder and Chief Executive Officer, said cost and security concerns have become central in discussions with financial institutions.

"Whether I'm talking with a small bank or one of the biggest, the conversation comes back to cost and security," said Alekh Jindal, Co-founder and Chief Executive Officer, Tursio. "The largest companies are under the same pressure on AI spending as everyone else. AI costs don't grow in a straight line; they compound quietly, query by query, until someone gets the bill. Now our customers can see what every query costs and decide ahead of time how much they want to spend."

Cost controls

Many organisations initially encouraged staff to use AI tools heavily, only to find that routine tasks could consume large volumes of tokens and drive up costs quickly. Tursio cited reported spending limits at groups including Accenture, Apple, Tesla, Uber and JPMorgan as evidence that tighter controls are becoming more common.

Tursio is applying that approach to AI search in financial services workflows such as member insights, marketing campaigns, portfolio management and fraud detection. In its portal, customers can monitor the price of each query as it runs and review average costs across usage.

According to the company, a typical query across its current deployments costs about three cents, or about two cents when averaged across all databases. It added that estimates put the cost of a query through Fiserv's agentOS at about 20 cents.

Tursio's pricing model bills tokens directly to a customer's existing large language model subscription, such as Azure OpenAI, rather than reselling usage. The company said it lowers spending by reusing context across queries and routing each question through a query plan designed to limit model calls.

Security rules

The second part of the update focuses on governance. Customers have often had to recreate access controls when making enterprise data searchable through AI tools, even when those rules already exist in source systems, according to Tursio.

Under the new setup, the platform reads and enforces controls already configured in systems such as SQL Server and Snowflake. Existing hashing, masking and redaction rules carry over, while identity and access management integrates with Microsoft Entra ID.

The feature was developed in response to demand from a large enterprise customer with detailed access policies. Because the platform runs on premises, enforcement happens inside the customer's own security environment.

Jindal said this removes the need for a second layer of policy work before organisations can search their data with AI.

"Customers have already set up their access rules in SQL Server or Snowflake, and until now they've had to apply all of those rules a second time before they could search that data," said Jindal. "There's no need for that. We pick up the rules they already have and enforce them, so the security they've built stays intact. That matters more over time, because AI can surface data far faster than it could before, so any gap in permissions gets exposed quickly."

The cost tracking, budgeting and mirrored governance features are now available. Tursio's platform also connects with systems including Symitar, SQL Server, MySQL, PostgreSQL, Oracle, Cassandra and Microsoft Fabric, as well as tools such as Microsoft Teams, Copilot and Claude.