Coginiti 26.6 Release Notes

Coginiti 26.6 puts the model directly in the data transformation pipeline. LLM Blocks, a new first-class block type in CoginitiScript, let you embed AI reasoning alongside SQL in governed, publishable, testable pipelines that run on every platform Coginiti supports. This release also adds event-based triggers for Project Actions, a unified Settings area for all users, Semantic Layer performance and correctness improvements, and security and stability fixes.

LLM Blocks in CoginitiScript

CoginitiScript now supports LLM blocks as a first-class block type alongside SQL. You can define a prompt, declare a typed output schema, and use the result exactly like a SQL block: reference it in queries, join it with tables, publish it, test it, and schedule it. The model becomes a step in your pipeline, not a separate tool outside of it.

LLM blocks work on every database platform Coginiti supports, including Snowflake, Databricks, BigQuery, Redshift, PostgreSQL and SQL Server. Write once, run anywhere.

LLM blocks also work with any of Coginiti's supported model providers, including OpenAI GPT, Anthropic Claude, Google Gemini, and open-source models served locally through Ollama. You can match the model to the task, choosing a frontier model for complex reasoning or a local Ollama model for air-gapped and regulated environments, without changing your pipeline code.

Typed output schemas. Every LLM block declares the columns, types, and descriptions it returns. The AI generates structured data that conforms to the schema, and the result is queryable immediately as a standard resultset.

SQL and LLM blocks interoperate. Pass SQL results into an LLM prompt with print.Csv() to let the model reason over your data. Reference an LLM block's output in a SQL FROM clause or JOIN just like any other block. Chain LLM blocks together using iterator() for multi-step analysis. SQL and AI work together in a single dependency graph.

Publication, caching, and scheduling. LLM block results can be published to database tables using the same publication.Run() framework that SQL blocks use, including incremental append strategies. Enable result caching to prevent redundant LLM calls when multiple downstream blocks reference the same output. Schedule LLM-powered pipelines with Coginiti Actions just like any other CoginitiScript workflow.

Testable. Write CoginitiScript test blocks that validate LLM output against expected value ranges, referential integrity, and completeness, ensuring AI-generated data meets the same quality standards as your SQL transformations.

Large dataset processing. For datasets that exceed token limits, CoginitiScript supports row-level mapping, fixed-size batching, data-driven batching, and incremental publication patterns that keep each LLM prompt bounded while processing arbitrarily large tables.

To get started, see the LLM Blocks Tutorial and the Large Dataset Processing Tutorial.

Event-Based Triggers for Project Actions

Project Actions can now run in response to project lifecycle events, not just on a cron schedule. Alongside existing schedule-based actions, you can now trigger an action when a review is created or when a project is published to the Project Hub. This lets you automate workflows and enforce consistent processes around the events that matter, such as kicking off validation when a review opens or running downstream steps the moment a project ships.

Unified Settings for All Users

Every user now has a single, discoverable Settings area anchored in the sidebar. Previously, non-admin settings were spread across separate screens for password, API keys, and general preferences. Those settings now live together in one place, starting with General Preferences and API Keys, giving users the same consistent Settings experience administrators already have, regardless of role.

Semantic Layer Improvements

This release includes performance and correctness improvements to the Coginiti Semantic Layer, building on the general availability release in 26.2.

Security and Stability

This release includes security patches and dependency updates, along with bug fixes across platform services.

Get Started

Upgrade to Coginiti 26.6 to start building AI-powered data pipelines with LLM Blocks. Visit the documentation to walk through a complete example.