Skip to main content

AI in Coginiti overview

Coginiti puts AI in three places: a chat assistant for conversational work, inline tools in the SQL editor, and LLM Blocks for calling a model directly from CoginitiScript. This page maps each feature to the role and edition that can use it, and links to the guide that covers it in depth.

Background​

Coginiti's platform serves people with very different jobs: developers writing CoginitiScript, analysts exploring a semantic layer, and business consumers who just want an answer. A single AI experience can't safely serve all three - a consumer shouldn't see raw schemas, and a developer needs direct database and catalog access. Coginiti resolves this by giving each role a distinct assistant, gating capabilities by role and edition, and asking for explicit approval before anything changes state.

Chat and agents​

Coginiti's chat panel is powered by one of two personas, chosen automatically by your role and edition.

Forge is the admin/user assistant. It's available to every non-Consumer role, and to all Pro users. Forge can search and browse the Project Hub catalog, list database connections and read schemas, list and query semantic layers, search documentation (Team/Enterprise only), and search and fetch the web. It draws Mermaid diagrams and asks clarifying questions. Catalog changes - creating a project or item, deleting an item - always prompt for approval first. Pro Forge doesn't include semantic layer tools or documentation search.

Guide is the consumer assistant, automatically assigned to the Consumer role on Team and Enterprise. It answers in plain language from the semantic layer only: listing semantic layers, reading metadata, running semantic queries, and searching the web. Guide refuses to write raw SQL, expose schema details, or make catalog changes. Consumers open a new chat tab with Alt+T (Option+T on macOS - see the Command Palette Reference for the full mapping used across this page).

Why two personas, and how Coginiti picks one, is covered in Forge and Guide Explained.

Both personas ask before taking any action that changes state, with an Allow, Allow per session, or Deny prompt, and retry a failed model call up to three times before giving up - see how agent approvals work.

  • Using the AI chat - opening the panel (Alt+7 in the browser, Mod+7 in the desktop app), asking, stopping, regenerating, and approving actions: see Using the AI chat
  • Chat history and sharing - searching, renaming, and sharing past chats: see Using the AI chat
  • Limiting what the AI can see - restricting which databases Forge and Guide can access: see How to Configure AI Assistant

AI in the SQL editor​

Inside the SQL editor, AI works inline rather than through chat.

  • Inline AI assistant (Mod+L) edits SQL in place using your selection, open tabs, and catalog items as context, and shows the change as a reviewable diff. Available to User and Admin roles.
  • AI quick actions give one-click help while you write and run queries: Explain SQL (Mod+Alt+E) and Optimize SQL (Mod+Alt+O) from the Edit and context menus, an Explain Error button on failed query output, and AI explanations of nodes in the Explain Plan panel.

See How to Use AI in the SQL Editor for both.

LLM Blocks in CoginitiScript​

Outside the chat panel, CoginitiScript can call a model directly from a script using LLM Blocks - useful for classification, extraction, or generation steps inside a data pipeline. See the LLM Blocks overview, the LLM Blocks how-to guide, and the LLM Blocks tutorial.

Administration​

  • Configure an AI provider - connecting Coginiti to Amazon Bedrock, Anthropic, Azure OpenAI, Google Gemini, Google Vertex AI, Ollama, or OpenAI, and how the admin setting overrides personal setup: see How to Configure AI Assistant
  • AI roles and editions matrix - exactly which features each role and edition unlocks: see AI Roles and Editions Reference
  • Data and privacy - what reaches your AI provider, what goes to web search, and how chat storage and sharing work: see AI Data and Privacy in Coginiti

Getting started​

New to AI in Coginiti? First Steps with AI in Coginiti walks through configuring a provider, having a first chat, making a first inline edit, and explaining a failed query.


See Semantic Layer Overview for the business-oriented data model that Guide, and semantic-layer-scoped Forge queries, run against.


tip

For more information about other exciting Coginiti features visit our documentation

Thank you for choosing us!​