Overview

Models, logic and agents that work on the ontology.

Converge AI brings language models to the ontology without giving them a way around its permissions.

Building blocks

Model roles and the gateway: every model call goes through one gateway that chooses a model by role, applies data rules, rate limits and records usage.
Logic: build LLM-powered functions from blocks without writing code.
Agents and the Agent Stack: agents that converse, query objects, call functions and propose actions, with human review of every write.
Evaluations and K-LLM ensembles: measure quality and combine several models for a more reliable answer.
Decisions API: ask a decision model structured questions about a state.

Same permissions as the user. AI tools run with the caller's identity. An agent cannot read an object its user cannot read.
Writes through actions only. Models change data only by applying actions, and write tools wait for a person.
Data stays governed. Data classified as restricted is refused by the gateway unless the deployment explicitly allows it.
Untrusted content stays data. Tool results and external content are passed to models marked as untrusted, never as instructions.

In the workspace, AI applications are grouped under AI in the navigation rail.