Multiple AI agents combined, enforces agent policies, collaboration with teammates in real time
Databricks, the Data and AI company, launched Omnigent on June 15. Omnigent acts as a harness that lets organisations manage different AI agent systems under one roof. It helps compose multiple AI agents, control them with advanced policies, and collaborate live with teammates.
It combines multiple models and techniques without rewriting code. It follows contextual policies of each agent respectively, tracks actions and enforces LLM cost budgets and permissions at the harness layer. Users can share live agent sessions through shared links and review files with each other, where one can review, comment and access agents in real time.
The user interface is the same for all users where messages and files are in and text streams and tools access them. A common API wraps both the terminal-based coding agents and agent SDKs. Agents can be launched in user’s own machines or hosted sandboxes, where users can switch between multiple AI agents with one-line changes. It provides its own sandbox that locks down OS access and intercept and transform network requests, providing privacy. Users can manage multiple individual processes and servers via cloud systems like Kubernetes and Terraform.
Databricks launched this initially as an Alpha / Beta release under the Apache 2.0 license. It is still early in its lifecycle, so some roadmap items, like advanced custom YAML policies, are still rolling out.
