All videos
0:00 / 0:00
research

AI Stack Explained in 3 Layers (LLM, Agent Harness, Omnigent)

Databricks29 June 2026Watch on YouTube

Part of series

Ep. 3 · Omnigent: AI Agent Orkestratie

Databricks introduceert Omnigent, het open-source meta-framework dat meerdere AI-agenten combineert in één gecontroleerde, interoperabele workflow.

View the series

What you'll learn

  • The AI stack has gained a third layer after the model and the harness: a meta-harness that connects separate agents.
  • Each agent now forms its own closed world, where session, context and history do not cross over between agents.
  • Databricks solves this with Omnigent, an open-source meta-harness that routes questions to the best-fitting agent through one runner.
  • Example agents Debbie the router and Polly the orchestrator show parallel querying and task splitting with mutual review.
  • Users can define their own agent harness and run it locally or through the managed Databricks version.

Frequently asked questions

What problem does Omnigent solve?
Every AI agent is its own closed world, where session, context and history do not carry over to another agent. Omnigent is a meta-harness on top of all agents that routes a question to the best-fitting agent through one runner.
Which example agents appear in the video?
Debbie is the router who sends a question to multiple agents in parallel and waits for all answers. Polly is an orchestrator who splits one task across multiple agents and then has them review each other's work.
Where can I run Omnigent myself?
The meta-harness is open source and users can define their own agent harness. You can run it locally or through the managed Databricks version, which is in beta with documentation updated on September 2, 2026.

Topics

agent orchestrationai architectureopen sourcedatabricks

Sources

What is known about this topic outside the broadcast, and where it says so.

Description from the channel

A new layer just got added to the AI stack. Here are the three layers that make it up, and where Omnigent fits in. Layer 1: Foundation Models Layer 2: The Agent Harness Layer 3: The Meta-Harness (Omnigent) Omnigent is a meta-harness that sits above the agents you already use (Claude Code, Codex, Pi, custom agents) and makes them work together as one system: compose multiple agents, govern them with policies, and collaborate on live sessions. It's open sourced by Databricks. 🔗 GitHub repo: https://github.com/omnigent-ai/omnigent