Loading…
Introducing Omnigent: A Meta-Harness to Combine, Control and Share Your Agents
Matei Zaharia, Kasey Uhlenhuth, Corey Zumar
- Source
- Databricks
- Published
- Added to Yomu
Summary
Omnigent is an open-source meta-harness designed to combine, control, and share agents across different harnesses, models, and interfaces. Databricks says users currently juggle multiple agents and copy context between them, while builders struggle to combine or replace harnesses with incompatible interfaces. Its runner wraps terminal-based agents and SDKs in sandboxed sessions with a uniform API for messages, files, streamed text, and tool calls; a server adds policies, sharing, and access through terminal, web, mobile, native Mac OS, and APIs. Features include live collaboration, hosted sandbox execution, contextual security and cost policies, OS isolation, and multi-harness authoring. Released in alpha under Apache 2.0, Omnigent is intended to provide a durable layer above changing agents and harnesses.
Context
Databricks describes users managing four or five agents at once and moving text between agents and collaboration tools. Agent builders also face incompatible harness interfaces, while composition, governance, and collaboration often span multiple harnesses, models, and people.
Approach / What changed
Omnigent places a common interface and server above terminal-based agents and agent SDKs. Runners create sandboxed sessions, while the server provides composition, live sharing, contextual security and cost policies, hosted execution, multiple client interfaces, OS sandboxing, and multi-harness authoring.
Takeaways
- A runner wraps terminal agents such as Claude Code, Codex, and Pi, along with agent SDKs, behind a common API for messages, files, text streams, and tool calls.
- Contextual policies can track session state, require approval after specified actions, restrict file writes, and pause an agent when its tracked LLM cost reaches a threshold.
- Omnigent can expose one connected agent through web, mobile, Mac OS, and APIs, and supports execution on local machines or hosted sandboxes including Modal and Daytona.