Databricks Unveils Omnigent, LTAP, Lakebase, and Genie at Data + AI Summit 2026
Deploy Omnigent to unify agent session APIs across your existing LLMs and configure spend controls.
Deploy Omnigent and configure agent session APIs, enable spend controls, and test persistent session sharing.
Summary
Databricks announced a suite of new data‑and‑AI operating system components at the 2026 Data + AI Summit. The open‑source Omnigent meta‑harness lets teams combine, control, and share agents across Claude Code, Codex, Cursor, Pi, and custom agents, providing a unified session API for files, streams, tool calls, and cancellation. LTAP (Lake Transactional Analytical Platform) is positioned as the first lake‑based HTAP solution, unifying the storage layer to deliver transactional and analytical workloads without collapsing query engines. Lakebase rethinks the database stack, emphasizing column‑oriented transactional writes and live operational context for agents. Genie introduces an agent runtime with RL fine‑tuning, document parsing models, and a Mosaic model strategy that blends open formats with AI. The announcement also highlighted agent security, contextual and stateful policies, spend controls, persistent sessions, cloud sandboxes, and collaboration features. Databricks’ rapid prototyping culture and scale—50‑60 million VMs a day—underscores its ambition to become the operating system for enterprise agents.
The release signals a shift from data storage to exposing the right slice of state, history, permissions, and business logic to AI systems at the moment of work, positioning Databricks as a key player in the emerging agent ecosystem.
Key changes
- Omnigent open‑source meta‑harness for combining, controlling, and sharing agents across Claude Code, Codex, Cursor, Pi, and custom agents
- LTAP first lake‑based HTAP solution unifying storage for transactional and analytical workloads
- Lakebase rethinks database stack with column‑oriented transactional writes and live operational context
- Genie agent runtime with RL fine‑tuning, document parsing models, and Mosaic strategy
- Agent security with contextual and stateful policies and spend controls
- Persistent sessions, cloud sandboxes, sharing, search, and collaboration features
- Open formats and AI integration across the stack
- Databricks’ rapid prototyping culture and 50‑60M VMs/day scale