Vercel's Eve vs. Databricks' Omnigent: Two Open-Source AI Agent Frameworks, Two Different Strategies

In a span of just a few days in June 2026, two major infrastructure companies — Vercel and Databricks — each open-sourced a new AI agent framework. The announcements share a common strategic bet: that the next major platform shift will be agent-driven, and that the winning approach will be open-source. But their products target very different problems.

Vercel's eve is a developer-centric framework for building, running, and scaling individual agents. Databricks' Omnigent is a meta-harness for orchestrating multi-agent systems, designed to sit above existing agent frameworks and provide a unified governance and collaboration layer. Each reflects a distinct theory of what enterprises need most right now.

Vercel — Eve: "Next.js for Agents"

What was announced

At its Ship 2026 conference on June 17, Vercel launched eve, an open-source agent framework published as the npm package eve under the Apache-2.0 license . The framework was extracted from the common architecture Vercel observed after building "hundreds of agents" internally .

Key features

• Agent-as-a-directory — every agent is a single folder containing agent.ts, instructions.md, tools/, skills/, subagents/, channels/, and schedules/. The file tree defines the agent's capabilities at a glance .
• Durable execution with automatic retries and state persistence .
• Sandboxed compute — agents get an isolated microVM for running code they write .
• Human-in-the-loop approvals — a single needsApproval flag on any tool gates expensive or destructive actions .
• Subagents — agents can delegate to sub-agents (same directory shape) nested inside subagents/ .
• Evals for testing agent behavior .
• Vercel Connect integration — agents request temporary, scoped credentials instead of storing long-lived tokens .
• Designed to feel like scaffolding, running, and deploying a Next.js app .

Internal adoption metrics

Vercel stated that it builds and runs its own agents on eve, and that more than 100 agents are in production at Vercel using the framework .

Strategic rationale

• Standardize agent development — Vercel argues that building agents today is "where the web was before frameworks," with everyone hand-rolling the same plumbing . eve aims to be what Next.js did for web apps: turn ad-hoc patterns into a single, opinionated structure.
• Deepen the Agent Stack — eve is the top-layer framework that ties together Vercel's existing agent infrastructure primitives (AI SDK, AI Gateway, Workflow SDK, Sandbox, Chat SDK, Connect) into one end-to-end developer experience .
• Drive platform adoption — by making agent development as easy as scaffolding a Next.js app, Vercel positions itself as the default deployment target for agents, just as it became the default for front-end apps .

Databricks — Omnigent: "Kubernetes for Agents"

What was announced

On June 13–18, 2026, Databricks open-sourced Omnigent, a "meta-harness for AI agents," released in alpha under the Apache 2.0 license . The project was announced by Databricks co-founder Matei Zaharia on X .

Key features

• Multi-agent composition — sits above existing agent systems (Claude Code, OpenAI Codex, Pi, custom agent SDKs) and provides a unified interface for routing tasks across them .
• Live collaboration — teams can share live agent sessions, review outputs together, and guide workflows in real time .
• Rich control policies — a stateful policy engine that tracks agent actions and enforces guardrails beyond prompt-based controls, covering security, cost, and access .
• Cloud execution environments with OS-level sandboxing .
• Vendor-neutral orchestration — designed to prevent lock-in to any single agent provider while maintaining centralized governance .

Internal adoption context

Databricks said Omnigent drew directly on lessons from its internal deployments of AI agents and from building agent-based products such as Genie . The broader Databricks platform serves more than 20,000 organizations and 70% of Fortune 500 companies . Databricks also offers a managed version of Omnigent as part of its commercial Agent Bricks platform .

Strategic rationale

• Solve fragmentation — enterprises running multiple agent systems (Claude Code, Codex, Pi, in-house) face integration complexity that scales non-linearly. Omnigent provides a single orchestration and governance layer above all of them .
• Open-source playbook — Databricks has a proven track record of open-sourcing foundational infrastructure (Apache Spark, MLflow, Delta Lake) to drive ecosystem adoption, then monetizing the managed enterprise version .
• Governance as a differentiator — as agents move from experiments to production, organizations need auditability, cost controls, and security policies that adapt to agent behavior, not just static prompts .
• Position as "Kubernetes for agents" — if widely adopted, Databricks' abstractions could become the default coordination protocol for multi-agent fleets, creating a natural on-ramp to its commercial platform .
• Complement to Agent Bricks — Omnigent extends Databricks' existing Agent Bricks platform with an open-source orchestration layer, while the proprietary Supervisor Agent and Unity Catalog governance remain part of the commercial offering .

Two strategies, one bet

Vercel's eve focuses on making the individual agent easy to build, test, and deploy — a horizontal tool for any developer building agentic features. Databricks' Omnigent focuses on managing the chaos of multiple agents across teams and vendors — a control plane for enterprise-scale deployment. Both companies are betting that open-source adoption will drive demand for their commercial platforms: Vercel's deployment infrastructure and Databricks' managed data and AI platform.

For developers evaluating these tools, the choice depends on the problem: use eve when you need to build a single agent quickly and ship it with a familiar developer experience; evaluate Omnigent when you need to orchestrate and govern a fleet of agents across different frameworks. The two are not directly competitive — in fact, they could be complementary, with eve-built agents eventually orchestrated by Omnigent.