Ora benchmarks all major AI agents on Vercel, finds eve outperforms rivals

Ora's founding is rooted in a gap Assaf Elovic discovered at his previous company, Tavily, which Nebius acquired earlier this year. Tavily built a web search engine for AI agents—solving half the problem. But an agent that finds a product still must actually use it. Elovic and co-founder Liad Yosef started Ora to measure how ready the web is for agents and to fix what isn't.

The core challenge Ora solved is infrastructure fragmentation. Every major agent framework—Claude Code, ChatGPT, Gemini, and others—expects its own environment and exposes steps differently. Running them side by side required separate runtimes for each harness, with full tracing of every step. This depth is what Ido Finder calls "one of the most valuable things ora brings to its customers": without the trace, a benchmark score is meaningless. When an agent stalls, the customer sees exactly which step failed and what the agent tried.

Ora's decision to run entirely on Vercel—front end, back end, and agent runtime sharing the same deployment path, logs, and authentication—became the foundation for both testing and internal operations. That consolidation feeds back into Ora's own engineering: 16 people shipping hundreds of commits a day, with coding agents handling day-to-day infrastructure work on eve. Finder saves a few hours a week from this setup; Elovic credits similar savings to how well coding agents build with Vercel's libraries. As Ora scales, it is splitting its platform into microservices, all on Vercel, so internal eve agents will deploy as one more service with no extra configuration.