Zo Computer Improved AI Agent Reliability 20x with Vercel

Zo Computer, a personal cloud platform with an AI agent, improved its system reliability 20-fold. The key to success was implementing Vercel AI Gateway and SDK, which unified operations across different model providers.

Problem: An adapter for every API

Zo Computer allows users to work with any AI model and upload their own keys. This means the platform needs to support OpenAI, Anthropic, MiniMax, GLM, Fireworks, and many others. Previously, each provider required a separate adapter—different image processing, different key management logic, different edge cases. When a new model launched weekly, engineers had to manually write an adapter, test it, and deploy it. For an 8-person team, this was a constant time drain. The initial metrics were grim: 98% success rate with 7.5% retry rate. Every 50th user message either failed or required a retry.

Solution: SDK + Gateway Instead of Hundreds of Integrations

Zo switched to Vercel AI SDK and AI Gateway—two layers that divided responsibility. AI SDK replaced all the custom adapter code. Instead of separate implementations for each provider, engineers got a single interface. Image processing is normalized, response formats are unified. AI Gateway handled the infrastructure layer:

• Automatic retry logic instead of manual management
• Routing between providers and fallback to alternative models
• Real-time monitoring and optimization
• New models are added via configuration without writing code

After implementation, Zo was able to add support for new models in under a minute.

Numbers: 20x Improvement and More

The results are impressive:

• Retry rate: 7.5% → 0.34% (reduced 20x)
• Success rate: 98% → 99.93%
• P99 latency: 131 sec → 81 sec (minus 38%)
• Time to add model: <1 minute without code

Rob Cheung, Zo Computer co-founder: "We didn't even realize the problem until we moved to Vercel. The numbers showed the truth—there were too many failures before."

What This Means

For startups like Zo, this means they're no longer dependent on internal infrastructure around each model. The team can focus on the product—on how users interact with the AI agent. Zo is targeting a million new users in 2026, and now the technology allows for it.