What happens if Lava goes down?

Reliability and Uptime

This is one of the most common questions from engineering teams evaluating Lava, and it's a great one. When you add any service to your critical path, you need to understand the failure modes.

Lava's Infrastructure

Lava's AI gateway runs on Vercel's Edge Network, which provides high availability and low latency globally. The gateway adds less than 20ms to your API calls. We have not experienced a major outage to date, but we plan for the possibility like any responsible infrastructure provider.

If Lava Goes Down

If Lava's gateway becomes unreachable, API calls routed through the gateway will fail. Here's how to handle this:

• SDK retries — Lava's SDK includes built-in retry logic. If the gateway doesn't respond, it can fall back to calling the AI provider directly using your own API keys.
• Bring Your Own Keys (BYOK) — If you use your own provider API keys with Lava's gateway, you already have the credentials to call providers directly as a fallback.
• POST request mode — If you use Lava's POST request integration (instead of the gateway), Lava is not in your critical path at all. You make the AI call yourself and send usage data to Lava asynchronously. If Lava is down, your requests still work — we just catch up on metering when we come back online.

If an AI Provider Goes Down

One of the advantages of using Lava's gateway is that you can switch between AI providers quickly. If Anthropic goes down, you can reroute traffic to OpenAI or Google with a configuration change — no code deployment needed. The gateway also supports fallback model routing in priority order.

Understanding the Risk

It's important to distinguish between two scenarios:

1. Lava gateway down — You fall back to direct provider calls. Usage during this period won't be metered, but your service stays up.
2. AI provider down — Lava's gateway helps by enabling fast model switching to an alternative provider.

The gateway adds a dependency, but it also adds resilience against provider outages. For teams that want zero additional dependency, the POST request integration mode keeps Lava entirely off your critical path.

SLAs

We're working on formal SLA commitments. In the meantime, our uptime has been consistent since launch, and we're happy to discuss reliability guarantees on a case-by-case basis for enterprise customers.