Cline routes agent traffic through Vercel AI Gateway
Cline, the open-source coding agent with over 1 million developers and 4 million installations, has moved its model request routing onto Vercel AI Gateway. The change affects how every request from the IDE reaches an inference provider: rather than connecting directly to a model endpoint, traffic first passes through Cline's infrastructure and then through the Gateway.
AI Gateway runs on the same global network that handles trillions of requests for Vercel-hosted applications. Each request terminates at one of more than a hundred worldwide points of presence, then travels across Vercel's private backbone to the nearest region instead of traversing the public internet.
This setup avoids the single-region bottleneck of custom proxies. Traffic dynamically selects the healthiest available path, and for in-cloud workloads most of the route between IDE and provider stays inside Vercel's managed network. The design targets lower round-trip times for interactive coding, where long streaming responses make network variance noticeable.
Measured gains in latency and reliability
Cline validated the switch with a week of live A/B testing, splitting real production traffic between its previous AI router and AI Gateway across its global user base. The results showed improvements in three areas:
- P99 streaming latency improved 10–14% across Cline's most-used models. The slowest 1% of requests represent mid-stream pauses and multi-step reasoning delays, so a shorter long tail means fewer visible stalls during agent workflows.
- API error rates dropped 43.8%. Fewer failures translate into fewer broken generations and retries during extended coding sessions.
- Model-specific gains were also recorded: Grok Code Fast 1 saw 9% faster P95 latency and 13.7% faster P99 latency, while Minimax M2 delivered 14.4% faster P99 latency alongside routing-related cost reductions.
The tested overhead for Gateway routing is sub-20ms for developers worldwide. Since the switch, Cline has become the largest opted-in application by traffic on the AI Gateway Leaderboards.
Transparent pricing and operational visibility
Pricing on the Gateway passes through model provider charges without markup on direct requests or Bring Your Own Key (BYOK) usage. Developers pay exactly what the provider lists for inference.
The Gateway also supplies Cline's team with telemetry, health checks, and per-request insight for diagnosing issues and planning capacity. That combination of predictable cost and observability aligns with Cline's stated approach: systems that are fast, understandable, and transparent about their behavior.
Infrastructure for open-source AI at scale
The migration reflects open-source AI tools moving into production engineering roles. For Cline, the Gateway provides lower latency variance, fewer failures, and clearer pricing without changing its open, extensible model. The practical effect is a coding agent that behaves more consistently across the global developer population that relies on it daily.
