AI infrastructure company Mangoboost on Oct. 12 held a press briefing for a new service launch at the Grand InterContinental Seoul Parnas in Seoul and unveiled its serverless AI inference platform, Mango Inference.

The company said Mango Inference is an inference service provided based on Mangoboost's full-stack AI infrastructure. It said users can use AI models without complex infrastructure building or configuration, and that it targets second-level deployment, low latency, maximum throughput and total cost of ownership efficiency.

Users can start AI inference with a single API key. Mango Inference is provided as a serverless service and will later be introduced as a dedicated inference service as well.

Users including SI partners, developers and startups can choose models that can be used immediately without separate settings or use dedicated endpoints. The company said users can access in one place the functions needed to run the service, including secure API key management, inference usage and status checks, and analysis of requests, traffic, latency and tokens, as well as payment history queries.

Mangoboost also cited its self-developed full-stack AI infrastructure as a competitive edge.

The company stressed that its AI optimisation software, LLMBoost, differs from existing inference software that runs only on GPUs by optimising the entire inference-serving process in which data flows from storage to GPUs. It said this removes bottlenecks and boosts inference performance and token efficiency.

Mangoboost CEO Kim Jang-woo (김장우) said its full-stack technology, designed in-house from hardware to software, is what differentiates Mango Inference. He said Mangoboost will set a new standard for fast, efficient and reliable inference.