Akamai has introduced two products in beta aimed at companies running AI agents: Akamai Inference, a managed model-serving service, and Akamai Agent Sandbox, an isolated environment where agents can execute the code they write. The company’s press release is dated October 27.
Akamai Inference
Akamai Inference is a managed service for open-weight and custom models. According to Akamai, it is reached through OpenAI-compatible APIs, so applications built against that interface can move over without being rewritten. Customers can choose serverless, pay-per-use inference or dedicated deployments, and requests run on Akamai’s global network. Techstrong, citing SiliconANGLE, reports the service launches with 12 open-weight models ranging from 8 billion to 2.8 trillion parameters. We have not independently verified that list. Akamai says capacity on NVIDIA B300 GPU clusters is available through its sales team.
Akamai Agent Sandbox
The sandbox addresses a problem that grows as agents write and run their own code: that code needs to be contained. Akamai describes Agent Sandbox as an open-source component of the Spin framework that uses WebAssembly isolation to set limits on compute, memory, outbound connections and system capabilities. Agents themselves run on Akamai Functions, while the code they generate runs in the separate isolated environment.
Akamai positions this against sandboxes that wrap traditional containers, saying it isolates at the runtime level instead. That is the company’s own comparison, and no independent testing of the claim was found. Spin is a project in the Cloud Native Computing Foundation’s sandbox tier and the technology came to Akamai through its acquisition of Fermyon, according to coverage; we found no CNCF statement about Agent Sandbox itself.
Where it fits
Both products are in beta, with the Spin SDK as the access channel for the sandbox. The sources we reviewed did not give pricing or a general-availability date.
The launch adds a CDN and cloud-network operator to a category that startups such as Daytona, which recently raised a $60 million Series B, have been building out. For teams evaluating agent infrastructure, the notable design choice is WebAssembly isolation with restricted outbound access rather than containers or microVMs, which should be tested against real workloads before adoption.