Moonshot AI has launched Kimi Hosted Agents, a cloud service that packages the agent runtime — what the company frames as Harness as a Service — as a managed product, removing the need for enterprises to build and maintain their own orchestration layer2. The service provides a 24/7 hosted environment for long-running tasks and multi-agent collaboration, configurable on Kimi's infrastructure.
The launch follows by roughly a day Moonshot AI's open-weight release of the 2.8-trillion-parameter Kimi K3.1 Pro model on November 4, which included day-zero Huawei Ascend inference support[1][2].
Runtime architecture and controls
Kimi Hosted Agents provides isolated sandboxes with network-access controls, session boundaries that operators can halt at any time, and execution logs exportable for audit. Usage reports break out sandbox hours, model inference, and plugin calls; IT Home cites a sandbox rate of 0.072 yuan per hour, with model and plugin charges billed separately.
The platform currently lists more than 140 skills and more than 30 plugins spanning web search, professional data queries, office documents, and content generation. Developers can also attach their own tools, data sources, and MCP services.
An optional security layer, Kimi Inference Hook, lets an enterprise's own or third-party safety service inspect task inputs and tool returns before each model call, blocking any content that fails the company's rules from reaching inference. Kimi Inference Hook is currently limited to enterprise-verified customers.
Templates and deployment modes
Moonshot AI is shipping a first wave of 12 official agent templates alongside the managed runtime. IT Home highlights eight finance-oriented templates covering modeling, deep research, morning briefs, IPO tracking, and portfolio monitoring. The broader set also includes game operations and SRE incident response. Templates can be copied into a project space and wired to private databases or internal reports.
Companies can point agents at internal systems for monitoring, research, and operations work, or embed them in customer-facing products as research or support assistants. Official documentation also describes self-hosted environments in which sandbox tools run on machines the customer controls while orchestration stays on the platform.
ANALYSIS By productizing the harness layer the day after open-weighting its flagship model, Moonshot AI is separating the model artifact from the revenue-bearing service: the weights travel freely, but the managed runtime, sandboxing, audit tooling, and inference-hook security carry the commercial load.