Moonshot AI launches Kimi Hosted Agents, a managed agent runtime with sandboxes at 0.072 yuan per hour

Moonshot AI has launched Kimi Hosted Agents, a cloud service that packages the agent runtime, or harness, as a managed product so enterprises no longer have to build and maintain that layer themselves. The details come from IT Home and Moonshot's open-platform documentation, as reported by Pandaily.

Moonshot frames the offering as Harness as a Service. Teams configure agents on Kimi's infrastructure and get a 24/7 hosted environment for long-running tasks and multi-agent collaboration.

Key details

• Use cases: 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.
• Isolation and control: Agents run in isolated sandboxes with network-access controls. Operators can stop sessions at any time, and execution logs can be exported for audit.
• Billing: Usage reports break out sandbox hours, model inference and plugin calls. IT Home cites a sandbox rate of 0.072 yuan per hour plus model and plugin charges. Model and plugin rates were not given.
• Catalog: The platform currently lists more than 140 skills and more than 30 plugins, covering web search, professional data queries, office documents and content generation.
• Extensibility: Developers can attach their own tools, data sources and MCP services.

Kimi Inference Hook

An optional security component, Kimi Inference Hook, lets an enterprise's own or third-party safety service inspect task inputs and tool returns before each model call. Only content that passes the company's rules proceeds to inference. The hook is currently limited to enterprise-verified customers.

Self-hosted environments

Official documentation also describes self-hosted environments. Sandbox tools run on machines the customer controls, while orchestration stays on the platform.

Agent templates

Moonshot is shipping a first set of official agent templates. IT Home highlights eight finance templates covering modeling, deep research, morning briefs, IPO tracking and portfolio monitoring. The documentation lists a broader first wave of 12 templates, which also include game operations and SRE incident response. Templates can be copied into a project space and wired to private databases or internal reports.

Context

The launch extends Kimi's product line from models and desktop agents into a hosted runtime that Moonshot says it will maintain and upgrade. It comes a day after Huawei Ascend announced day-zero inference support for the open-weight Kimi K3.1 Pro model.

What this means

Moonshot is moving responsibility for the agent harness, meaning the sandbox, session control, logging and tool integration, from the customer to the vendor. For enterprises, the controls described are the ones audit and security teams typically ask about: network-access limits, stoppable sessions, exportable logs and a pre-inference inspection step. The self-hosted option, where sandbox tools run on customer machines while orchestration stays on the platform, addresses cases where execution must stay on infrastructure the customer controls.

Billing is split into sandbox hours, model inference and plugin calls, so total cost will depend on workload mix rather than the 0.072 yuan hourly sandbox rate alone. The restriction of Kimi Inference Hook to enterprise-verified customers means the pre-inference screening feature is not yet broadly available.