On November 5, Moonshot AI officially launched Kimi Hosted Agents, opening the Harness framework that powers AI agents as a cloud service to enterprises and developers. According to ifeng Tech and official channels, companies can get a 24/7 hosted environment supporting long-horizon tasks and multi-agent collaboration without building or maintaining their own runtime.
The pain point is concrete: putting an agent into production previously took months of building a code-execution base — context and session management, tool calls, permission control, task interruption and recovery — plus continuous maintenance afterward. Hosted Agents lifts that foundation into the cloud: teams just define tasks and plug in tools, compressing the path from prototype to production, while Moonshot maintains the underlying environment.
The capability list includes 140+ Skills and 30+ plugins covering web search, professional data queries, office document processing and content generation, with support for private tools and data sources. On security, it offers isolated sandboxes, network access controls, session termination at any time, full execution trails with audit export, and integration with Kimi Inference Hook — a security component that inspects task inputs and tool outputs before every model inference, intercepting sensitive data upstream (currently open to enterprise-verified users).
The go-to-market path is laid out: the official template library ships 8 financial agent templates covering investment research modeling, deep research, morning briefings, IPO tracking and position monitoring, all running continuously in the hosted environment; SF Express and CSC Financial appear among early partners, with CSC noting Kimi K3's stability in identifying and attributing risk clues in long-horizon financial analysis.
Agent hosting is the natural extension of the "model companies selling picks and shovels" logic — above inference billing, runtimes, tool marketplaces and security audits are forming a new monetization layer. As Anthropic and OpenAI make sandboxed execution standard, Chinese model companies are keeping pace for the first time, and the barrier to moving agents from demo to production has genuinely dropped for teams without infrastructure groups. The real differentiator will be stability pricing for long-horizon tasks: whoever's hosted environment posts the longest uptime and lowest rework rates turns this layer into profit.