News · SI

Kimi launches Hosted Agents, a managed harness for long-running AI work

Moonshot AI's Kimi has formally launched Hosted Agents, a managed cloud harness for long-running and multi-agent work, with sandboxes, 140+ skills, finance templates and Inference Hook security checks.

KimiMoonshotAgentsEnterprise

Kimi Hosted Agents launch art for Moonshot AI harness-as-a-service
Image: Kimi / Moonshot AI

Moonshot AI's Kimi has formally launched Kimi Hosted Agents, packaging the agent harness as a cloud service so enterprises get a 7×24 managed runtime for long-running tasks and multi-agent collaboration without building that stack themselves[1][2]. Sign-up is live at platform.kimi.com[1].

01 / 04

Internal ops and product agents

Kimi positions the service for two patterns: agents wired into internal systems for monitoring and operations response, and agents embedded in customer-facing products for research or customer service[1]. Isolated sandboxes, network access controls, configurable boundaries, interruptible sessions and audit export are meant to keep those runs controllable and reviewable[1][2].

02 / 04

Skills, plugins and finance templates

Hosted Agents ships more than 140 Skills and 30-plus plugins spanning web retrieval, professional data, documents and content generation, with room for custom tools and data sources[1]. Official templates include eight finance workflows covering research modelling, deep research, morning briefs, IPO tracking and portfolio monitoring that teams can copy into their own project space[1][2].

03 / 04

Security and pricing

Enterprises can pair Hosted Agents with Kimi Inference Hook so pre-inference checks inspect task inputs and tool returns against company rules before the model runs, currently for enterprise-verified users[1]. Pricing combines sandbox runtime at ¥0.072 per hour with model token rates and plugin call charges[2][3].

04 / 04

Why it matters

Most of the cost of production agents is still harness work: sandboxes, sessions, tools, recovery and audit. By selling that layer as a managed service with finance-ready templates and an optional Inference Hook, Kimi is trying to shorten the path from prototype to a supervised, long-running agent inside the enterprise stack[1][2].

FAQ

Frequently asked questions.

01

What are Kimi Hosted Agents?

A managed cloud harness from Moonshot AI's Kimi that runs long-running agent tasks and multi-agent collaboration without enterprises building their own runtime framework.

02

What can enterprises build with it?

Internal ops and monitoring agents, plus customer-facing product agents such as research and customer-service assistants, using isolated sandboxes and configurable boundaries.

03

What skills and templates ship with it?

More than 140 Skills and 30-plus plugins covering web search, data, documents and content, plus official templates including eight finance workflows such as research modelling, morning briefs, IPO tracking and portfolio monitoring.

04

How does security work?

Agents run in isolated sandboxes with network access controls, interruptible sessions and audit export. Hosted Agents can also plug into Kimi Inference Hook for pre-inference security checks for enterprise-verified users.

05

How is it priced?

Sandbox runtime is billed at ¥0.072 per hour, plus model token rates and plugin call charges. Sign-up starts at platform.kimi.com.

That’s the story

Building with what’s new?

Spectrum Web Co ships products, places talent, builds social narratives and builds custom business tools as monthly services.

Related stories

SI Kimi K3.1 Pro is live, with open weights due by 4 NovemberMoonshot says Kimi K3.1 Pro, a post-trained Kimi K3 with a 1-million-token context, is available now, with weights due by 4 November 2026. Moonshot scores it at 65.7% on Terminal-Bench 4.0, just behind Claude Opus 5.5, while Artificial Analysis scores that test at 67% and ranks the model fourth on its Intelligence Index.Read the storySI Google DeepMind gives its Antigravity agent cheminformatics skills, turning days of drug-design work into about an hourGoogle DeepMind has released Cheminformatics Skills for Google Antigravity. Built on the open-source RDKit toolkit with ChEMBL and PubChem connectors, they let an agent mine the literature for compounds, run matched molecular pair analysis, train potency models and propose new macrocycles with synthesis routes. DeepMind is also offering 10,000 scientists its top Antigravity usage tier free for six months.Read the storySI Fireworks uses its first Forge conference to launch 1,500 token-per-second inference and run its engine on customers' own GPUsAt its first Forge conference, Fireworks put Ultra-fast inference into public preview at up to 1,500 tokens per second, added a half-price Background Mode and multi-day prompt caching for agents, launched Nexus Desktop, and previewed Fireworks Sovereign for running its inference engine on customer-owned infrastructure.Read the story