---
type: "post"
id: "moonshot-launches-kimi-hosted-agents-as-cloud-harness-f11e04"
slug: "moonshot-launches-kimi-hosted-agents-as-cloud-harness-f11e04"
title: "Moonshot launches Kimi Hosted Agents as cloud Harness-as-a-Service"
description: "Moonshot’s Kimi Open Platform presents Hosted Agents, a managed cloud harness for teams building long-running or asynchronous agents."
retrieval_nugget: "Moonshot’s Kimi Open Platform presents Hosted Agents, a managed cloud harness for teams building long-running or asynchronous agents. It combines execution, session persistence, recovery and operational controls with skills, plugins and templates, shifting some infrastructure work from a product team to a vendor service."
published_at: "2026-11-05"
updated_at: "2026-11-07"
record_date: "2026-11-05"
date_kind: "published_at"
topics: ["mcp"]
entities: ["Moonshot AI","Kimi Hosted Agents","Kimi Open Platform","Kimi Inference Hook","MCP"]
related_patterns: ["harness-architecture-outlives-model-choice"]
novelty_label: "new_fact"
what_changed: "Moonshot’s Kimi platform now offers Hosted Agents, a managed cloud harness with persistent sessions, sandbox execution, recovery, and usage-based billing."
prior_refs: ["https://newruntime.com/posts/moonshot-publishes-kimi-k3-1-pro-open-weights-1cfc2b/","https://newruntime.com/posts/moonshot-launches-kimi-k3-1-pro-on-the-d876cd/"]
source_url: "https://platform.kimi.com/docs/hosted-agents/overview"
source_urls: ["https://platform.kimi.com/docs/hosted-agents/overview"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:moonshot-launches-kimi-hosted-agents-as-cloud-harness-f11e04"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/moonshot-launches-kimi-hosted-agents-as-cloud-harness-f11e04/","markdown":"https://newruntime.com/posts/moonshot-launches-kimi-hosted-agents-as-cloud-harness-f11e04.md","json":"https://newruntime.com/posts/moonshot-launches-kimi-hosted-agents-as-cloud-harness-f11e04.json"}
markdown_url: "https://newruntime.com/posts/moonshot-launches-kimi-hosted-agents-as-cloud-harness-f11e04.md"
---

# Moonshot launches Kimi Hosted Agents as cloud Harness-as-a-Service

## Retrieval answer

Moonshot’s Kimi Open Platform presents Hosted Agents, a managed cloud harness for teams building long-running or asynchronous agents. It combines execution, session persistence, recovery and operational controls with skills, plugins and templates, shifting some infrastructure work from a product team to a vendor service.

## Summary
Moonshot’s Kimi Open Platform presents Hosted Agents, a managed cloud harness for teams building long-running or asynchronous agents. It combines execution, session persistence, recovery and operational controls with skills, plugins and templates, shifting some infrastructure work from a product team to a vendor service.

## What happened
Moonshot’s documentation describes Kimi Hosted Agents as a service that runs configurable agents on the Kimi Open Platform’s infrastructure. The intended users are enterprises and developers who want to build either internal business applications—such as monitoring, research or operations response—or agents embedded in customer-facing products. The service is positioned against the platform’s model inference API: direct inference suits teams that want a custom harness and finer control, while Hosted Agents targets long-running or asynchronous work without requiring teams to build their own sandbox and session infrastructure.

The hosted runtime covers several pieces that teams otherwise have to assemble. The platform stores complete session history as events that can be retrieved; runs built-in tools in a cloud sandbox; and takes responsibility for execution-layer retries and scheduling. Documentation says that after a platform failure, execution can resume with behavior consistent with the pre-interruption state. Custom tools remain client-executed. The service also offers isolated sandboxes, configurable network permissions, session interruption, traceable execution and audit export. Certified enterprise users can run sandbox tools in a self-hosted environment on their own machines, while model calls, orchestration and session management remain platform-managed.

The launch materials list more than 140 skills and more than 30 plugins for areas including web search, specialist data, office documents and content generation. Developers can also connect custom skills, plugins, MCP services and tools. The first set includes 12 official agent templates spanning financial research and modeling, regulatory monitoring, fund reconciliation, game development and advising, SRE incident response, and overseas community operations. Templates can be copied into a project and adapted to internal systems or private databases.

The platform says Hosted Agents are pay-as-you-go. Charges consist of model tokens, sandbox runtime duration and plugin calls; the overview points to separate pricing documentation but gives no rates. Tasks can be launched through the kha CLI, web console or API. The Kimi Inference Hook, which checks task inputs and tool returns against enterprise-defined security rules before inference, is available only to certified enterprise users by arrangement with sales. Self-hosted sandbox environments also require enterprise certification.

## What the archive adds
This is a new product event, distinct from earlier Moonshot model releases. [Moonshot AI quietly rolls out Kimi K3.1 Pro and K3.1 Pro Highspeed in Kimi Code and the Kimi app](https://newruntime.com/posts/moonshot-ai-quietly-rolls-out-kimi-k3-1-94b950/) covered access to K3.1 Pro models in Kimi Code and the app, not a hosted runtime. Earlier coverage establishes Moonshot’s model and product context; it does not report this managed harness service.

## Trend signal
Hosted Agents supports [Harness architecture outlives model choice](https://newruntime.com/patterns/harness-architecture-outlives-model-choice/): a vendor is productizing the runtime around models—sessions, tools, execution and recovery—as a service in its own right. The docs establish the scope of the offering, but do not yet show adoption or prove that outsourcing this layer is preferable to a custom harness.

## Why now
The documentation frames harness construction as a months-long effort involving execution environments, session and context management, tool calls, permissions, and interruption and recovery. Hosted Agents is presented as a way to shorten the path from prototype to application deployment, with Moonshot maintaining the underlying runtime.

## Why it matters
For a team moving an agent from a demo to an ongoing service, the hard work is not only choosing a model. It also has to preserve session state, run tools inside controlled environments, handle failures and retries, and make activity auditable. Hosted Agents offers one bundled alternative to building those components, and its catalog and templates could reduce initial integration work for supported use cases.

That convenience comes with a consequential boundary: built-in tools execute in Moonshot’s cloud sandbox, while custom tools execute on the client side; self-hosted sandboxes and some security capabilities require enterprise certification. Teams should map their data, network and audit requirements against these boundaries before treating the service as a production shortcut. The documented billing dimensions also make workload testing important: token consumption is only one component of cost, alongside sandbox time and plugin calls. Later coverage: [Kimi launches Hosted Agents as harness-as-a-service](https://newruntime.com/posts/kimi-launches-hosted-agents-as-harness-as-a-c6d7f8/).

## Limits and uncertainty
The supplied overview does not state specific prices, service-level commitments, regional availability, maximum task duration, resource limits or performance results. It does not independently verify reliability, security effectiveness or the breadth and quality of the listed skills and plugins. The failure-recovery and audit features are product claims, not demonstrated outcomes. Enterprise certification gates access to self-hosted execution and Inference Hook, and the page does not establish what contractual or data-retention terms apply to each deployment. Teams will need the detailed pricing and service documentation and a workload-specific evaluation before estimating cost or suitability.

## Reader takeaway
Kimi Hosted Agents is a managed alternative to building the operational harness around an agent: sessions, sandboxed tools, retries, recovery and scheduling are bundled with skills, plugins and templates. It may accelerate long-running agent applications, but the choice depends on control requirements, enterprise access gates and total usage costs—not model-token price alone.

## Open question
How much operational work and total cost will teams actually save compared with running their own harness?

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Mcp](https://newruntime.com/topics/mcp.md): Explore the mcp topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)