---
type: "post"
id: "kimi-launches-hosted-agents-as-harness-as-a-c6d7f8"
slug: "kimi-launches-hosted-agents-as-harness-as-a-c6d7f8"
title: "Kimi launches Hosted Agents as harness-as-a-service"
description: "Moonshot has launched Kimi Hosted Agents, a managed cloud runtime for long-running and multi-agent workflows."
retrieval_nugget: "Moonshot has launched Kimi Hosted Agents, a managed cloud runtime for long-running and multi-agent workflows. It bundles execution, controls and integrations rather than requiring teams to build and maintain their own harness, making the operational boundary and cost model the central questions for builders."
published_at: "2026-11-06"
updated_at: "2026-11-07"
record_date: "2026-11-06"
date_kind: "published_at"
topics: ["engineering-management"]
entities: ["Moonshot AI","Kimi Hosted Agents","Kimi Inference Hook","SF Intra-city","CITIC Securities"]
related_patterns: ["ai-native-orgs-move-to-review-and-orchestration"]
novelty_label: "same_event"
prior_refs: ["https://newruntime.com/posts/moonshot-launches-kimi-hosted-agents-as-cloud-harness-f11e04/"]
source_url: "https://www.kimi.com/news/kimi-hosted-agents"
source_urls: ["https://www.kimi.com/news/kimi-hosted-agents"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:kimi-launches-hosted-agents-as-harness-as-a-c6d7f8"
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/kimi-launches-hosted-agents-as-harness-as-a-c6d7f8/","markdown":"https://newruntime.com/posts/kimi-launches-hosted-agents-as-harness-as-a-c6d7f8.md","json":"https://newruntime.com/posts/kimi-launches-hosted-agents-as-harness-as-a-c6d7f8.json"}
markdown_url: "https://newruntime.com/posts/kimi-launches-hosted-agents-as-harness-as-a-c6d7f8.md"
---

# Kimi launches Hosted Agents as harness-as-a-service

## Retrieval answer

Moonshot has launched Kimi Hosted Agents, a managed cloud runtime for long-running and multi-agent workflows. It bundles execution, controls and integrations rather than requiring teams to build and maintain their own harness, making the operational boundary and cost model the central questions for builders.

## Summary
Moonshot has launched Kimi Hosted Agents, a managed cloud runtime for long-running and multi-agent workflows. It bundles execution, controls and integrations rather than requiring teams to build and maintain their own harness, making the operational boundary and cost model the central questions for builders.

## What happened
Moonshot says Kimi Hosted Agents is available through platform.kimi.com as a cloud service for businesses and developers. The product is intended to support agents that run continuously, handle long tasks and collaborate across multiple agents. The company describes two target settings: agents integrated into internal systems for monitoring or operations response, and agents embedded in customer-facing products such as research or customer-service assistants.

The runtime includes isolated sandboxes and network-access controls. Customers can define operational boundaries, stop a session, inspect execution records and export audit data. Kimi also says usage reports and a defined billing structure are provided so customers can track consumption and costs by agent project. The source does not specify sandbox resource limits or retention periods for audit records.

The announcement positions the service as an alternative to building an in-house harness, which Moonshot says can require months of work for code execution, context and session management, tool calls, permissions, and interruption and recovery. Kimi says it maintains and upgrades the underlying environment. Developers can define tasks, connect tools and use more than 140 Skills and 30-plus plugins covering web search, specialist data queries, office-document processing and content generation. Customers can also connect their own tools and data sources.

Pricing combines sandbox runtime at ¥0.072 per hour, model inference tokens and fees per plugin call. No complete example bill or comparative cost is provided in the announcement. Eight initial finance-agent templates cover tasks including research and modeling, daily regulatory-case reporting, IPO tracking, portfolio monitoring and fund reconciliation. Templates can be copied into a project and adapted to internal reports, private databases and business processes.

Kimi Inference Hook is an optional security component for checking task inputs and tool outputs before each model inference. Enterprises define the checks; the hook allows or rejects content so only approved material reaches inference. The announcement says this component is currently available only to verified enterprise users by arrangement. SF Intra-city and CITIC Securities are quoted describing operational and finance use cases, but the source does not provide measured deployment outcomes.

## What the archive adds
This is a new Hosted Agents launch, not a restatement of earlier Kimi coverage. The archive has covered [Kimi K3.1 Pro on the Kimi API platform](https://newruntime.com/posts/moonshot-launches-kimi-k3-1-pro-on-the-d876cd/) and [Kimi’s 3D Make workflow](https://newruntime.com/posts/moonshot-ai-launches-3d-make-in-kimi-letting-d35704/); those concern the model and application capabilities, not this managed runtime.

## Trend signal
The release supports the pattern of AI-native organizations shifting effort toward orchestration and review: the product emphasizes task control, audit export and usage reporting around agent work. The announcement does not establish that customers measure accepted outcomes or review time, so it is evidence of product direction rather than proof of operating-model adoption. [AI-native organizations move toward review and orchestration](https://newruntime.com/patterns/ai-native-orgs-move-to-review-and-orchestration/).

## Why now
Moonshot frames the launch around the engineering and ongoing maintenance burden of building a reliable harness before an agent can enter production. Its proposed response is to move that runtime work into a managed service while teams focus on business workflows.

## Why it matters
For a team moving from a prototype to a persistent agent, execution infrastructure is a product decision, not just implementation detail. A hosted sandbox, network controls, session interruption and exportable logs could reduce the amount of runtime machinery a team must own, while keeping explicit points for controlling what an agent can access and reviewing what it did.

The trade-off is that the team adopts a vendor’s environment and billing model. The stated ¥0.072 hourly sandbox charge is only one component: inference tokens and plugin calls add costs, and the announcement gives no worked workload estimate. Teams should therefore compare a representative task end to end, including usage, audit needs, access boundaries and the effort of integrating private systems, rather than comparing the hourly rate alone. The finance templates may shorten setup for specific workflows, but their presence is not evidence of reliable results in a team’s own data or process. Earlier coverage: [Moonshot launches Kimi Hosted Agents as cloud Harness-as-a-Service](https://newruntime.com/posts/moonshot-launches-kimi-hosted-agents-as-cloud-harness-f11e04/).

## Limits and uncertainty
The announcement does not disclose sandbox specifications, concurrency or runtime quotas, geographic availability, service-level commitments, data-retention terms, or a full price schedule for tokens and plugins. It does not report independent evaluations, quantified customer savings, error rates or production acceptance rates. The two customer quotations describe potential and qualitative experience, not measured outcomes. The Inference Hook’s restricted availability also means the described input and tool-output checks may not be accessible to every customer. Builders should verify these operational and contractual details before treating the service as production-ready for a particular workload.

## Reader takeaway
Kimi Hosted Agents sells managed execution around agents: sandboxing, controls, logs, integrations and templates, with the runtime billed alongside inference and plugin calls. It may shift work away from maintaining a harness, but the announcement alone cannot show whether it is cheaper or safer for a specific deployment. Test a representative workflow and confirm the limits, security terms and full cost before committing.

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: [Engineering Management](https://newruntime.com/topics/engineering-management.md): Explore the engineering-management 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)