Kimi

Agent·runs kimi·by Hivemind

Moonshot's Kimi coding CLI.

hive agents install kimi

Source and documentation →

What it can do on the canvas

—Resumes a session
—Reports when a turn ends
—Can be supervised by another agent

launches and reads status; other agents cannot collect its replies.

Options

Set per tile, before the agent starts.

Model--model
Approval

Installing the CLI itself

Hivemind runs a command that is already on your machine — it never installs one for you. This is the publisher's own instruction.

curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash

The manifest

The whole file, as the app reads it. Everything above is derived from it.

Read agent.yaml

# Kimi — built-in provider manifest.
# Authoring format for the def half. See packages/hive-agents/src/manifest.ts.
manifestVersion: 1
id: kimi
label: Kimi
bin: kimi
enabled: true
caps:
  promptDelivery: typed
  turnSignal: false
  resume: none
  supervise: human
  blockedDetection: true
icon:
  viewBox: 0 0 24 24
  attrs:
    fill: currentColor
  shapes:
  - path:
      d: "M21.765.351C22.998.351 24 1.353 24 2.586S22.998 4.82 21.765 4.82h-1.974c-.15 0-.26-.12-.26-.26V2.586A2.237 2.237 0 0 1 21.765.35M9.41 13.388l8.447-8.377c.16-.16.07-.471-.14-.471h-4.55s-.1.02-.14.06l-9.099 9.029c-.14.14-.35.02-.35-.21V4.81c0-.15-.1-.27-.221-.27H.22c-.12 0-.22.12-.22.27v18.57c0 .15.1.27.22.27h3.137c.12 0 .22-.12.22-.27v-3.79c0-.08.03-.16.08-.21l2.826-2.796c.07-.07.16-.08.241-.03l7.546 5.551a8.9 8.9 0 0 0 4.018 1.493c.12.01.23-.11.23-.27V19.76c0-.14-.08-.25-.19-.26a5.8 5.8 0 0 1-2.355-.942l-6.533-4.73c-.14-.09-.15-.32-.03-.441"
options:
- id: model
  label: Model
  flag: --model
- id: mode
  label: Approval
  values:
    yolo:
    - --yolo
    auto:
    - --auto
    plan:
    - --plan
  unattended: auto
install:
  url: https://moonshotai.github.io/kimi-code/en/guides/getting-started
  command: "curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash"
detect:
  rules:
  - when:
      any:
      - contains: allow?
      - contains: confirm?
      - contains: approve?
      - contains: proceed?
      - contains: '[y/n]'
      - contains: (y/n)
    then: blocked
  - when:
      any:
      - contains: thinking
      - contains: processing
      - contains: generating
      - contains: waiting for response
      - contains: ctrl+c to cancel
      - contains: ctrl-c to cancel
    then: working
  default: idle
note: launches and reads status; other agents cannot collect its replies.