New Research: Supply Chain Attack on Axios Pulls Malicious Dependency from npm.Details →
Socket
Book a DemoSign in
Socket

agentk8

Package Overview
Dependencies
Maintainers
1
Versions
27
Alerts
File Explorer

Advanced tools

Socket logo

Install Socket

Detect and block malicious and high-risk dependencies

Install

agentk8

Multi-LLM Council Terminal Suite - Three-stage consensus with GPT, Gemini, and Claude

latest
npmnpm
Version
2.3.8
Version published
Maintainers
1
Created
Source

AGENT-K

Multi-LLM Council Terminal Suite

╭──────────────────────────────────────────────── AGENT-K v2.3.7 ────────────────────────────────────────────────╮
│      ___________________        ____....-----....____             Welcome to AGENT-K                           │
│     (________________LL_)   ==============================                                                     │
│         ______\   \_______.--'.  `---..._____...---'              Pack Intelligence System                     │
│         `-------..__            ` ,/                                                                           │
│                     `-._ -  -  - |                                                                              │
│                         `-------'                                                                              │
╰────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯

A sophisticated terminal UI for orchestrating AI agents with multi-LLM consensus. Built with React + Ink, featuring Council mode for GPT + Gemini + Claude collaboration.

Features

  • Council Mode - Multi-LLM consensus with GPT, Gemini, and Claude working together
  • Three Execution Modes - Normal (confirm edits), Plan (approval first), Auto (no confirmations)
  • Smart Context Selection - RLM-inspired file selection using LLM reasoning
  • Scout Agent - Real-time research with intelligent codebase navigation
  • Beautiful UI - Dark teal/purple theme with live token tracking
  • Command History - Use ↑/↓ arrows to browse past messages

Installation

npm install -g agentk8

Homebrew (macOS/Linux)

brew tap de5truct0/agentk
brew install agentk8

pip (Python backend)

pip install agentk8

From Source

git clone https://github.com/de5truct0/agentk.git
cd agentk
npm install && npm run build && npm link

Requirements

  • Node.js 18+
  • Claude Code CLI - Install from claude.ai/code
  • Python 3.10+ - For Council mode (optional)
  • API Keys - For multi-LLM: OPENAI_API_KEY, GEMINI_API_KEY, ANTHROPIC_API_KEY

Quick Start

# Start interactive session (Development mode)
agentk8

# Start ML Research mode
agentk8 --mode ml

# Show help
agentk8 --help

Execution Modes

ModeCommandBehavior
Normal/normalExecute tasks, confirm before edits
Plan/planCreate plan for approval before executing
Auto/autoNo confirmations, execute everything directly

Use Shift+Tab to cycle between modes, or type the command.

Council Mode

Council mode enables multi-LLM consensus using three stages:

Stage 1: Initial Analysis
  → GPT-4, Gemini, Claude analyze independently

Stage 2: Cross-Review
  → Each model reviews others' analyses

Stage 3: Chairman Synthesis
  → Claude synthesizes final consensus

Toggle with /council command. Requires API keys for all three providers.

Commands

CommandDescription
/helpShow available commands
/clearClear chat history
/statusShow session status
/normalNormal mode (confirm edits)
/planPlan mode (approval first)
/autoAuto mode (no confirmations)
/councilToggle Council mode
/modelsShow available LLM models
/exitExit AGENT-K

Keyboard Shortcuts

KeyAction
Shift+TabCycle execution mode
/ Browse command history
Esc EscExit (double-tap)
Ctrl+UClear input line
Ctrl+AJump to start of line
Ctrl+EJump to end of line

Agent Teams

Development Mode (Default)

AgentRole
OrchestratorTask decomposition, coordination
EngineerCode implementation, debugging
TesterUnit/integration tests, coverage
SecurityOWASP review, vulnerability detection

ML Mode (--mode ml)

AgentRole
OrchestratorML project lifecycle coordination
ResearcherLiterature review, SOTA analysis
ML EngineerModel implementation, training
EvaluatorMetrics, benchmarking, experiments

Smart Context Selection

The Scout agent uses RLM-inspired (Recursive Language Models) context selection:

  • Scans project file tree
  • Asks LLM to select most relevant files for the query
  • Reads only the selected files
  • Falls back to pattern matching if LLM fails

This dramatically improves context relevance compared to naive "top N files" approaches.

Tech Stack

  • React + Ink - Terminal UI framework
  • TypeScript - Type-safe codebase
  • LiteLLM - Multi-LLM Python backend
  • Claude Code CLI - AI backend

Project Structure

agentk/
├── src/
│   ├── cli.tsx              # CLI entry point
│   ├── components/
│   │   ├── App.tsx          # Main application
│   │   ├── StatusBar.tsx    # Footer with mode indicator
│   │   ├── ChatMessage.tsx  # Message display
│   │   └── Input.tsx        # Input with history
│   └── lib/
│       ├── claude.ts        # Claude CLI integration
│       └── council.ts       # Multi-LLM council
├── python/
│   └── agentk/
│       ├── council.py       # Council consensus logic
│       ├── scout.py         # Smart Context Selection
│       └── llm.py           # LiteLLM wrapper
└── package.json

License

MIT License - see LICENSE for details.

AGENT-K v2.3.7 - Multi-LLM Council Intelligence

GitHubnpmPyPI

Keywords

claude

FAQs

Package last updated on 16 Jan 2026

Did you know?

Socket

Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.

Install

Related posts