Skip to main content

muonroi-cli Overview

muonroi-cli is a TypeScript/Bun CLI that replaces single-model AI assistants with a multi-provider orchestration layer. It routes each prompt through a six-stage intelligence pipeline, runs adversarial multi-model debates for high-stakes decisions, and persists behavioral memory across sessions.

Prerequisites​

Installation​

To install globally via NPM (requires Node.js ≥ 20):

npm install -g muonroi-cli

To install globally via Bun (requires Bun ≥ 1.3):

bun add -g muonroi-cli

Once installed, run the TUI directly:

muonroi-cli

Alternative — Prebuilt Standalone Binary (requires repository access)​

Linux / macOS:

# Note: Requires github authorization if the repository is private
curl -fsSL https://raw.githubusercontent.com/muonroi/muonroi-cli/master/install.sh | bash

Windows PowerShell:

# Note: Requires github authorization if the repository is private
irm https://raw.githubusercontent.com/muonroi/muonroi-cli/master/install.ps1 | iex

Quick Start​

muonroi-cli                                      # interactive TUI
muonroi-cli "fix the flaky test in auth.test.ts" # with starting prompt
muonroi-cli --prompt "run tests" --format json # headless / CI mode
muonroi-cli models # list models with pricing
muonroi-cli doctor # health check

First Run​

On first launch, muonroi-cli checks for ~/.muonroi-cli/user-settings.json. If the file is missing or no API key is configured, the CLI prompts for a key and writes the config. Once credentials are set, the interactive TUI starts automatically.

Run muonroi-cli doctor at any time to verify provider connectivity and configuration.

Supported Providers​

ProviderModelsKey
AnthropicClaude Opus 4.7 / Sonnet 4.6 / Haiku 4.5MUONROI_API_KEY
OpenAIGPT-4o, GPT-4o-mini, o3, o4-miniOPENAI_API_KEY
GoogleGemini 2.5 Pro / FlashGOOGLE_API_KEY
DeepSeekDeepSeek V4 Flash / ProDEEPSEEK_API_KEY
xAIGrok 3, Grok 3 MiniXAI_API_KEY
SiliconFlowQwen, GLM, InternLMSILICONFLOW_API_KEY
OllamaAny local modelKeyless — http://localhost:11434

Multi-Provider Setup​

Role-based routing and council debates require at least two providers. Place the following in ~/.muonroi-cli/user-settings.json:

{
"apiKey": "sk-ant-your-key",
"providers": {
"anthropic": { "apiKey": "sk-ant-..." },
"deepseek": { "apiKey": "sk-..." }
},
"roleModels": {
"leader": "claude-sonnet-4-6",
"implement": "deepseek-v4-flash",
"verify": "claude-sonnet-4-6",
"research": "deepseek-v4-flash"
}
}

roleModels maps each agent role to a specific model. The router uses these assignments when running the Prompt Intelligence Layer pipeline. See CLI Settings Reference for the full schema.

Architecture​

Request Pipeline​

Every prompt passes through the following stages:

User prompt
→ Redactor
→ PIL (Prompt Intelligence Layer)
→ Router
→ Provider
→ Vision Proxy
→ Tool Loop
→ Sub-agent cap
→ Vision Bridge
→ Output guardrails
→ Auto-compact
→ Session storage

Core Subsystems​

SubsystemDescription
Multi-Model CouncilAdversarial multi-model debate for high-stakes decisions
Prompt Intelligence Layer (PIL)Six-layer pipeline routing each prompt to the optimal model
Experience Engine (EE)Persistent behavioral memory across sessions

Source Layout​

PathPurpose
src/orchestrator/Agent loop, auto-compact, council runner
src/council/Multi-model debate engine
src/pil/Prompt Intelligence Layer
src/router/Role-based and tier-based routing
src/providers/Multi-provider factory
src/ee/Experience Engine client and hooks
src/tools/Built-in tools (bash, file, grep, LSP, schedule)

Development​

git clone https://github.com/muonroi/muonroi-cli.git
cd muonroi-cli && bun install

bun run dev # run from source
bun run typecheck # type check
bun run test # vitest
bun run lint # biome check
bun run build:binary # standalone binary

Generate the demo GIF (requires vhs):

vhs docs/demo.tape