krew-cli lets you run GPT, Claude, Gemini (and any OpenAI-compatible provider) in a single terminal session

zhing2006 — Wed, 25 Mar 2026 09:56:49 +0000

I built a CLI that lets you chat with multiple LLM models (GPT, Claude, Gemini) in one terminal.

What if you could host a roundtable meeting with multiple AI models — all in your terminal?

That's exactly what krew does. It lets you run GPT, Claude, Gemini (and any OpenAI-compatible provider) in a single terminal session, with shared context so agents can see and build on each other's answers.

Quick Start

Get started in 3 commands:

npm install -g @zhing2026/krew
krew config init
krew

The config wizard walks you through setting up providers and agents — no manual config file editing needed.

@ Addressing — Talk to Any Agent

Use @ to control who you're talking to:

› @all What's the best way to handle errors in Rust?
  (all agents respond in order)

› @opus Can you elaborate on the Result type?
  (only Claude responds)

› Tell me more
  (continues with the last respondent)

#Whisper — Private Messages

Want to ask one agent to privately evaluate another's answer?

› @all Propose an architecture for a chat app
  (both agents answer publicly)

› #opus What are the weaknesses in GPT's proposal?
  (only opus sees this — other agents see a placeholder)

You can even create private whisper groups:

› #opus #gemini Discuss the tradeoffs between these approaches
  (only opus and gemini see each other's replies)

AI-to-AI Routing

When an agent's reply @mentions another agent, that agent is automatically dispatched. You can sit back and watch them collaborate (or argue).

Built-in Tools

Agents aren't just chatbots — they can take action:

Tool	Description
`read_file`	Read file content
`write_file`	Create or overwrite files
`edit_file`	Search-and-replace editing
`shell`	Execute shell commands
`glob`	File pattern matching
`grep`	Content search with regex
`fetch_url`	Fetch and parse web pages
`activate_skill`	Load specialized skill instructions

All file operations are sandboxed to your project directory.

MCP Integration

Extend agent capabilities via Model Context Protocol servers — both stdio and HTTP transports are supported:

[[mcp_servers]]
name = "filesystem"
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "."]

Skill System

Define reusable skill packages with a SKILL.md file. Agents automatically discover available skills and activate them when needed:

my-skill/
├── SKILL.md          # Skill definition (name, description, instructions)
├── scripts/          # Helper scripts
└── references/       # Reference materials

Custom Slash Commands

Create your own commands as Markdown files with argument substitution and bash preprocessing:

---
description: Review code for issues
argument-hint: <file_path>
---

Please review the following file: $ARGUMENTS

Here are the recent changes:
!`git diff --cached`

Save as .krew/commands/review.md, then use /review src/main.rs in your session.

Session Management

Persistence — Every message is saved in real-time. Crash? No problem.
Resume — Pick up any previous conversation with /resume
Rewind — Fork from any point in history with /rewind
Auto-compact — Automatic context compression when token limits approach

Prompt Mode for CI/CD

Run one-shot prompts from scripts:

# Code review in CI
git diff HEAD~1 | krew -p "@opus review these changes for bugs"

# JSON output for parsing
krew -p "@all hello" --format json

More Features

Streaming output — Markdown rendering with syntax highlighting and per-agent color coding
Thinking/reasoning — Display model thinking process (configurable: low/medium/high)
Web search — Provider-native web search (OpenAI, Anthropic, Gemini)
Per-agent sampling — Configure temperature, top_p, max_tokens per agent
Project instructions — AGENTS.md files auto-injected into system prompts
Config management — krew config init/add/del/list/doctor for full config CRUD

Supported Providers

Provider	Examples
OpenAI	GPT-5.2
Anthropic	Claude Opus 4.6, Sonnet 4.6
Google	Gemini 3.1 Pro (+ Vertex AI)
OpenAI-Compatible	OpenRouter, LiteLLM, any compatible API

Built with Rust

Single static binary. Zero runtime dependencies. 5 platforms (Windows, Linux x64/arm64, macOS x64/arm64).

GitHub: https://github.com/ZHing2006/krew-cli

Install: npm install -g @zhing2026/krew
Config: krew config init
Run: krew

Feedback and contributions welcome!

DEV Community: zhing2006