agents-cli
run any coding agent on your subscription not API costs
Last verified:
What is agents-cli?
agents-cli is the missing toolchain for CLI coding agents, designed to run powerful AI agents at scale. It serves as a unified interface that lets you run any model through any CLI, keeping your preferred interface (like Claude Code) while swapping in different underlying models such as Kimi K2.5, MiniMax M2.5, GLM 5, Qwen3 Coder, or DeepSeek through OpenRouter. The tool is built for engineers and developers who use multiple coding agents and want to prevent config drift across their tools.
Key features include rotating across multiple accounts to avoid usage limits, chaining agents in Unix pipe pipelines across different models, pinning agent versions per project (like .nvmrc but for AI), running parallel agents with DAG dependencies via teams, searching and replaying AI history across all agents, installing skills/MCP servers/commands once and syncing them everywhere, unified config management in ~/.agents/ that syncs to each harness's native format, layered configuration (system, user, team, project), a local browser driven via Chrome DevTools Protocol, scheduling agents on cron for recurring tasks, and Keychain-backed secrets management without plaintext .env files.
Agents CLI is for developers and engineers who use multiple AI coding agents like Claude Code, Codex, Gemini CLI, Cursor, OpenCode, Copilot, Amp, Goose, and Roo. It's ideal for teams who want open, local tooling without cloud SaaS dependency, for CI/CD pipelines that need scripted agent chains, and for anyone tired of managing separate configs for each agent.
agents-cli pricing
Pricing model: Freemium
Free and open-source. Available via curl (curl -fsSL agents-cli.sh | sh), bun (bun install -g @phnx-labs/agents-cli), or npm (npm install -g @phnx-labs/agents-cli). Also available as alias 'ag'. No paid tiers mentioned - 100% local with no cloud SaaS dependency.
agents-cli pros
- Run any model through any CLI interface with OpenRouter support
- Rotate across multiple accounts automatically to avoid rate limits
- Chain different agents in Unix pipe pipelines
- Pin agent versions per project like .nvmrc for AI
- Run parallel agents with DAG dependencies via teams feature
- Search and replay AI history across all agents
- Install skills, MCP servers, commands once and sync everywhere
- Unified ~/.agents/ config synced to every harness's native format
- Four-layer configuration stacking (system, user, team, project)
- Local browser via Chrome DevTools Protocol, no cloud service needed
- Schedule recurring agent tasks with cron support
- Keychain-backed secrets with no plaintext .env files
- 100% local and open-source, no cloud SaaS dependency
- Supports 11+ agents: Claude, Codex, Gemini, Cursor, OpenCode, Copilot, Amp, Goose, Roo, Kiro, OpenClaw
- Live status watching with agents teams start --watch
agents-cli cons
- macOS Keychain integration limits secrets to Apple ecosystem
- Requires multiple installation dependencies (curl, bun, or npm)
- Browser feature requires local Chrome/Chromium installation
- Steep learning curve with many commands and concepts
- No official Windows-specific keychain alternative mentioned
- Cloud runner feature still coming (not yet available)
- Team collaboration requires setting up shared .agents/ repo
- May be overkill for users who only use one coding agent
- Configuration layering complexity can be confusing
- No built-in GUI, entirely CLI-based
Frequently asked questions about agents-cli
What is Agents CLI?
Agents CLI is the missing toolchain for CLI coding agents that lets you run powerful agents at scale. It provides a unified interface to run any model through any CLI, rotate across accounts to avoid rate limits, chain agents in pipelines, pin versions per project, run parallel agents with DAG dependencies, search AI history, sync skills/MCP/commands across all agents, manage unified config, use a local browser via CDP, schedule cron routines, and store secrets in Keychain.
Which AI coding agents does it support?
Agents CLI supports Claude Code, Codex, Gemini CLI, Cursor, OpenCode, Copilot, Amp, Goose, Roo, Kiro, and OpenClaw - over 11 different coding agents from a single CLI interface.
How do I install Agents CLI?
You can install via curl with 'curl -fsSL agents-cli.sh | sh', via bun with 'bun install -g @phnx-labs/agents-cli', or via npm with 'npm install -g @phnx-labs/agents-cli'. It is also available as the alias 'ag'.
How does config sync work across harnesses?
~/.agents/ is the canonical config source. Write commands as markdown, rules as AGENTS.md, and hooks as scripts. Agents CLI syncs them into each harness's native format: markdown for Claude and Gemini, TOML for Codex, .cursorrules for Cursor. The config includes commands/, skills/, mcp/, hooks/, memory/, and permissions/ directories.