Skip to main content

Agent setups

How to set up common coding agents. Start with the guide for your level, then check the per-agent notes for specifics.

Each setup links its sources.Updated June 2026

12 setups

Your first coding agent. Pick one and learn it.

Setup guide

One subscription agent and a short rules file. No MCP yet.

Beginner

You use one agent and want more from it.

Setup guide

Add a few tools, save repeated workflows, and add a review step.

Intermediate

Multiple agents, scheduled runs, and gates you trust.

Setup guide

Orchestrate workers, route models by cost, and verify before merge.

Advanced

Code stays on your machine. No tokens leave the box.

Setup guide

An open model served locally, driven by a local-first agent.

Local-only

Agents that run on a schedule or on events, not just when you type.

Setup guide

Scheduled runs, hooks, and a message channel wired into the agent.

Automation

Hermes

Local operator loop

A local loop: one model scopes a task, Claude Code builds it, audits gate the result, and every run ends with a summary you can parse.

Multi-agent orchestratorIn your stack

Claude Code

Anthropic

Terminal agent with long context, subagent fan-out, and computer use.

Terminal agent

Cursor — Agent Mode

Anysphere

IDE-native agent with codebase indexing and parallel background agents.

IDE agent

GitHub Copilot — Coding Agent

GitHub · Microsoft

Assign an issue, get a pull request with tests and a self-review.

Cloud / async

OpenCode

sst / Anomaly

Open, model-agnostic terminal coding agent — plan, then build, on the provider you choose.

Open-source terminal

Aider

Open source

Terminal pair-programmer that commits every change atomically and works with any LLM.

Git-native pair

OpenClaw

Open source · self-hosted

A self-hosted agent you run on your own box. Personality and memory live in plain Markdown files; it talks to you over chat apps.

Personal agent runtime