7 best agent harnesses in 2026

A model on its own answers questions. To do work, it needs something around it that reads files, runs tools, asks for approval, and keeps going until the job is done. That something is the harness, and it matters as much as the model.

What is an agent harness?

Anthropic's engineering blog defines it this way: "An agent harness (or scaffold) is the system that enables a model to act as an agent." OpenAI describes the work of handling context, tools, and human approval, and says "that surrounding execution system is the harness."

In practice a harness does four things:

• It gives the model tools, such as reading and writing files, running commands, and searching the web.
• It manages what the model sees, by choosing which files and earlier messages to send.
• It decides what needs your approval.
• It loops. The model acts, sees the result, and acts again.

There are two layers. Some harnesses talk to the model directly, such as Claude Code, Codex, and OpenCode. Others sit on top of those and add an interface, such as T3 Code, Conductor, and Mimzie. The second kind matters for cost, because a harness that runs the vendor's own agent can use your subscription. Which apps can use your Claude or ChatGPT subscription? explains why.

The 7 best agent harnesses

1. Claude Code
2. OpenAI Codex
3. OpenCode
4. Cursor
5. T3 Code
6. Conductor
7. Mimzie

1. Claude Code

• Best for: Developers who want Anthropic's own agent
• Pricing: Included with Claude Pro ($20 a month) and higher plans
• What we like: Skills, plugins, and subagents, on a flat monthly plan

Anthropic describes Claude Code as "an agentic coding tool that reads your codebase, edits files, runs commands, and integrates with your development tools. Available in your terminal, IDE, desktop app, and browser."

It runs on a Claude plan, so heavy use costs a flat monthly price. It runs Claude models only.

Claude Code pricing

Included with Pro at $20 a month, Max at $100, Ultra at $200, and Team and Enterprise plans. It is not in the Free plan. Source: Claude pricing.

2. OpenAI Codex

• Best for: Developers who pay for ChatGPT
• Pricing: Included in every ChatGPT plan, with limits on Free and Go
• What we like: One agent across the terminal, the desktop app, the IDE, and the cloud

Codex is OpenAI's agent. It runs in the ChatGPT desktop and mobile apps, on the web, in a command-line tool, in an IDE extension, and in the cloud. ChatGPT Work, OpenAI's agent for office tasks, shares its usage limits. The command-line tool needs Plus or higher.

OpenAI also runs a program, Sign in with ChatGPT, that lets partner apps draw on a subscriber's plan.

Codex pricing

Included in ChatGPT Free, Go ($8 a month), Plus ($20), Pro (from $100), Business, Edu, and Enterprise. Source: OpenAI's pricing docs.

3. OpenCode

• Best for: Developers who want an open-source agent that works with any model
• Pricing: Free. OpenCode Go has curated models from $10 a month
• What we like: MIT licensed, and it works with any model

OpenCode calls itself "the open source coding agent." It works with dozens of providers and can sign in with a ChatGPT plan or GitHub Copilot. It cannot use a Claude plan. Anthropic allows that only in its own agent, so Claude in OpenCode needs an API key.

Several other products are built on it, including OpenWork.

OpenCode pricing

The agent is free and open source. Go starts at $10 a month. Source: opencode.ai.

4. Cursor

• Best for: Developers who want the agent inside a code editor
• Pricing: Free, then from $8 a month
• What we like: The editor and the agent are one product

Cursor is a code editor with an agent built in. You see every change as a diff in the file, which is the same idea Mimzie applies to documents.

Cursor sells its own plans. A Cursor staff reply on its forum says signing in with a ChatGPT subscription "isn't supported."

Cursor pricing

• Free: $0
• Plus: $8 a month
• Premium: $30 a month
• Super: $100 a month
• Ultra: $200 a month
• Teams: from $30 per user a month

Source: Cursor pricing.

5. T3 Code

• Best for: Developers who use several agents and want one window for them
• Pricing: Free and open source
• What we like: It runs the agents you already have and handles no credentials

T3 Code sits on top of other harnesses. Its readme lists Claude Code, Codex, Cursor, Grok Build, OpenCode, and Google Antigravity. You install each provider's tool and log in through it, so each agent runs on your own plan. It has desktop, web, and mobile apps, and the readme calls it "very very early."

T3 Code pricing

Free, under the MIT license. Source: the T3 Code repository.

6. Conductor

• Best for: Developers on a Mac who run several agents at once
• Pricing: Free for now
• What we like: Each agent gets its own isolated copy of the code

Conductor is a Mac app that runs Claude Code, Codex, and Cursor agents in parallel, each in its own workspace, with review and merge built in. Its FAQ says "by default, Conductor uses the auth tokens already saved on your machine", so a Claude plan works.

Conductor pricing

Conductor's FAQ says it does not charge right now and plans to charge later for team features. Source: conductor.build.

7. Mimzie

• Best for: People whose work is documents, spreadsheets, and slides
• Pricing: Free
• What we like: The same agents developers use, with editors for Office files

Every other harness on this list is built for code. Mimzie is built for office work. It runs Claude Code, Codex, Antigravity, and Grok Build on your computer, each signed in with your own plan, and gives them editors for .docx, .xlsx, and .pptx files.

By default, changes show up as tracked edits in the document, and you accept or reject each one. Bots split larger jobs between them. It runs on macOS, Windows, and Linux.

Mimzie is in beta, and it has no scheduled tasks.

Mimzie pricing

Free. You bring the Claude, ChatGPT, Google, or Grok plan you already have. It also has its own agent with free models.

Which agent harness should you pick?