# Agent Harnesses `FLARE` supports three agent harnesses. All three implement the {class}`~milp_flare.harness.base.Harness` interface. | Harness | Models | Billing | Credential | | --- | --- | --- | --- | | [Claude Code](claude_code.md) | Anthropic Claude | Claude subscription | `CLAUDE_CODE_OAUTH_TOKEN` (OAuth) | | [Codex](codex.md) | OpenAI GPT | ChatGPT subscription | `~/.codex` (OAuth) | | [OpenCode](opencode.md) | **All** Providers | Provider API | `_API_KEY` | Each harness page covers: 1. **Choosing a plan** (or API account) — what to sign up for and where to track usage. 2. **Authentication** — the one-time setup step that allows `FLARE` to invoke the agent. 3. **Using the harness** — configuring the harness in Python for use by `FLARE`. 4. **Cost tracking** — how to track the cost of agent usage. ```{toctree} :maxdepth: 1 :hidden: claude_code codex opencode ```