Supported agents
The agent CLIs Open Free Max drives — Claude Code, Codex, Antigravity, Cursor Agent, OpenCode, Vibe and Kimi Code — and how provider support is structured.
OFM is bring-your-own-agent. It drives the official CLIs you already have installed and logged in; it never bundles, updates or proxies them.
Supported CLIs
| CLI | Provider | Notes |
|---|---|---|
| Claude Code | Anthropic | Full session resume by id, naming, fork |
| Codex | OpenAI | Interactive CLI |
| Antigravity | Interactive CLI (agy) |
|
| Cursor Agent | Cursor | Interactive CLI |
| OpenCode | Open source | Interactive agent CLI |
| Vibe | Mistral | Interactive CLI |
| Kimi Code | Moonshot AI | Interactive CLI |
Provider adapters
Each provider is described by a small declarative adapter that knows how to:
- launch the binary in interactive mode (with the right session/resume flags);
- apply your settings (model, effort, permission mode, allowed tools, MCP config, …) as launch flags — configured once in Settings;
- read its screen to detect state and confirmation prompts.
Adding a new CLI is one adapter — the rest of the app is provider-agnostic.
Autonomy levels (A / B)
Not every CLI exposes its confirmation prompts identically, so OFM grades providers:
- Level A — driven (Claude Code): OFM reliably classifies its screen and confirmation boxes, so it can fully auto-resolve and run autonomously per your rules.
- Level B — assisted (others, initially): OFM drives them exactly the same way, but their confirmations always go to the human queue rather than being auto-answered blindly. No CLI is ever excluded; a level-B provider just asks for a hand more often.
The level is shown next to the CLI when you launch a session, so you know up front whether it will run hands-off or surface confirmations.
Authentication
You sign in where you already do — inside each official CLI. OFM never sees your tokens or keys. See Subscription vs API.