Beyond Claude and Codex, Vanta CLI drives any OpenAI-compatible host through Codex's provider config — OpenRouter, Groq, Together, or a local Ollama server. A registry resolves a provider name to a base URL and key, and a loopback Chat→Responses bridge papers over the API differences so they all behave like Codex-proper.
Name a host on --provider and it resolves to the openai engine pointed at that host's base URL and key variable. OpenRouter, Groq, Together, and local are built in; add your own with one entry in the defaults file.
Modern Codex speaks only the OpenAI Responses API, but most third-party hosts expose only /chat/completions. Vanta CLI starts a local bridge: Codex talks Responses to localhost; the bridge forwards Chat Completions to the real upstream with the right key and headers. Hosts that serve Responses natively can opt out.
Tool-calling quality varies by model — route to a frontier or coding-tuned model and the harness behaves like Codex-proper. Vanta CLI warns, never blocks, on an unrecognized model.
A first-party agent CLI behind one REPL and contract. Swap Claude for Codex mid-session; Vanta drives it exactly like it drives everything else.
ExploreClaude runs in-process via the Agent SDK; OpenAI Codex runs as a subprocess. Same REPL, same tools, same cards — switch mid-session without losing the thread.
ExploreOne LLM plans the work, spawns agents, reviews their diffs, and merges. You talk to it like a tech lead.
ExploreVanta Studio is launching soon. Join the waitlist and start directing a team.