Claude vs. Codex: Choosing Your AI Agent Engine Inside Poko Motion

Most AI video tools pick a model for you and never mention it again. Poko Motion puts the choice in Settings, under something called the agent engine.

This isn't a cosmetic option. It changes which AI models show up in the picker, which subscriptions you can connect, and how billing works. Here's what actually happens when you switch, and how to decide which one to leave it on.

What "agent engine" means

The agent engine is the coding agent that reads your project, writes the script, builds the scenes, and applies every chat edit you type. Poko gives you two:

• Claude Code - the default, running on Poko's existing agent runtime
• Codex - OpenAI's coding agent, connected through Poko credits, ChatGPT, your own API key, or a gateway

Switching between them starts a fresh agent session. Your project files and chat history stay exactly where they were - you're changing who's doing the work, not starting the project over.

This is different from the model picker inside the chat box. The engine decides which family of models is even available to choose from; the model picker just lets you pick a specific one from that family for a given message.

Claude Code: the default path

Claude Code is what every new project uses unless you change it. It needs no setup and no keys.

If you switch to bring-your-own-key billing, you get two ways to connect:

• Claude subscription - sign in with your Claude Pro or Max account and skip API keys entirely
• Anthropic API key - paste a key if you'd rather pay per token

There's also an optional field for a Gemini API key, which unlocks Gemini models as an experimental extra alongside Claude.

Codex: the OpenAI path

Codex exists for people who specifically want GPT models, or who already have a ChatGPT plan they'd rather put to work. On Poko credits, it just runs. On bring-your-own-key, you get:

• ChatGPT connection - sign in and use the Codex access already included in your plan
• OpenAI API key - paste a key for direct, pay-per-token billing

An optional xAI key unlocks Grok models the same way Gemini does on the Claude Code side.

Poko credits vs. bring your own key

Every project starts on Poko credits: no keys, no connections, Poko routes the request for you. That's the whole setup, and it's fine to never touch this section at all.

Switching to bring-your-own-key changes who gets billed. Your provider - Anthropic, OpenAI, or whoever's behind your key - bills you directly, and your Poko credits aren't touched for agent or model usage.

That's a real advantage if you're already paying for a Claude or ChatGPT plan and don't want to double-pay for the same tokens through Poko credits on top of it.

There's one carve-out worth knowing: sound effects and background music still draw from Poko credits even in BYOK mode, since BYOK support for audio generation isn't live yet. Everything else - the script, the scenes, every chat edit - runs on whichever billing mode you chose.

The models you actually get

Whichever engine you're on determines the model list in the chat composer's picker:

• Claude Code unlocks Claude Fable 5, Opus 4.8, Sonnet 5, Sonnet 4.6, and Haiku 4.5, spanning Premium, Fast, and Instant tiers
• Codex unlocks the GPT-5.6 family (Sol, Terra, Luna) plus GPT-5.5, GPT-5.4, and GPT-5.4 Mini

You're not locked into one model per project either. You can switch models mid-conversation - use a Premium-tier model for the first build, then drop to a Fast or Instant tier for small chat edits that don't need as much reasoning.

Some models also support a thinking toggle for deeper reasoning on harder edits. A few run it automatically ("adaptive"), some let you switch it on or off per message, and instant-tier models skip it entirely to stay fast.

Bringing your own gateway

Pro and Agency plans can point either engine at a custom gateway instead of a named provider - OpenRouter, a self-hosted model, or infrastructure your team already runs.

The protocol matters here. Codex needs an endpoint that speaks OpenAI's Responses API. Claude Code needs one that speaks Anthropic's Messages API. A standard Ollama or LM Studio server, which typically only exposes Chat Completions, won't work as-is - you'd need a compatible proxy in front of it.

So which one should you actually pick?

For most projects, leave it on Claude Code. It's the default for a reason: zero setup, and it's what Poko's agent runtime was built around first.

Switch to Codex if any of these are true:

• You already have a ChatGPT plan and would rather use the Codex access bundled with it
• You specifically want GPT-family models for a project
• Your team's existing AI infrastructure is already built around OpenAI's API

Neither choice is permanent or risky to test. Switching engines doesn't touch your video, your scenes, or your chat history - it only changes which agent is listening the next time you type a message.

If you're not sure, that's the point: stay on the default, and treat the other engine as a lever you can pull later, not a decision you need to get right on day one.