Choosing an Agent
Which AI agent and subscription plan should you use with Agent Integration Kit?
Not sure which agent to use? This guide breaks down every option by capability, cost, and what real users experience — so you can pick the right one for your workflow.
Quick Recommendation
We recommend Claude Code and Codex CLI. They're both excellent — Claude Opus 4.6 and GPT-5.3 Codex are close in capability, and most power users set up both and switch between them. Each outperforms the other on different tasks.
- Claude Code with Claude Max ($100 or $200/month) — best for complex multi-step Blueprint work
- Codex CLI — free even with a free ChatGPT account, generous usage limits, and GPT-5.3 Codex is exceptionally strong
Decision Flowchart
| Your Situation | Recommended Agent | Plan | Monthly Cost |
|---|---|---|---|
| Want to try it for free | Codex CLI | ChatGPT (free) | $0 |
| Best value paid option | Codex CLI | ChatGPT Plus ($20) | $20 |
| Want the best complex workflow performance | Claude Code | Claude Max ($100) | $100 |
| Maximum usage, no compromises | Claude Code + Codex CLI | Claude Max ($200) + ChatGPT Plus ($20) | $220 |
| Already have a ChatGPT Pro subscription | Codex CLI | ChatGPT Pro ($200) | $200 |
| Want to use multiple models flexibly | OpenRouter | Pay-per-token | Varies |
| Already have GitHub Copilot | Copilot CLI | Copilot subscription | Existing sub |
Agent-by-Agent Breakdown
Claude Code — Best for Complex Workflows
Claude Code with Opus 4.6 excels at complex, multi-step Unreal Engine work. Where it really shines over Codex is deep Blueprint editing, intricate animation systems, and long multi-file workflows where maintaining context is critical.
Strengths:
- Excels at complex Blueprint creation — handles node graphs, state machines, and multi-file workflows with the most consistency
- Understands Unreal Engine architecture deeply (GAS, motion matching, behavior trees, etc.)
- Excellent at following long, multi-step instructions without losing context
- Native ACP support — full in-editor chat with context attachments, tool calls, session history
Which Claude plan?
| Plan | Price | Usage | Verdict |
|---|---|---|---|
| Claude Pro ($20) | $20/month | ~5 queries per 5 hours | Not recommended. Runs out extremely fast with AIK — most users hit limits within minutes of real work. |
| Claude Max ($100) | $100/month | ~5x Pro usage | Good. Enough for daily development work. Best value tier. |
| Claude Max ($200) | $200/month | ~20x Pro usage | Best. Heavy users and teams. Equivalent to ~$2,000-3,000 of API usage. |
| API Key | Pay-per-token | Unlimited (pay as you go) | Expensive. A single complex session can cost $10-50+. Only recommended if you need no rate limits at all. |
Claude Pro ($20) is not enough for serious work. This is the most common disappointment users report. The rate limits are very low — you'll get about 5 queries every 5 hours. If you're choosing Claude, get the $100 plan minimum.
Setup: Claude Code Setup Guide →
Codex CLI — Best Free Option & Strong All-Rounder
OpenAI's Codex CLI with GPT-5.3 Codex is an exceptionally capable agent that works with any ChatGPT account — including the free tier. It gives you more usage than Claude Pro ($20) at zero cost, and the new GPT-5.3 Codex model is genuinely competitive with Claude Opus 4.6.
Strengths:
- Works with a free ChatGPT account — no subscription required to get started
- More generous usage limits than Claude at every price tier — even the free tier gives more usage than Claude Pro ($20)
- GPT-5.3 Codex model is very strong — outperforms Claude on certain task types (structured data, some C++ patterns, research)
- Good at Blueprint editing, C++ generation, and structured tasks
- Solid reasoning capabilities for debugging and research
- Native ACP support — works with in-editor chat
Which ChatGPT plan?
| Plan | Price | Verdict |
|---|---|---|
| ChatGPT Free | $0 | Great starting point. Free Codex usage — more generous than Claude Pro ($20). |
| ChatGPT Plus | $20/month | Best value. More usage, better models. The sweet spot for most users. |
| ChatGPT Pro | $200/month | Maximum usage. For heavy users who prefer OpenAI's ecosystem. |
Weaknesses:
- Can be slower than Claude — thinks longer before responding
- Claude Opus 4.6 still edges ahead on very complex multi-step Blueprint workflows and deep Unreal Engine architecture tasks
Best for: Getting started with AIK at zero cost. Many experienced users keep Codex set up alongside Claude and switch between them depending on the task.
Setup: Codex CLI Setup Guide →
Gemini CLI — Not Recommended
We do not recommend Gemini for use with AIK. Gemini is the worst-performing agent across all tasks — Blueprint editing, animation, multi-step workflows, and tool usage. If you're looking for a free option, use Codex CLI instead (free with a free ChatGPT account).
Gemini CLI has a free tier, but its performance with AIK is consistently poor compared to every other agent.
Weaknesses:
- Worst performance of all agents — struggles with nearly every AIK task type
- Frequently refuses to edit Blueprints entirely — tells you it "cannot open .uasset files" and generates C++ instead
- Does not use AIK tools reliably — often ignores available tools or uses them incorrectly
- Weak understanding of Unreal Engine graph-based workflows
- Multiple users have switched away from Gemini after trying it
Strengths:
- Free tier with a Google account
- Can handle basic text questions and code explanations
If you're considering Gemini because it's free: Codex CLI is also free (with a free ChatGPT account) and performs dramatically better at every task. There is no reason to choose Gemini over Codex.
Setup: Gemini CLI Setup Guide →
OpenRouter — Flexible Multi-Model Access
OpenRouter gives you access to dozens of AI models through a single API key. No CLI installation needed — just paste your key in settings.
Strengths:
- Quickest setup — API key only, no CLI to install
- Access to Claude, GPT, Gemini, Llama, and many other models
- Pay only for what you use
- Good for testing different models
Weaknesses:
- Pay-per-token pricing adds up quickly with complex workflows
- No ACP support — uses a built-in native client instead of the full agent protocol
- Token limits depend on your credit balance (you may hit 402 errors if balance is low)
Best for: Users who want to try multiple models, or quick one-off tasks where you don't want to set up a CLI.
Setup: OpenRouter Setup Guide →
Copilot CLI — For GitHub Users
If you already have a GitHub Copilot subscription, you can use it with AIK through the Copilot CLI.
Strengths:
- Uses your existing GitHub Copilot subscription
- Native ACP support with in-editor chat
- Easy setup if you already have Copilot
Weaknesses:
- Smaller community of users testing it with AIK — less real-world feedback
- Performance varies depending on the underlying model Copilot routes to
Best for: Users who already pay for GitHub Copilot and want to avoid additional subscriptions.
Setup: Copilot CLI Setup Guide →
Cursor — Not Recommended for AIK
Cursor is a popular AI-powered IDE, but it has significant limitations when used with Agent Integration Kit.
Strengths:
- Can access AIK tools via MCP
Weaknesses:
- No ACP support — cannot use the in-editor chat window, context attachments, or session history
- Frequently refuses to create Blueprints autonomously — tells you to "do it manually" or suggests Python scripts instead
- You cannot control how Cursor prompts the underlying model, so it often doesn't use AIK tools effectively
- Must use Cursor's own UI instead of the Unreal Editor chat
Several users have reported frustration with Cursor refusing to create Blueprints, saying things like "edit_graph cannot directly create component function calls." This is a Cursor prompting limitation — the same tools work perfectly when used through Claude Code.
Best for: Users who already use Cursor as their IDE and want basic MCP tool access. For full AIK capabilities, use Claude Code instead.
Setup: Cursor Setup Guide →
Common Questions
Can I switch agents mid-project?
Yes. You can switch agents at any time from the chat window dropdown. Your project files and assets remain the same — the agent is just the AI that reads and modifies them.
Can I use multiple agents?
Yes — and we recommend it. Claude Opus 4.6 and GPT-5.3 Codex are close in overall capability, but each outperforms the other in different areas. Many power users set up both Claude Code and Codex CLI, then switch between them depending on the task. You can switch agents at any time from the chat window dropdown.
Do I need an internet connection?
The plugin itself is fully local — nothing is uploaded to Betide Studio's servers. However, cloud-based AI providers (Claude, ChatGPT, Gemini) require an internet connection. For fully offline use, you can run a local LLM through OpenCode with Ollama, though results will vary depending on the model.
I have Claude Pro ($20). Is it worth upgrading?
If you're using AIK regularly, yes. Claude Pro's limits (~5 queries per 5 hours) mean you'll spend more time waiting than working. The $100 Claude Max plan is the most common recommendation from active users.
Is the API more cost-effective than a subscription?
No. API pricing for models like Claude Opus is extremely expensive — a single complex session can cost $10-50+. The $200/month Claude Max subscription gives you roughly $2,000-3,000 worth of API calls. Always use a subscription plan over API credits.