Agent Settings

Agent settings let you control which AI model powers each agent and add custom instructions that shape how they work. This is the place to go when you want the AI to follow your specific guidelines or when you want to experiment with different models.

Choosing an AI Model

Each agent in StarZero Agents can be powered by a different AI model. You can change the model from the Customize → Agents page for persistent defaults, or switch models mid-conversation using the Chat Settings popover (the gear icon next to the message input). Changes made in the popover apply to the current session immediately.

You'll find options from several providers:

• Google Gemini — Google's AI models
• Anthropic Claude — Anthropic's AI models
• OpenAI GPT — OpenAI's models
• xAI — xAI's Grok models
• DeepSeek — DeepSeek's models
• MoonShotAI — MoonShotAI's models (Kimi)
• Qwen — Alibaba's Qwen models
• Zhipu (GLM) — Zhipu AI's models
• OpenRouter — Access to additional models from various providers

Different models have different characteristics. Some are faster, some are more creative, and some are better at precise analytical work. The default models have been chosen to work well for each agent's role, but you're free to experiment. The Motion Graphics Agent has its own shorter list of models suited to design work.

Model Comparison Stats

When browsing models in the selector, each model shows two rows of icons under its name to help you compare:

• Intelligence (graduation caps) — How capable the model is at complex, multi-step tasks, rated at the effort level StarZero Agents actually runs it at. Based on independent industry benchmarks.
• Cost (coins) — How much the model typically costs to finish a whole task, not just its price per word. A model with a higher list price can still cost less overall if it gets the job done with less back-and-forth. More coins means more expensive.

Each row has five icons that fill from left to right, including partial fills. The fill level is relative — a model with 4 caps of intelligence is among the most capable currently available, while 2 is average. Hover over a row to see which stat it shows. If a model hasn't been benchmarked yet, that row appears struck through, meaning the rating is unknown.

Not sure which model to pick? Stick with the defaults. They've been selected to balance quality and speed for each agent's specific job.

When a Model Is Retired

The model list changes over time as providers release new models and retire old ones. If an agent was set to a model that's no longer available, it switches back to its default model automatically — in your saved settings and in older chats you reopen. You can pick a different model any time from the selector.

Adding Custom Rules

Custom rules are standing instructions written in plain English that an agent follows in every conversation. Think of them as permanent notes to your production team.

How to Add Rules

1. Open Customize → Agents tab
2. Select the agent you want to configure
3. Type your rule in the custom rules field
4. Save your changes

Example Rules by Agent

The Producer:

• "Always default to 9:16 aspect ratio unless I specify otherwise"
• "Use word-by-word captions on every project"
• "Keep cutdowns under 60 seconds unless I ask for longer"

Search Agent:

• "Always include timestamps in search results"
• "Prefer clips that are at least 5 seconds long"
• "Exclude clips from the intro and outro of episodes"

Analysis Agent:

• "Keep summaries under 200 words"
• "Always identify speakers by name when quoting"
• "Focus on actionable insights over general observations"

Motion Graphics Agent:

• "Use the brand colours #0A84FF and #111111 and the Inter font"
• "Keep lower-thirds in the bottom-left corner, about half the frame width"
• "Prefer quick, subtle motion over bouncy easing"

Critique Agent:

• "Pay special attention to caption timing accuracy"
• "Flag any silent gaps longer than 1 second"
• "Check that the first 3 seconds have strong visual engagement"

Tips for Writing Rules

• Be specific. "Always use captions" is clear. "Make it good" is not helpful.
• Keep rules general. Rules apply to every conversation, so make sure they're broadly useful. For one-off instructions, just include them in your chat message.
• Don't overload. A few clear rules are better than a long list. Too many rules can conflict with each other or slow things down.

Critique Settings

The Critique Agent can review your work at different points. Choose the option that fits how you like to work:

• Inline QC — The Critique Agent reviews work as it happens during editing. You get feedback in real time, which is great for interactive, hands-on sessions.
• Task Agent — Quality review happens after the Task Agent finishes a multi-step workflow. Good for autonomous tasks where you want a final check before you review.
• Disabled — No automatic quality reviews. Choose this if you prefer to evaluate everything yourself.

YOLO Mode

YOLO Mode lets agents act autonomously by removing confirmation prompts. You can toggle it from the Customize → Agents page, or mid-conversation using the Manual / Auto switch in the Chat Settings popover (the gear icon next to the message input). When enabled:

• Image and video generation is auto-approved — The AI can start generating images and videos without asking you first. Since generations spend credits, keep an eye on your usage while YOLO Mode is on.
• Library-wide scans are auto-approved — The AI can analyze all or part of your library without asking. For large libraries, this can use significant credits since each item is processed by AI.

YOLO Mode is off by default. Turn it on only when you're comfortable with the AI acting on your behalf without pausing for confirmation.

Resetting to Defaults

If your customizations aren't working out, you can always clear your custom rules and switch back to the default models. There's no risk in experimenting — you can always go back to where you started.