Specialized Agents

Behind the scenes, StarZero Agents has a team of specialized agents that each handle a specific type of work. You don't need to call them directly — the Producer brings them in when needed. But understanding what each one does can help you get better results.

Content Search Agent

What it does: Finds clips in your video library based on specific, factual criteria.

Best for:

• Finding clips by visual elements ("sunset shots," "close-up of the product")
• Searching dialogue and transcripts ("when she says 'game changer'")
• Locating clips by keyword or topic ("clips about pricing strategy")
• Filtering by filename or video source

Not for: Subjective or interpretive queries like "the funniest moment" or "the most emotional part." Those go to the Analysis Agent.

The Search Agent is fast and efficient. It searches everything StarZero Agents knows about your library to find exactly what you're looking for. Results appear as clips you can preview and select.

Content Analysis Agent

What it does: Handles complex, interpretive questions about your content. This is the thinker of the team.

Best for:

• Summarizing episodes or conversations
• Extracting themes and key topics
• Comparing different clips or episodes
• Understanding narrative structure and flow
• Identifying emotional moments, tone shifts, and energy levels
• Answering subjective questions ("what's the most compelling moment?")

When a question hinges on a visual detail, the Analysis Agent can look directly at an image or a specific video frame — for example, reading text on a slide, checking what's on screen at a given moment, or confirming a product's color. It can also use web research when it needs background context — for example, looking up a guest's bio or fact-checking a claim mentioned in the video.

Task Agent

What it does: Executes structured, multi-step tasks autonomously. This is the workhorse that takes a complex job and runs with it.

Best for:

• Multi-step workflows (search, select, arrange, caption, render)
• Recipe execution (following step-by-step templates)
• Complex projects where you want the AI to work independently

When you use the /task command or launch a recipe, the Task Agent takes over. It creates a to-do list, works through each step, and manages the entire workflow. It can also call in the Critique Agent to review quality.

During task execution, the AI works independently — you can watch the progress through the to-do list in the chat. To stop a running task, use the stop button.

Transcript Fixing Agent

What it does: Corrects misheard or mistranscribed words in your video transcripts.

Best for:

• Fixing names that were transcribed incorrectly
• Correcting industry-specific terms the transcription got wrong
• Fixing any words that don't match what was actually said

Corrections are precise — it knows exactly which word to fix and when it appears in the video. It's fast and focused, handling corrections quickly without disturbing anything else in your project.

Critique Agent

What it does: Reviews rendered videos for quality issues. Think of it as your quality control reviewer who watches the final output with a critical eye.

Best for:

• Checking for cut-off words at clip boundaries
• Spotting flash frames or awkward transitions
• Evaluating pacing and flow
• Identifying framing issues
• Suggesting specific fixes with timestamps

The Critique Agent watches your rendered video and provides actionable feedback. Instead of vague comments like "the middle section is weak," it gives specific notes like "the word 'strategy' is cut off at 0:23 — extend the clip by 0.5 seconds."

The quality check is context-aware: speech-related warnings (cut words, mid-sentence cuts) only apply to clips whose audio is audible, while visual checks like flash frame detection run on every clip regardless of volume. This means muting a clip won't generate unnecessary speech warnings, but visual issues are always caught.

When inline QC is enabled, quality check results appear directly in the chat as a visual summary. Each issue is listed with its code (e.g., W0001) and a description of the problem, so you can quickly scan for anything that needs attention. If a warning is intentional, you can suppress it by code so it won't appear in future quality checks — see Suppressing Warnings for details.

You can configure when the Critique Agent runs: during editing (inline QC), after task completion, or disabled entirely. See Configuring Agent Behavior for details.

Seeing Agent Activity

When StarZero Agents delegates to a specialist, you'll see it in the chat as an expandable section. Click to expand and see:

• Which agent was called and why
• What the agent searched for, analyzed, or reviewed
• The results and recommendations

This transparency lets you understand how the AI team is working on your request and gives you the information you need to provide feedback or make decisions.