Agent Guide

The Poolday agent is a general-purpose video editing system that can handle a wide range of tasks autonomously. Here are some examples of what it can do: • Edit and assemble video compositions from raw footage • Generate new visual assets using AI models (images, video clips, effects) • Add and sync audio, music, and voiceovers • Create and animate text, captions, and graphics • Apply transitions, effects, and color grading • Call one or multiple genAI models on one or multiple files (AI UGC, lipsync, background swap, etc.) • Translate and localize content across languages • Analyze footage and make creative decisions • Crop or resize footage into any target aspect ratio • Choose and cut footage based on user preferences • And more... The agent coordinates multiple specialized sub-agents, each responsible for a specific capability. Together, they plan, execute, and refine edits with minimal human input. You can chat and interact with the agent when unsure. You can ask it for any sort of help. It has full knowledge of what it can do and cannot do and can give you tips.

Skills are knowledge you teach the agent about your specific preferences. They help the agent understand what you mean when instructions are subjective. For example, a prompt like "Pick the best moments from this video" is subjective. "Best" means different things to different people: • For a lifestyle creator: moments with smiling faces, laughter, and dynamic movement • For a gaming channel: scenes with many characters on screen, explosions, and intense action • For a podcast: only the guest's answers, skipping the interviewer's questions By creating a skill, you teach the agent what "best" means for you. Once the skill is created, your prompt can simply say: "Pick the best moments from my video" The agent will apply your criteria automatically. How to create a skill: Describe what you want the agent to remember. Be specific about your preferences, style, and any constraints. Example: "Create a skill for picking highlights. I want moments with high energy: people laughing, fast movement, or dramatic reactions. Skip anything slow or with long pauses." Skills are saved to your account and can be updated or refined over time.

Questions Mode lets you encourage the agent to ask you questions before starting work. When enabled: The agent will ask clarifying questions before executing your request. This is useful when: • Your request has multiple valid interpretations • You want more control over creative direction • You're working on something new and want to iterate on the approach When disabled: The agent will make reasonable assumptions and start working immediately. This is faster when you trust the agent's judgment or have given clear instructions. We highly recommend using Questions Mode early on. It's a great way to learn how the agent thinks and what information it needs to produce the best results. You can toggle Questions Mode on or off at any time during a conversation.

The problem with most AI tools is that if you don't like the output, you have to re-prompt and regenerate the entire video. Poolday is built differently. The agent builds controls just for you based on your requests, so you can refine and tweak specific parts without regenerating everything. Examples of generative UI: • Image grids to select from multiple generated options • Sliders to adjust parameters (speed, intensity, duration) • Color pickers for brand colors • Before/after comparisons • Timeline previews These controls appear contextually based on what you're working on. You can interact with them directly instead of describing changes in text. You can also instruct the agent to build specific controls, either directly in your prompt or as part of a skill. For example: "Always show me a slider to adjust video speed before rendering." The agent learns from your selections and will adapt its suggestions over time.

Some tasks take time. We prioritize quality and accuracy over speed, which means the agent may take up to an hour (or more) for certain tasks. The agent can work in the background while you do other things. How it works: • Start a task and the agent begins processing • You can close the tab or work on other projects • The agent continues running on our servers • Return to review and refine the output Typical durations: • Simple edits: ~10 minutes • Medium complexity edits (most common): ~45 minutes • Complex tasks: up to 2 hours You don't need to watch the agent work. Come back when it's done and continue the conversation to make adjustments. You can also start multiple conversations in parallel. Each agent works independently, so you can have several tasks running at the same time.

The agent self-QAs its work. Before delivering results, it reviews the output to catch errors and ensure quality. In addition to this built-in QA, you can create a custom skill (see the Skills section) to deepen the level of QA and specify exactly what the agent should check for. Examples of custom QA skills: • Check if there's any alcohol visible in the video • Flag any swear words or inappropriate language • Verify brand colors are used correctly • Ensure captions are properly synchronized • Confirm no copyrighted music is present By combining built-in QA with custom skills, you get thorough quality control tailored to your specific requirements.

The agent can work with multiple inputs and produce multiple outputs in a single conversation, with a single prompt. Multiple inputs: • Upload several video clips, images, or audio files • Reference multiple assets in your prompt • The agent understands relationships between files Multiple outputs: • Request variations (e.g., "Create 5 versions with different hooks") • Generate content for multiple platforms (Instagram, TikTok, YouTube) • Produce a series of related videos from one prompt Example: "Here are 10 product photos. Create a 15-second video for each one, using my brand style skill. Export in 9:16 for TikTok and 1:1 for Instagram." The agent will process all inputs and deliver all outputs, organizing them clearly for review.

Slash commands are shortcuts for prompts you use often. Once you find prompts you use repeatedly, you can create a slash command for them. Instead of typing the full prompt each time, just type /your-shortcut-name and the agent will execute it. Example: Instead of typing "Add my brand captions with coral highlights and Montserrat Bold" every time, create a slash command called /captions and use that instead. Slash commands can also be generalist. Instead of hardcoding everything, you can make a slash command that keeps the core prompt but asks for variables like assets or parameters each time. Just ask the agent to create a slash command and describe what should be fixed vs. what should be flexible. Type "/" to see your available commands.

Every project you create can be reused as a template for future work. How to reuse a composition: 1. Type @ in your prompt 2. Click on Compositions 3. Select the composition you want to reuse 4. In your prompt, specify what you'd like to adjust from the original What gets preserved: • Timeline structure and timing • Effects and transitions • Text styles and animations • Audio placement and levels What you can change: • Individual clips and images • Text content • Colors and branding • Duration adjustments This is powerful for creating consistent content at scale. Build once, reuse many times.

If you ask the agent to do something that isn't physically possible, it won't be able to do it. The agent relies on existing AI models, so if no model exists to perform a task, the agent can't either. Examples of what's not supported: • Removing burned-in captions: If you upload a video with captions already baked into the pixels, the agent cannot remove them. No AI model can reliably extract and erase burned-in text. To translate captions, you need to provide the original video without captions, or a separate subtitle file. • Generating gameplay with accuracy: The agent cannot create realistic, accurate gameplay footage. Game visuals require precise physics, mechanics, and real-time rendering that generative models can't replicate faithfully. • Downloading assets from the web: The agent cannot browse the internet or download files from URLs. All assets must be uploaded directly by you. • Crawling websites for information: The agent cannot crawl or scrape websites to gather information about your brand, product, or any other content. If you need the agent to know specific details, you must provide them directly in your prompt or through uploaded files. When in doubt, ask the agent whether something is possible before starting a project.