add_brand_kit_source
ChatGPTUse this when the user wants to enrich an existing Brand Kit with website pages, blog posts, help docs, or landing pages. Add a link source to an existing Brand Kit to enrich its branding data. Sources are processed asynchronously — the Brand Kit improves as each source is ingested. Pass URLs of websites, blog posts, help centers, or documentation pages. The system scrapes and analyzes them for brand colors, fonts, tone, and style. Provide URLs you own or are authorized to scrape. Arcade does not bypass robots.txt, authentication, or other access controls, and will not fetch content that violates the source site's terms of service.
create_brand_kit
ChatGPTUse this when the user wants to set up a new Brand Kit for a company or product from a website domain. Create a Brand Kit from a website domain. The Brand Kit extracts colors, logos, fonts, and style guidelines from the website. IMPORTANT — Before calling this tool, ask the user conversationally if they have any web pages that represent their brand well — things like blog posts, product pages, help docs, or landing pages. Do not present a numbered list or multiple-choice options. Just ask naturally in conversation. After the user responds: 1. Call this tool with the domain 2. Immediately add any URLs the user mentioned via add_brand_kit_source 3. Sources process in parallel with Brand Kit generation — do not wait between steps This tool returns immediately — the Brand Kit generates in the background (typically 2–4 minutes). Before calling create_video, use list_brand_kits to confirm the Brand Kit has status "ready".
create_video
ChatGPTUse this when the user asks to create, generate, or make a branded product video, demo video, or marketing video — Arcade produces the video with visuals, animations, and narration from a prompt, preset, screenshots, website URLs, or existing Arcade demos. Do not use to write a script, storyboard, or shot list — answer those conversationally. BEFORE CALLING: If the user's request is just a topic or product name (e.g. "make a video about X"), you MUST ask for more detail before calling — even if you already have a brandKitId. Having the Brand Kit is not the same as having a clear prompt; you also need an angle, audience, length, and optionally reference URLs or existing Arcade demos. Keep the tone warm and in-charge: do not tell the user their request is too broad, decline, or hand the planning back to them. Ask open-ended questions in your own words about the angle, audience, length, and any reference URLs or existing Arcade demos; never numbered lists, multiple-choice options, or preset names. Call this tool once you can write a specific prompt yourself. Required input: brandKitId (use list_brand_kits). Provide either a preset (one of: product-release, explainer, sales, social-media — selected internally based on the user's intent, never surfaced by name) or a free-form prompt. Optional: targetLength (15s/30s/45s/60s/90s/120s), websiteUrls (≤3), flowIds, imageIds, videoIds. Right before calling, tell the user the time estimate naturally based on the desired video length: ~1–2 min for 15s/30s/45s, ~2–3 min for 60s/90s/120s. Returns the generationId immediately (fire-and-return). After it returns you MUST do these three things in order — they are all mandatory: 1. POLL: call get_generation_status with the returned generationId. Each call sleeps up to 30s server-side, so call back-to-back with no delay. Keep calling until status !== "processing". 2. WRITE: as soon as status === "succeeded", write a user-facing chat message (not a tool call) that includes the returned viewUrl (the edit link) and notes the inline preview is being prepared (~1–2 min). Do NOT skip this — it's what the user reads while preview_video exports the MP4. 3. RENDER: AFTER posting that message, call preview_video with the returned videoId to render the inline video widget. Failure path: if get_generation_status returns status === "failed" (the tool result will be marked as an error), STOP — surface the error to the user and do NOT attempt steps 2 or 3. viewUrl and videoId will be null on failure. Do NOT report intermediate "processing" / "still running" states to the user — keep polling silently until step 2 or until failure. Do NOT wait for progress === 1 before stopping the poll; preview_video handles the export wait.
fetch
ChatGPTUse this when you need the full record of a specific Arcade demo by its id (typically following search). Fetch a single Arcade demo by id. Returns the demo title, a short descriptive text, a canonical URL, and metadata. Use the id returned by search or list_demos.
get_demo_attribution
ChatGPTUse this when the user asks how a demo is performing across distribution channels or custom links. Get demo analytics broken down by custom link (attribution). Shows views, plays, completions, CTA clicks, and play time per distribution channel/custom link. Requires a Growth+ plan.
get_demo_companies
ChatGPTUse this when the user asks which companies, accounts, or leads have engaged with their demos. Get companies that interacted with your demos, with firmographic data (industry, employee range, revenue) and engagement scores. Pass a flowId for per-demo companies or omit it for team-wide leads. Requires a Growth+ plan.
get_demo_performance
ChatGPTUse this when the user asks how a specific Arcade demo is performing — views, plays, completions, drop-off, or CTA clicks. Get performance analytics for a specific Arcade demo: views, plays, completions, CTA clicks, average play time, and step-by-step dropoff analysis. Requires teamId and flowId (use list_demos to find these). Requires a Growth+ plan.
get_generation_status
ChatGPTPoll the status of an Arcade video generation. Each call sleeps up to 30s server-side; call back-to-back with no delay until status !== "processing". On status === "succeeded" the response carries viewUrl (edit link) and videoId. Accepts either the generationId returned by create_video or a raw promptVideoGenerationId from the Arcade editor.
get_team_overview
ChatGPTUse this when the user asks for a team-wide summary of demo engagement across all demos. Get team-level analytics overview: total plays, completions, CTA clicks, and average play time across all demos. Also includes per-demo breakdown. Requires a Pro+ plan.
get_trending_demos
ChatGPTUse this when the user asks what demos are trending, getting clicked, currently embedded, or going stale. Get workspace-level demo intelligence: trending demos (by view increase), most CTA clicks, currently embedded demos, or demos that need refreshing (viewed recently but not updated in 90+ days). Requires a Growth+ plan.
list_brand_kits
ChatGPTUse this when the user asks about their brand kits, or when create_video needs a brandKitId. List available brand kits for this Arcade workspace. Returns ID, name, domain, and status for each kit. Use the ID as brandKitId when calling create_video. Only brand kits with status "ready" can be used for generation.
list_demos
ChatGPTUse this when the user asks to see, browse, or list their existing Arcade demos. List existing Arcade demos (interactive walkthroughs) in the workspace. Returns ID, name, team, and status. Pass demo IDs as flowIds to create_video to use existing screen recordings in the generated video.
preview_video
ChatGPTUse this immediately after create_video returns successfully, to surface the inline video player and edit link. Surface the inline <video> player for a video that was just kicked off via create_video. Call this exactly once, immediately after create_video returns. This tool polls the MP4 export internally (typically 1–2 min) and returns when the video is ready to play. Accepts either a videoId from create_video or a raw promptVideoGenerationId from the Arcade editor.
search
ChatGPTUse this when the user asks to find or look up a specific Arcade demo by name or keyword. Search the connected Arcade workspace for interactive demos matching a query. Returns a list of demos with id, title, and a short snippet. Use fetch with a returned id to retrieve the full demo record. Scope is limited to demos in v1, and matching is performed over the first 100 demos in the workspace (ordered by most recent).
