Back to .md Directory

Pitch Video Script -- Predix

Provides a timed, scene-by-scene script for a 4.5-minute pitch video demoing an agent-native prediction market on Stacks.

May 2, 2026
0 downloads
1 views
ai agent mcp claude cursor
View source

What this file does

Provides a timed, scene-by-scene script for a 4.5-minute pitch video demoing an agent-native prediction market on Stacks.

When to use it

  • You are recording a product demo video for a DeFi or agent platform
  • You need a structured script with timing, screen actions, and voiceover
  • You want to show both human and AI agent trading in one video
  • You are pitching a Bitcoin-layer-2 project to developers or investors

Assumes this stack

StacksClarityPyth NetworkVercelUpstash RedisMCP

Pitch Video Script -- Predix

Target duration: 4 minutes 30 seconds Format: Screen recording with voiceover (no face camera required, but recommended for intro/outro)


SCENE 1: Hook (0:00 -- 0:30)

[Screen: Predix landing page, dark theme, BTC price chart animating in real time]

Voiceover:

"What happens when you give AI agents direct access to a prediction market on Bitcoin?

This is Predix -- the first agent-native prediction market built on Stacks. Users and AI agents trade 1-minute BTC price predictions with zero gas fees and fully automated on-chain settlement.

Let me show you how it works."

[Action: Briefly hover over the UP/DOWN buttons and countdown timer to draw attention to the interface.]


SCENE 2: The Problem (0:30 -- 1:00)

[Screen: Simple slide or text overlay on screen -- three bullet points appearing one by one]

Voiceover:

"Prediction markets today have three core problems.

First, they are slow. Rounds last hours or days. By the time you get a result, you have lost interest.

Second, they are expensive. Every bet costs gas. For small trades, gas can eat a significant portion of your position.

Third, they are human-only. There is no standard way for an AI agent to discover a market, authenticate, and trade autonomously.

Predix solves all three."


SCENE 3: Live Demo -- Placing a Bet (1:00 -- 2:00)

[Screen: Back to Predix live at predix.live. Show a round in progress.]

Voiceover:

"Every round lasts 60 seconds. You see the live BTC price from the Pyth oracle, the countdown timer, and the current pool sizes for UP and DOWN."

[Action: Click on UP, enter an amount (e.g., 5 USD), confirm in Xverse wallet.]

"I am placing a 5 dollar bet on UP. Notice that the gas fee is zero -- the platform sponsors every transaction. The user just signs and confirms."

[Action: Show the bet appearing in the pool. Wait for the countdown to reach zero.]

"When the round ends, the settlement cron fetches the close price from Pyth Benchmarks, validates it against the circuit breaker, and calls resolve-and-distribute on-chain. Winners receive their payout automatically. There is nothing to claim."

[Action: Show the result -- win or loss appearing on screen. If a win, show the confetti animation.]

"That is the entire flow. Sixty seconds, zero gas, automatic payout."


SCENE 4: Agent Platform (2:00 -- 3:00)

[Screen: Split between code editor (showing SDK code) and the Predix agent leaderboard page]

Voiceover:

"But Predix is not just for humans. It is agent-native from the ground up.

AI agents can integrate through four paths: a REST API with full OpenAPI specification, an MCP server for Claude and Cursor, a TypeScript SDK, and a Python SDK with LangChain support."

[Action: Show the MCP config in a code editor -- the JSON block with npx @predix/mcp.]

"Here is an MCP configuration for Claude Desktop. Add six lines of JSON and Claude can read the market, analyze opportunities, and place bets autonomously."

[Action: Switch to a terminal or Claude Desktop showing an agent calling predix_market and predix_place_bet.]

"The agent calls predix_market to check the current round. It evaluates the odds. It calls predix_place_bet. The platform builds an unsigned transaction, the agent signs locally -- its private key never leaves the machine -- and the platform sponsors and broadcasts."

[Action: Show the agent leaderboard at predix.live/agents.]

"Agents compete on a public leaderboard alongside human traders. You can see their win rate, P&L, ROI, and total volume. This is the first prediction market where AI and humans trade on equal terms."


SCENE 5: Technical Architecture (3:00 -- 3:45)

[Screen: Architecture diagram from the README or ARCHITECTURE.md -- the ASCII diagram or a cleaner visual version]

Voiceover:

"Under the hood, Predix is built entirely on Stacks and finalized on Bitcoin.

The smart contracts are written in Clarity. The main market contract uses a gateway-only pattern -- all calls route through a thin proxy that enforces access control and round validation. The gateway and sponsor addresses are timelocked, requiring 144 blocks -- about 24 hours -- to change.

Settlement uses Pyth Network as the oracle. A two-layer circuit breaker validates prices both server-side and on-chain, rejecting any settlement that shows signs of manipulation or stale data.

The entire platform runs serverless on Vercel, with Upstash Redis for optimistic state, nonce tracking, and agent key management."

[Action: Briefly show the contracts folder in the IDE -- predixv7.clar, gatewayv6.clar, test-usdcx.clar.]

"The token is a standard SIP-010 implementation. Every piece of Predix leverages Stacks: Clarity for trust, sponsored transactions for UX, and post-Nakamoto block times for viable on-chain timing."


SCENE 6: Jackpot (3:45 -- 4:00)

[Screen: Jackpot page at predix.live/jackpot]

Voiceover:

"One percent of all fees stay in an on-chain jackpot treasury. Every day at 9 PM Eastern, a Bitcoin block hash is used as verifiable randomness to draw a winner. Early and large bettors earn multiplied tickets, incentivizing active participation."


SCENE 7: Closing (4:00 -- 4:30)

[Screen: Predix landing page with the logo centered. Links to GitHub, docs, and live demo below.]

Voiceover:

"Predix is a production-grade prediction market with a complete agent platform -- REST API, MCP server, two SDKs, webhooks, and discovery manifests.

It is live on Stacks testnet at predix.live. The GitHub repo includes full documentation, smart contract source, and the agent SDK packages.

We built this to prove that DeFi on Bitcoin can be fast, frictionless, and agent-ready. This is Predix. Thank you."


Recording Notes

Technical setup:

  • Record at 1080p minimum (1920x1080).
  • Use a clean browser window (no bookmarks bar, no other tabs visible).
  • Dark mode for both the browser and code editor to match the Predix theme.
  • Record audio separately if possible for cleaner sound.

Key moments to capture:

  1. The countdown timer reaching zero and the round resolving (Scene 3).
  2. The wallet popup showing zero gas fee (Scene 3).
  3. The MCP config JSON and an agent placing a bet (Scene 4).
  4. The agent leaderboard with real data (Scene 4).
  5. The jackpot treasury balance and draw history (Scene 6).

Pacing:

  • Do not rush. Let the UI breathe between transitions.
  • Pause for 1-2 seconds after making a point before moving to the next topic.
  • Keep cursor movements deliberate. Avoid aimless mouse movement.

If a round does not resolve during recording:

  • Pre-record a round resolving and splice it in during editing.
  • Alternatively, use the history page to show a resolved round.

Fallback for agent demo:

  • If the MCP server is not configured during recording, show the TypeScript SDK code running in a terminal instead.
  • The key point is demonstrating programmatic agent access with local signing.

What's inside

7 scenes with timestamps, screen actions, voiceover lines, and recording notes including fallback plans

Change this for your project

  • Replace predix.live with your own demo URL
  • Replace @predix/mcp with your own MCP package name
  • Replace predixv7.clar, gatewayv6.clar, test-usdcx.clar with your contract filenames
  • Replace 9 PM Eastern with your own jackpot draw time

Where it goes

Keep it in your repository where the agent or team that needs it will read it.

Worth borrowing

  • Split video into problem, live demo, agent platform, architecture, and closing for clear narrative flow
  • Include fallback plans for demo failures (pre-recorded clips, alternative agent paths)
  • List specific key moments to capture on screen so the editor knows what to highlight

Related Documents