Overview
Permissions
Capabilities
Network Operations
Go back to agents list
prediction-finder
An AI Agent that uses deepsearch and twitter APIs to fetch predictions.
Prediction Finder Documentation
Overview
The Prediction Finder is a core component of the Prediction Swarm Agents ecosystem that discovers and identifies explicit and implicit predictions across social media, with a primary focus on X/Twitter platform. It uses advanced Natural Language Processing (NLP) to distinguish genuine predictions from general commentary.
Core Functionality
The Prediction Finder identifies predictions by:
- Detecting explicitly stated predictions (direct statements about future events)
- Discovering implicitly phrased predictions (indirect references to future outcomes)
- Filtering results for relevance with confidence scoring
- Categorizing predictions by timeframe and conditions
Command Line Interface
Basic Usage
# Find predictions on a specific topic
bun run find-predictions --topic "<topic>" [--limit <number>] [--output <file>]
Examples
# Find Tesla stock predictions with default limit
bun run find-predictions --topic "Tesla stock price" --limit 20 --output predictions.json
# Find cryptocurrency predictions
bun run find-predictions --topic "Bitcoin price" --limit 50
# Find political predictions
bun run find-predictions --topic "2024 election results" --output election-predictions.json
Parameters
--topic(required): The topic to search for predictions about--limit(optional): Maximum number of predictions to find (default varies)--output(optional): File path to save results in JSON format
Output Format
The Prediction Finder returns predictions in the following JSON structure:
{
"prediction_text": "Tesla stock will reach $300 by end of 2023",
"confidence_score": 0.87,
"implicit": false,
"topic_relevance": 0.95,
"timeframe": "months",
"has_condition": false,
"post_id": "1234567890",
"post_url": "https://x.com/username/status/1234567890",
"author_username": "elonmusk",
"author_name": "Elon Musk",
"post_date": "2023-01-15T14:30:00.000Z",
"topic": "Tesla stock price"
}
Field Descriptions
- prediction_text: The extracted prediction statement
- confidence_score: Algorithm's confidence that this is a genuine prediction (0.0-1.0)
- implicit: Boolean indicating if the prediction was stated indirectly
- topic_relevance: How relevant the prediction is to the searched topic (0.0-1.0)
- timeframe: Predicted timeframe category (immediate, days, weeks, months, years, decades)
- has_condition: Whether the prediction includes conditional elements ("if X then Y")
- post_id: Unique identifier of the social media post
- post_url: Direct link to the original post
- author_username: Username of the prediction maker
- author_name: Display name of the prediction maker
- post_date: Timestamp when the prediction was posted
- topic: The topic that was searched for
Technical Requirements
- Node.js 22+
- Bun runtime
- 4GB RAM minimum
- API keys for:
- Language models (for NL