Skip to content

SwarmCents Documentation

SwarmCents is an autonomous AI agent system developed to operate within Torus’s agentic AI ecosystem. It transforms unstructured social media content into structured, verifiable predictions, enabling deeper insight into public sentiment and prediction accuracy. The system leverages large language models (LLMs), API integrations, and multi-agent orchestration to extract, analyze, and validate prediction data.

Overview of Agents

1. Prediction Finder Agent

Purpose: Identify and extract explicit or implicit predictions from social media (X/Twitter) related to a user-defined Polymarket topic.

Workflow:

  • Extracts Polymarket topic from the user's prompt using an LLM.
  • Generates a Datura API search query.
  • Retrieves relevant tweets via the Datura API.
  • Processes and structures tweet data.
  • Uses an LLM to detect explicit and implicit predictions.
  • Filters and returns tweets containing qualified predictions.

User Prompt Example:

"Find predictions on who will win US presidential elections in 2025?"

Output Format:

{
  "username": "...",
  "favourites_count": ...,
  "is_blue_verified": ...,
  "tweet_text": "...",
  "like_count": ...,
  "created_at": "...",
  "tweet url": "..."
}

2. Prediction Profiler Agent

Functionality 1: Predictor Profile Builder

Purpose: Generate a profile for an X user based on their prediction behavior and topics.

Workflow:

  • Retrieves recent tweets for a given user.
  • Batches and analyzes tweets using an LLM to detect predictions.
  • Filters out non-predictive tweets.
  • Identifies patterns in predictions: topics, confidence levels, styles, and behavioral trends.

User Prompt Example:

"Build profile for @elonmusk."

Output Format:

{
  "handle": "...",
  "total_tweets_analyzed": ...,
  "prediction_tweets": [...],
  "prediction_count": ...,
  "prediction_rate": ...,
  "analysis": {
    "topics": {
      "politics": ...,
      "crypto": ...,
      ...
    },
    "confidence_level": "...",
    "prediction_style": "...",
    "patterns": ["...", "..."],
    "summary": "..."
  }
}

Functionality 2: Credibility Score Calculator

Purpose: Evaluate the reliability of an X user's predictions by verifying their historical accuracy.

Workflow:

  • Generates predictor profile.
  • Verifies past predictions using the Prediction Verifier Agent.
  • Calculates a credibility score: true / total predictions.
  • Outputs user statistics, verified predictions, and a credibility summary.

User Prompt Example:

"Calculate the credibility score for @elonmusk."

Output Format:

{
  "handle": "...",
  "credibility_score": ...,
  "prediction_stats": {
    "total": ...,
    "true": ...,
    "false": ...,
    "uncertain": ...
  },
  "verified_predictions": [...],
  "profile_summary": "..."
}

3. Prediction Verifier Agent

Purpose: Assess the truth value of a given prediction related to a Polymarket topic using external information sources.

Workflow:

  • Extracts relevant Polymarket topic using an LLM.

  • Searches for confirmation using:

  • Google Search API

  • Datura Web Search API

  • Analyzes source data to determine veracity via LLM.

  • Labels prediction as True, False, or Uncertain based on evidence.

User Prompt Example:

"Trump will win in 2025."

Output Format:

{
  "result": "...",  // True, False, or Uncertain
  "summary": "...",
  "sources": [...]
}

System Highlights

  • Agent Autonomy: Each agent operates with minimal user input and performs multi-step reasoning and action-taking.
  • LLM-Powered Reasoning: LLMs are used at multiple decision points—topic extraction, prediction identification, behavioral analysis, and result synthesis.
  • API Integration: Seamless access to Datura and Google APIs ensures robust data gathering and verification.
  • Structured Insights: Outputs are designed for downstream consumption in applications like dashboards, analytics engines, or credibility trackers.

Usage Examples

  1. Discovering Predictions
User: Find predictions on Ethereum price by Q4 2025
System: [Returns structured tweets containing related predictions]
  1. Building a Prediction Profile
User: Build profile for @naval
System: [Returns analysis including topics, style, confidence, and summary]
  1. Calculating Credibility
User: Calculate the credibility score for @balajis
System: [Returns credibility score and verification details]
  1. Verifying a Prediction
User: Bitcoin will hit $100k by end of 2024
System: [Returns verification status, summary, and evidence sources]

System Architecture Diagram

+--------------------------+
|    User Input Prompt     |
+------------+-------------+
             |
             v
+------------+-------------------+
|     Agent Dispatcher           |
+---+------------+--------+------+
|               |              |
v               v              v
+-----------+ +-----------+ +-----------+
|  Finder   | |  Profiler | | Verifier  |
|  Agent    | |  Agent    | |  Agent    |
+-----------+ +-----------+ +-----------+
                   |               ^
                   | (calls)       |
                   +---------------+

Outputs from all agents:
- Finder ➜ Structured Predictions
- Profiler ➜ Predictor Profiles & Credibility Scores
- Verifier ➜ Verified Prediction Outcomes

             v
+------------------------------------------+
|            Final Structured Output        |
|  (Predictions, Profiles, Verifications)   |
+------------------------------------------+

This diagram illustrates the multi-agent orchestration across components to convert social media predictions into actionable intelligence.