Skip to main content

​
What is an AI Oracle?

An AI oracle is a system that uses artificial intelligence to fetch real-world data and resolve onchain prediction markets automatically. delphAI combines:
  • AI agents that understand questions and fetch relevant data
  • Smart contracts that enforce resolution onchain
  • Cryptographic proofs that verify data authenticity

​
Why AI for Oracles?

Traditional oracles rely on humans or simple scripts. AI oracles offer unique advantages:

Natural Language Understanding

AI can parse complex questions and understand resolution criteria written in plain English

Multi-Source Verification

AI can query multiple data sources simultaneously and cross-verify information

Context Awareness

AI understands nuance and can handle edge cases that simple scripts would miss

Adaptive Learning

AI improves over time based on resolution history and feedback

​
How delphAI Works

​
Architecture Overview

​
Resolution Process

1

Schedule Registration

When a platform requests resolution:
oracle.requestResolution(
    marketId,
    "Will BTC reach $100k by Dec 31, 2025?",
    resolutionTime,
    "Use CoinGecko API. Check price at 23:59 UTC."
)
delphAI registers the market and schedules the AI agent.
2

Automated Trigger

At the exact resolution time:
  • Blockchain triggers the oracle contract
  • Oracle activates the assigned AI agent
  • No human intervention needed
3

Data Collection

The AI agent:
  • Parses the resolution criteria
  • Identifies required data sources
  • Queries APIs (CoinGecko, news APIs, etc.)
  • Collects multiple data points for verification
4

AI Analysis

The AI evaluates the data:
  • Applies logic from resolution criteria
  • Cross-references multiple sources
  • Checks for anomalies or inconsistencies
  • Calculates confidence score
5

Onchain Submission

If confidence threshold is met:
  • AI submits the resolution to smart contract
  • Includes cryptographic proof of data sources
  • Emits event for listening platforms
  • Market is marked as resolved

​
Data Sources

delphAI’s AI can access various types of data:

​
Financial Data

  • CoinGecko: Comprehensive crypto price data
  • CoinMarketCap: Market cap and price information
  • DEX Aggregators: Onchain price feeds
  • CEX APIs: Binance, Coinbase, Kraken
Example: “BTC price on Dec 31, 2025 at 23:59 UTC”
  • Stock APIs: Yahoo Finance, Alpha Vantage
  • Forex: Exchange rate providers
  • Commodities: Gold, oil, wheat prices
  • Economic Indicators: GDP, unemployment, inflation
Example: “S&P 500 closing price on market close”

​
Sports & Events

  • Official League APIs: NBA, NFL, FIFA
  • Sports Data Providers: ESPN, The Sports DB
  • Live Score APIs: Real-time game data
Example: “Lakers vs Warriors final score on Feb 15, 2025”
  • Box Office: Movie revenue data
  • Streaming: Netflix/Spotify charts
  • Awards: Oscar, Grammy, Emmy results
Example: “Best Picture winner at 2025 Oscars”

​
Real-World Data

  • Weather APIs: OpenWeatherMap, NOAA
  • Climate Data: Temperature, precipitation
  • Natural Events: Hurricanes, earthquakes
Example: “Temperature in NYC on Christmas Day at noon”
  • Election Results: Official government sources
  • Polling Data: Verified polling organizations
  • Legislative Actions: Bill tracking, votes
Example: “Winner of 2026 US midterm election in District X”

​
Onchain Data

  • Token Transfers: Whale movements
  • Smart Contract Events: Protocol actions
  • DeFi Metrics: TVL, volume, fees
  • NFT Sales: Floor prices, volume
Example: “Total ETH staked on Ethereum by Jan 1, 2026”

​
AI Capabilities

​
Natural Language Processing

The AI can understand complex questions: Simple: “Will BTC reach $100k?”
  • Direct price check
Complex: “Will Bitcoin’s price on CoinGecko exceed $100,000 at any point during December 2025?”
  • Requires historical data scanning
  • Time range analysis
  • Peak detection
Nuanced: “Will the Lakers win more than 50% of their regular season games in 2024-25?”
  • Requires counting wins/losses
  • Calculating percentage
  • Understanding sports seasons

​
Multi-Source Verification

For critical resolutions, AI queries multiple sources: 
// Example: Verifying election result
sources = [
  "Official Government API",
  "Associated Press",
  "Reuters",
  "Multiple news outlets"
]

// AI cross-references all sources
// Only resolves if consensus is reached

​
Edge Case Handling

AI can handle unexpected situations:

Data Source Down

Automatically falls back to secondary sources specified in criteria

Ambiguous Outcome

Flags market for manual review if confidence is too low

Partial Data

Attempts to infer outcome from available related data

Contradictory Sources

Applies weighting based on source reliability

​
Confidence Scoring

Every resolution includes a confidence score:
ScoreMeaningAction
95-100%CertainAuto-resolve immediately
80-94%High confidenceAuto-resolve with proof
60-79%ModerateFlag for verification
0-59%Low confidenceEscalate to dispute
Factors affecting confidence:
  • Source agreement (all sources say same thing)
  • Data quality (official vs unofficial)
  • Criteria clarity (unambiguous vs vague)
  • Historical accuracy (source reliability)

​
Security & Trust

​
How to Trust AI Resolutions?

Every resolution includes:
  • List of APIs queried
  • Exact data received
  • Timestamps of queries
  • Source URLs
All verifiable by anyone onchain.
AI submits cryptographic proofs:
  • Data signatures from sources
  • Hash of raw data
  • Proof of execution
Impossible to fake or manipulate.
If resolution is contested:
  • Users can stake tokens to dispute
  • Community reviews the data
  • Potential re-resolution if error found
  • Disputer rewarded if correct
AI agents have reputation scores:
  • Based on resolution accuracy
  • Disputed resolutions lower score
  • High-reputation agents get more markets

​
Limitations & Risks

Be aware of limitations:
  • AI relies on external data sources (if all sources are wrong, AI will be wrong)
  • Extremely ambiguous questions may not resolve with high confidence
  • New types of questions may require AI training/updates
  • Highly subjective questions are not suitable for AI resolution

​
Best Practices for Platforms

When integrating delphAI:

​
Write Clear Resolution Criteria

✅ Good Example: 
"Use CoinGecko API endpoint /simple/price for Bitcoin (BTC).
Check price at 2025-12-31 23:59:59 UTC.
If price >= $100,000, resolve to YES.
If price < $100,000, resolve to NO.
If CoinGecko is down, use CoinMarketCap as backup."
❌ Bad Example: 
"Check if Bitcoin reaches 100k by end of year"

​
Specify Reliable Sources

  • Use established, reputable data providers
  • Specify exact API endpoints when possible
  • Provide backup sources for redundancy
  • Avoid sources known for downtime

​
Set Appropriate Resolution Times

  • Allow sufficient time for data to be available
  • Consider timezone differences
  • Account for API update delays
  • Avoid microsecond-level precision unless necessary

​
Next Steps

Resolution Criteria

Learn how to write effective resolution criteria

Market Lifecycle

Understand the full market lifecycle

Oracle Contract

Explore the smart contract interface

AI Reliability

Deep dive into AI reliability and accuracy
⌘I