Skip to main content

​
General Questions

delphAI is an AI-powered onchain oracle that automatically resolves prediction markets. We use artificial intelligence running in Trusted Execution Environments (TEE) to fetch data, analyze it, and determine market outcomes without human intervention.We only resolve markets - we don’t run prediction market platforms. Platforms integrate our oracle to get automated, trustless resolution.
  1. A prediction market platform requests resolution from delphAI
  2. At the scheduled time, our AI (running in TEE) activates
  3. AI fetches data from specified sources
  4. AI analyzes data against resolution criteria
  5. AI submits the outcome onchain with cryptographic proof
  6. Platform receives the result and handles payouts
All resolution logic runs in secure hardware (TEE), making it tamper-proof.
If you’re a developer: Integrate delphAI into your prediction market platform for automated resolutionIf you’re a user: Use prediction market platforms that integrate delphAI for trustless, automated outcomesIf you’re a researcher: Study our open-source AI logic and TEE implementation
delphAI is being deployed on:
  • Ethereum
  • Polygon
  • Arbitrum
  • Base
More chains coming soon. Check the latest deployment addresses in our contracts overview.

​
Technical Questions

A TEE is a secure area of a processor that provides hardware-level isolation and encryption. It guarantees:
  • Code Integrity: Nobody can modify the AI code running inside
  • Data Confidentiality: Data processed inside TEE is encrypted
  • Attestation: Cryptographic proof that genuine code is running in genuine hardware
We use Intel SGX and AMD SEV for TEE security.
Every resolution includes:
  1. TEE Attestation: Cryptographic proof the resolution came from secure hardware
  2. Code Hash: Identifier of the exact AI version that resolved
  3. Data Proofs: Evidence of data sources queried and responses received
  4. Timestamp: When resolution occurred
You can verify all of this independently. See Security Overview for details.
delphAI can query any publicly accessible data source:
  • Crypto Prices: CoinGecko, CoinMarketCap, DEX APIs
  • Sports: ESPN, official league APIs, The Sports DB
  • Weather: OpenWeatherMap, NOAA
  • Financial: Yahoo Finance, Alpha Vantage
  • Onchain Data: Etherscan, The Graph, blockchain data
  • News/Events: Official sources, Reuters, AP
You specify data sources in your resolution criteria.
Resolution accuracy depends on:
  1. Criteria Clarity: Well-defined criteria = higher accuracy
  2. Data Sources: Reliable sources = better results
  3. Question Objectivity: Objective questions easier to resolve
The AI includes a confidence score (0-100) with each resolution. We recommend only accepting resolutions with 95%+ confidence.
If the AI’s confidence score is below the threshold:
  • Resolution is not automatically submitted
  • Market is flagged for review
  • Platform can implement dispute/manual resolution fallback
  • Resolution can be retried after more data is available
You set the confidence threshold in your integration.
Costs include:
  • Resolution Fee: Small fee per market resolution (TBA)
  • Gas Costs: Standard Ethereum gas for transactions
  • Platform Fees: Any fees your platform charges users
Pricing details will be announced before mainnet launch.

​
Integration Questions

Basic integration steps:
  1. Deploy your prediction market contract
  2. Call delphAI.createMarket() with creation fee when creating markets
  3. Store the returned marketId
  4. Poll delphAI.getMarket(marketId) to check resolution status
  5. When resolved, use the outcome to distribute winnings on your platform
See our Quickstart Guide for detailed instructions.
No! delphAI handles all infrastructure:
  • TEE servers
  • AI execution
  • Data fetching
  • Onchain submission
You just interact with our smart contract.
Yes! We provide testnet deployments on:
  • Sepolia
  • Mumbai (Polygon)
  • Arbitrum Sepolia
  • Base Sepolia
Test your full integration before mainnet launch.
You don’t need to implement any interface! Simply call delphAI functions:
interface IDelphAI {
    function createMarket(
        string memory question,
        string memory description,
        string[] memory possibleOutcomes,
        uint256 resolutionTimestamp
    ) external payable returns (uint256);

    function getMarket(uint256 marketId) external view returns (Market memory);
    function marketCreationFee() external view returns (uint256);
}
See Oracle Contract for full API.
Yes! While designed for prediction markets, delphAI can resolve any binary or multiple-choice question:
  • Insurance payouts (e.g., flight delay insurance)
  • Conditional payments (e.g., milestone-based releases)
  • Automated decision making
  • Data verification
Any use case requiring trustless, automated oracle resolution.

​
Security Questions

delphAI implements multiple security layers:
  1. TEE Security: Hardware-level code and data protection
  2. Audited Contracts: Smart contracts audited by leading firms
  3. Cryptographic Proofs: All resolutions include verifiable proofs
  4. Open Source: AI logic is transparent and auditable
See Security Overview for complete details.
TEE compromise is extremely difficult but we plan for it:
  • Multiple TEE technologies (Intel SGX, AMD SEV)
  • Attestation verification catches compromised TEE
  • Platforms can reject resolutions from unknown TEE versions
  • Dispute mechanisms for contested resolutions
Additionally, TEE vulnerabilities are rare and quickly patched by hardware vendors.
Resolution criteria should specify backup data sources:
Primary: CoinGecko API
Backup: CoinMarketCap API
Fallback: Retry after 1 hour
If all sources fail, resolution is delayed until data is available. Platforms can implement emergency resolution after timeout.
Yes, platforms can implement dispute mechanisms:
  1. User stakes tokens to dispute resolution
  2. Community reviews the evidence (attestation + proof)
  3. If dispute valid: User gets stake back + reward
  4. If dispute invalid: Stake is slashed
Dispute logic is implemented by the platform, not delphAI.
Audits are currently in progress. See Audits for status and timeline.Upon completion, all audit reports will be published publicly.

​
Resolution Criteria Questions

Good resolution criteria must be:
  1. Specific: Name exact data sources and APIs
  2. Clear: Unambiguous YES/NO or outcome logic
  3. Timed: Exact resolution time with timezone
  4. Complete: Handle edge cases and failures
  5. Verifiable: Based on objective data
Example:
Question: "Will BTC reach $100k by Dec 31, 2025?"
Source: CoinGecko API at 2025-12-31 23:59:59 UTC
Logic: If BTC >= $100,000 → YES, else → NO
Backup: CoinMarketCap if CoinGecko down
See Best Practices for detailed guide.
Avoid:
  • ❌ Vague questions: “Will BTC do well?”
  • ❌ Subjective criteria: “Will the movie be good?”
  • ❌ No data source: “Check the price”
  • ❌ Ambiguous timing: “End of year”
  • ❌ No backup plan: Single data source with no fallback
These lead to low confidence scores or resolution failures.
No. delphAI only resolves objective, verifiable questions:✅ “Will BTC price be >= $100k?” (objective, verifiable) ❌ “Will BTC price be ‘high’?” (subjective, vague)✅ “Will Lakers win vs Warriors?” (objective, score-based) ❌ “Will Lakers play well?” (subjective opinion)For subjective decisions, consider using human voting oracles instead.
Define edge case behavior in your criteria:For ties:
If teams have same score → Resolve to TIE outcome
OR use tiebreaker rules from official source
For exact thresholds:
If price = exactly $100,000 → YES (use >= not >)
For cancelled events:
If game cancelled → Return all funds
OR postpone until rescheduled game
See Best Practices for more examples.

​
Troubleshooting

Possible reasons:
  1. Data source delay: Source hasn’t updated yet
  2. Low confidence: AI waiting for more data/confirmation
  3. Network congestion: Transaction pending in mempool
  4. Criteria unclear: AI unsure how to resolve
Check resolution status with getResolution(marketId). If still pending after expected time, contact support.
Check:
  1. ✅ Does your contract implement IPlatformCallback?
  2. ✅ Is the function marked external?
  3. ✅ Is it callable by delphAI oracle address?
  4. ✅ Does it have access control restrictions?
  5. ✅ Is your contract verified on block explorer?
Test callback manually on testnet before mainnet.
Low confidence usually means:
  • Vague criteria: Make criteria more specific
  • Unreliable sources: Use established data providers
  • Ambiguous question: Rephrase question clearly
  • Data unavailable: Wrong data source or endpoint
  • Too early: Resolution time before data available
Review your resolution criteria and data sources.
Verify:
  1. You’re using the correct attestation verification library
  2. You’ve approved the code hash for this AI version
  3. Attestation hasn’t expired (check timestamp)
  4. TEE address is authorized
See Security Best Practices for proper verification.

​
Still Have Questions?

Troubleshooting

Detailed troubleshooting guide

Telegram Community

Join our community for support

Documentation

Read the full documentation

GitHub

View code and examples
⌘I