The architecture page preserves the deeper execution, routing and escrow material while the overview remains concise.
See technical details
Decentralized AI infrastructure
AI without one central gatekeeper.
Most AI applications depend on a small number of model and cloud providers. BitcoinAI is building an open coordination network where applications can request AI, providers can perform useful work, results can be verified, and services can settle through blockchain infrastructure.
- AI Orchestration
- Model-Agnostic
- Planned
- Verifiable Receipts
- Proof-of-AI
- Canonical
- Qualifying BATT Fee Split
- 90 / 10
- Canonical
- Concurrent AI Job Target† (Engineering target; subject to benchmark validation.)
- 100K–500K
- Research target
The AI stack
BitcoinAI handles trust. BatteryAGI handles intelligence.
- Layer 1 · ApplicationUser / Application
- Person
- dApp
- Enterprise application
- Autonomous agent
- External chain
- Layer 2 · BitcoinAIBitcoinAI
- AI request
- Quote reference
- Escrow
- Job state
- Smart contract
- Cross-chain coordination
- Layer 3 · BatteryAGIBatteryAGI
- Model routing
- Fusion AI
- Provider selection
- Task orchestration
- Layer 4 · Off-chainDistributed AI Providers
- Inference
- Retrieval
- Storage
- Specialist agents
- Verification
- Relays
- Layer 5 · BitcoinAI + BATTBitcoinAI Verification + BATT Settlement
- Proof-of-AI receipt
- Result commitment
- Verification
- Settlement instruction
- Provider compensation
- BitcoinAI
- Trust, proof, state, smart contracts, coordination and settlement
- BatteryAGI
- Model-agnostic AI orchestration and execution
- BATT
- AI-service settlement asset — not the BTCAI gas or security asset
Heavy AI compute stays off-chain. Compact proof, state, and settlement remain on BitcoinAI.
From request to verified result
A complete AI transaction has state, proof, and settlement.
- 1QUOTEDBatteryAGI prices the request.
- 2ESCROWEDValue is locked against AI_JOB_ID.
- 3SENT_
TO_ BATTERYThe job is dispatched for execution. - 4EXECUTINGProviders run the models off-chain.
- 5RESULT_
RETURNEDResult and receipt come back. - 6PROOF_
VERIFIEDBitcoinAI checks the Proof-of-AI receipt. - 7PAYMENT_
RELEASEDEscrow releases after verification. - 8BATT_
SETTLEDProviders are paid in BATT. - 9COMPLETEState is final and auditable.
Failure paths
Current design- TimeoutNo result within the window
- Provider failureExecution could not complete
- Invalid proofReceipt fails verification
- Rejected resultResult fails policy or checks
- Refund / cancelEscrow returns per final rules
BTCAI escrow should not be treated as immediately spent before successful verification.
See Settlement ArchitectureCompact evidence
The receipt connects AI execution to blockchain state.
| Field | Description |
|---|---|
| jobId | Unique AI_JOB_ID |
| requester | Who asked and funded the job |
| providerId | Executing provider |
| modelId | Model identifier |
| modelVersion | Model / artifact version |
| inputHash | Commitment to the input |
| outputHash | Commitment to the output |
| configHash | Inference configuration |
| executedAt | Execution timestamp |
| nonce | Anti-replay nonce |
| receiptHash | Hash of the receipt |
| signature | Provider signature |
| verificationClass | Tier 0–4 |
| teeAttestationHashoptional | Optional hardware evidence |
| zkProofHashoptional | Optional zk proof |
| paymentRefoptional | Escrow / settlement reference |
| sourceChainoptional | Where the request began |
| destinationChainoptional | Where the result returns |
Hash(
chain_id
|| job_id
|| requester
|| provider_id
|| input_hash
|| output_hash
|| nonce
)Richer than a hash
A hash alone proves integrity. The receipt adds who ran the job, which model, when, under which verification class, and how it was paid.
Conceptual receipt structure; final encoding is implementation-defined.
Model-agnostic by design
Use the right model for the job.
- Frontier proprietary APIse.g. OpenAI-family APIs
- Open-weight LLMse.g. DeepSeek-class models
- Local / private modelsEnterprise-hosted
- Multimodal modelsText + vision + audio
- Code modelsGeneration and review
- Image modelsGeneration and analysis
- Speech / audio modelsTranscription and voice
- Specialist industry modelsDomain-tuned
- Retrieval systemsRAG and vector search
- Future providersOpen to new entrants
Example model class / potential provider
Named model families are examples of model classes BatteryAGI is designed to route to. They are not announced partnerships or confirmed integrations.
Explore AI
Start simple, then open the technical layer you need.
Build the AI economy
Bring AI execution, proof, and settlement together.
Build an AI application, connect compute, create an autonomous agent, or integrate BitcoinAI's verification and settlement architecture into a multichain workflow.