Integrating with AI Link (Partner Flow Overview)
This page describes what partners need to support in order to integrate AI inference into their applications using AI Link. The goal is to provide a clear operational picture of the integration flow, without requiring deep knowledge of Nesa’s internal execution or cryptographic design.
From a partner’s perspective, AI Link integration is stateful, asynchronous, and reusable.
Integration Prerequisites
Before submitting inference requests through AI Link, partners should ensure the following:
The application is deployed on a supported external chain
Users can interact with a designated smart contract to maintain a deposit balance
The frontend or application can submit requests and poll for results asynchronously
No assets, contracts, or application logic need to be migrated to the Nesa chain.
Partner-side Request Lifecycle
From the perspective of a partner application, each AI request follows a small number of consistent phases.
1. One-time deposit
Users deposit collateral once to enable AI inference requests. This deposit is:
Reusable across multiple requests
Not locked per request
Used only for post-execution settlement
As long as the user maintains the minimum required balance, no additional approvals are required for subsequent requests.
2. Request submission
When a user initiates an AI action, the application submits an inference request. The request is acknowledged immediately, allowing the application to continue without waiting for AI execution to complete.
At this stage:
The request is accepted for processing
No inference is performed synchronously
No blocking occurs on the originating chain
3. Asynchronous processing
AI inference is executed asynchronously on the Nesa network. During this time:
The application does not need to remain connected
No further on-chain interaction is required
The request progresses independently in the background
This design ensures that AI execution does not interfere with normal application flow or block user interactions.
4. Result polling and retrieval
The application periodically checks the status of the request. Once processing is complete:
Results become available for retrieval
The application can consume or display outputs as needed
Results can be used to trigger additional on-chain or off-chain logic
Polling intervals and timeouts are configurable to suit application needs.
5. Automatic settlement
After inference completes, fees are calculated based on actual execution costs. Settlement occurs automatically:
Fees are deducted from the user’s existing deposit
No upfront locking is required
Accounting reflects real usage rather than estimates
From the partner’s perspective, settlement is fully managed by the system.
Operational Characteristics
AILink is designed to minimize integration complexity while preserving strong guarantees:
Fully asynchronous — no synchronous waiting on inference
Reusable deposits — no per-request approvals
No blocking on confirmations — request recording and execution are decoupled
Failure-aware — requests can report processing, completion, or failure states
What Partners Do Not Need to Manage
Partners integrating with AILink do not need to:
Host or scale AI models
Manage inference infrastructure
Handle cryptographic key material manually
Perform on-chain settlement logic per request
These responsibilities are handled by the Nesa network.
Summary
Integrating with AILink requires only a small number of partner-side capabilities: maintaining a reusable deposit, submitting requests, polling for results, and consuming outputs. The complexity of AI execution, settlement, and infrastructure management is abstracted away, allowing partners to focus on application logic and user experience.
Last updated