Skip to main content

IntelligenceProvider

@totemsdk/intelligence

@totemsdk/intelligence / IntelligenceProvider

Interface: IntelligenceProvider

Provider-neutral intelligence interface.

Implementors wrap a concrete inference runtime (QVAC, remote LLM gateway, embedded model host, …). The provider is compute-only: it cannot sign, cannot move value, and must never hold private keys.

Properties​

capabilities​

readonly capabilities: readonly `intelligence:${string}`[]

Domains the provider currently supports.

displayName​

readonly displayName: string

Human-readable provider name, e.g. 'QVAC In-situ Inference'.

id​

readonly id: string

Stable provider identifier, e.g. 'qvac'.

isReady​

readonly isReady: boolean

True if the provider is connected / ready.

version​

readonly version: string

Wrapped provider runtime version.

Methods​

cancel()​

cancel(requestId): Promise<IntelligenceOutcome<void>>

Cancel an in-flight operation by request id.

Parameters​

requestId​

string

Returns​

Promise<IntelligenceOutcome<void>>

close()​

close(): Promise<void>

Release provider resources. Further invoke calls should reject.

Returns​

Promise<void>

invoke()​

invoke<T>(op): Promise<IntelligenceOutcome<T>>

Execute a single inference operation.

Implementations MAY throw IntelligenceError for hard failures but should prefer returning IntelligenceErrorResult for operational failures so the caller can inspect code/retryable without try/catch.

Type Parameters​

T​

T = unknown

Parameters​

op​

IntelligenceOperation<T>

Returns​

Promise<IntelligenceOutcome<T>>

invokeStream()​

invokeStream(op): AsyncIterable<IntelligenceStreamChunk>

Execute a streaming inference operation.

Returns an async iterable of stream chunks. If the operation is not stream-capable, the implementation must throw NOT_IMPLEMENTED.

Parameters​

op​

IntelligenceStreamOperation

Returns​

AsyncIterable<IntelligenceStreamChunk>