Interface: TypedNetworkObserverOptions

Defined in: packages/agentos/src/memory/retrieval/typed-network/TypedNetworkObserver.ts:66

Construction options for the observer.

Properties

llm

llm: ITypedExtractionLLM

Defined in: packages/agentos/src/memory/retrieval/typed-network/TypedNetworkObserver.ts:68

LLM adapter implementing the 6-step extraction call.

---

maxTokens?

optional maxTokens: number

Defined in: packages/agentos/src/memory/retrieval/typed-network/TypedNetworkObserver.ts:70

Max output tokens. Default 4096 (Hindsight extractions are typically 50-200 facts × ~30 tokens each).

---

temperature?

optional temperature: number

Defined in: packages/agentos/src/memory/retrieval/typed-network/TypedNetworkObserver.ts:72

Temperature. Default 0 for deterministic extraction.

---

timeoutMs?

optional timeoutMs: number

Defined in: packages/agentos/src/memory/retrieval/typed-network/TypedNetworkObserver.ts:80

Per-attempt request timeout in milliseconds. When the underlying llm.invoke() does not resolve within this window the attempt is abandoned and the observer falls through to the retry path. Used to prevent stale TCP sockets / hung OpenAI requests from deadlocking long-running ingest pipelines. Default 30 000 ms (30 s).