Skip to main content

Interface: IAIGatewayConfig

Defined in: types.ts:130

Everything the gateway needs: one config per child, settings and persistence seams only. No class instance is in it; the gateway creates and injects its children itself.

Written by the host application and passed to the AIGateway constructor; createGatewayDeps hands each member to the child of the same name. A member that is optional has a runtime default of {}: the child runs with its own defaults.

Properties​

aiProfileStore​

aiProfileStore: IAIProfileStoreConfig

Defined in: types.ts:134

Where the profiles live: the repository of the AIProfileStore.

contextAssembler​

contextAssembler: IContextAssemblerConfig

Defined in: types.ts:212

How the system prompt is assembled: the prompt-cache switches (promptCaching, sharePromptCacheAcrossTenants) and the tokens reserved for the answer.

Example​

{promptCaching: true, sharePromptCacheAcrossTenants: false} - the stable part of every call is cached, per tenant

defaultCheapModelId​

defaultCheapModelId: string

Defined in: types.ts:222

Platform-wide cheap model for the helper calls (memory, query rewrite, judge) when a profile names none.

embedder​

embedder: IEmbedderConfig

Defined in: types.ts:152

The embedding model, its dimensions and the batch size (Embedder).

fileStore​

fileStore: IFileStoreConfig

Defined in: types.ts:140

Where uploaded files live, the allowed MIME types and the size cap: the FileStore.

guardrails?​

optional guardrails?: IGuardrailsConfig

Defined in: types.ts:203

The application's input and output rules, their timeout and the prompt length cap (Guardrails).

Default​

{} - the built-in checks only

humanGate​

humanGate: IHumanGateConfig

Defined in: types.ts:188

How long a pause waits for the human, and whether the model may ask the user a question (HumanGate).

kbIngestor?​

optional kbIngestor?: IKbIngestorConfig

Defined in: types.ts:220

How many documents are ingested at once (KbIngestor).

Default​

{}

kbRetriever?​

optional kbRetriever?: IKbRetrieverConfig

Defined in: types.ts:162

How many chunks a knowledge-base search returns and the minimum score (KbRetriever).

Default​

{}

llmClient​

llmClient: ILLMClientConfig

Defined in: types.ts:205

The provider credentials, the per-call timeout and the retries of a model call (LLMClient).

mdChunker?​

optional mdChunker?: IMdChunkerConfig

Defined in: types.ts:150

How a Markdown document is cut into chunks (MdChunker): the token bounds of a chunk and the overlap.

Default​

{}

modelRegistry​

modelRegistry: IModelRegistryConfig

Defined in: types.ts:132

The models a profile may name, with their caps and prices: the repository of the ModelRegistry.

promptTemplateStore​

promptTemplateStore: IPromptTemplateStoreConfig

Defined in: types.ts:136

Where the prompt templates live, and the templates seeded on the first run: the PromptTemplateStore.

queryRewriter​

queryRewriter: IQueryRewriterConfig

Defined in: types.ts:156

Whether a question is rewritten before the knowledge-base search (QueryRewriter).

runStore​

runStore: IRunStoreConfig

Defined in: types.ts:186

Where paused runs live: the repository of the RunStore.

skillRegistry?​

optional skillRegistry?: ISkillRegistryConfig

Defined in: types.ts:184

What the gateway applies to the skills of every profile and call (D124): the caps maxSkills (default 50) and maxBodyChars (default 50 000). The skills themselves are on the profile (IAIProfile.skills) and on the call (IGatewayAskInput.skills), never here.

Default​

{}

threadStore​

threadStore: IThreadStoreConfig

Defined in: types.ts:138

Where threads and turns live, and how long a call's lease on a thread holds: the ThreadStore.

toolPolicy?​

optional toolPolicy?: IToolPolicyConfig

Defined in: types.ts:176

The rules the ToolPolicy applies to every run: which permission unlocks which tools, and the guarded workflows (requires).

Default​

{}

toolRegistry?​

optional toolRegistry?: IToolRegistryConfig

Defined in: types.ts:169

The application's tools, the tool timeout, the output cap and the error report (ToolRegistry). The gateway adds its built-in tools (the user-memory tools, dyna_aigw_ask_user_question) in front of them.

Default​

{} - no tools of the application; the built-in ones are still there

trace​

trace: ITraceConfig

Defined in: types.ts:214

Where the trace entries live: the repository of the Trace.

usageGuard​

usageGuard: IUsageGuardConfig

Defined in: types.ts:197

The app's usage adapter, asked before every model call and told after it.

The gateway holds no token logic: who may spend what is the app's decision, made in adapter.check; what was spent is reported to adapter.addUsage.

Example​

{adapter: new InMemoryUsageAdapter()} - tests and the sandbox: approves everything, records every call

userMemoryStore​

userMemoryStore: IUserMemoryStoreConfig

Defined in: types.ts:144

The user memory: its switch, its repository, the caps of the facts and the extraction cadence (UserMemoryStore).

vectorStore​

vectorStore: IVectorStoreConfig

Defined in: types.ts:154

Where the chunks and their vectors live (VectorStore).

workingMemory​

workingMemory: IWorkingMemoryConfig

Defined in: types.ts:142

The thread memory: its switch and its strategy (WorkingMemory).