@nemesis-oss/ollama-sdk 1.9.0 → 1.10.0

package/dist/index.d.cts

@@ -1,12 +1,12 @@
-import { F as FetchLike, M as Middleware, L as Logger, R as RequestLifecycleHook, H as HttpClient, c as OllamaStreamEvent, A as AbortableAsyncIterable, d as OllamaStreamEventType, P as ProgressStreamResult, B as BinaryBody, O as OllamaRuntime, C as ChatStreamResult, G as GenerateStreamResult, e as RawUsageSource, T as TokenUsage } from './runtime-D2W9TfCJ.cjs';
-export { D as DoneEvent, E as ErrorEvent, f as HttpBody, g as HttpClientOptions, h as HttpRequestOptions, i as LifecycleErrorEvent, j as LifecycleRetryEvent, k as LifecycleStartEvent, l as LifecycleSuccessEvent, m as LogFn, n as MessageEvent, o as MiddlewareContext, N as NANOS_PER_MS, p as NANOS_PER_SECOND, q as NOOP_LOGGER, r as NextFunction, s as OllamaAbortError, t as OllamaAgentMaxIterationsError, u as OllamaAgentMaxToolCallsError, v as OllamaAuthError, w as OllamaBadGatewayError, x as OllamaClientError, y as OllamaClientErrorOptions, z as OllamaErrorRequestContext, J as OllamaErrorResponseContext, K as OllamaGenericClientError, Q as OllamaIncompatibleModelError, U as OllamaMcpError, V as OllamaModelRoutingError, W as OllamaNetworkError, X as OllamaNotFoundError, Y as OllamaOverloadedError, Z as OllamaQuotaExceededError, _ as OllamaRateLimitError, $ as OllamaRequestTooLargeError, a0 as OllamaRequestValidationError, a1 as OllamaResponseValidationError, a2 as OllamaServerError, a3 as OllamaServerVersionUnknownError, a4 as OllamaSkillInvalidError, a5 as OllamaSkillNotFoundError, a6 as OllamaStreamError, a7 as OllamaTimeoutError, a8 as OllamaToolTimeoutError, a9 as OllamaToolValidationError, aa as OllamaUnsupportedCapabilityError, ab as RequestContext, ac as RequestLifecycleEvent, ad as ResponseContext, ae as SseEvent, af as ThinkingEvent, ag as TokenEvent, ah as ToolCallEvent, ai as composeMiddleware, aj as createConsoleLogger, ak as extractUsage, al as mapError, am as parseSseStream } from './runtime-D2W9TfCJ.cjs';
-import { T as ThinkingMetadata, M as ModelResponse, S as ShowRequestOptions, a as ShowResponse, P as PullRequestOptions, b as ProgressResponse, c as PushRequestOptions, C as CreateRequestOptions, D as DeleteRequestOptions, d as StatusResponse, e as CopyRequestOptions, f as PsResponse, V as VersionResponse, R as RequestCancellationOptions, g as ToolParameters, h as ThinkValue, i as ToolCall, j as ChatRequestOptions, k as ModelOptions, l as ToolDefinition, F as FormatOption, m as ChatResponse, n as Message, G as GenerateRequestOptions, o as GenerateResponse, E as EmbedRequestOptions, p as EmbedResponse, q as EmbeddingsRequestOptions, r as EmbeddingsResponse, W as WebSearchRequestOptions, s as WebSearchResponse, t as WebFetchRequestOptions, u as WebFetchResponse, U as UsageRequestOptions, B as BalanceRequestOptions, v as VisionInput } from './types-DVcX_zMp.cjs';
-export { L as ListResponse, w as Logprob, x as LogprobToken, y as ModelDetails, z as Role, A as ToolCallFunction, H as ToolFunctionDefinition, I as ToolProperty, J as UsageRange, K as UsageScope, N as WebSearchResult, O as imageStringNeedsResolution, Q as resolveImageInput } from './types-DVcX_zMp.cjs';
+import { F as FetchLike, M as Middleware, L as Logger, R as RequestLifecycleHook, H as HttpClient, c as OllamaStreamEvent, A as AbortableAsyncIterable, d as OllamaStreamEventType, P as ProgressStreamResult, B as BinaryBody, O as OllamaRuntime, C as ChatStreamResult, G as GenerateStreamResult, e as RawUsageSource, T as TokenUsage } from './runtime-BsdBbuOv.cjs';
+export { D as DoneEvent, E as ErrorEvent, f as HttpBody, g as HttpClientOptions, h as HttpRequestOptions, i as LifecycleErrorEvent, j as LifecycleRetryEvent, k as LifecycleStartEvent, l as LifecycleSuccessEvent, m as LogFn, n as MessageEvent, o as MiddlewareContext, N as NANOS_PER_MS, p as NANOS_PER_SECOND, q as NOOP_LOGGER, r as NextFunction, s as OllamaAbortError, t as OllamaAgentMaxIterationsError, u as OllamaAgentMaxToolCallsError, v as OllamaAuthError, w as OllamaBadGatewayError, x as OllamaClientError, y as OllamaClientErrorOptions, z as OllamaErrorRequestContext, J as OllamaErrorResponseContext, K as OllamaGenericClientError, Q as OllamaIncompatibleModelError, U as OllamaMcpError, V as OllamaModelRoutingError, W as OllamaNetworkError, X as OllamaNotFoundError, Y as OllamaOverloadedError, Z as OllamaQuotaExceededError, _ as OllamaRateLimitError, $ as OllamaRequestTooLargeError, a0 as OllamaRequestValidationError, a1 as OllamaResponseValidationError, a2 as OllamaServerError, a3 as OllamaServerVersionUnknownError, a4 as OllamaSkillInvalidError, a5 as OllamaSkillNotFoundError, a6 as OllamaStreamError, a7 as OllamaTimeoutError, a8 as OllamaToolTimeoutError, a9 as OllamaToolValidationError, aa as OllamaUnsupportedCapabilityError, ab as RequestContext, ac as RequestLifecycleEvent, ad as ResponseContext, ae as SseEvent, af as ThinkingEvent, ag as TokenEvent, ah as ToolCallEvent, ai as composeMiddleware, aj as createConsoleLogger, ak as extractUsage, al as mapError, am as parseSseStream } from './runtime-BsdBbuOv.cjs';
+import { T as ThinkingMetadata, M as ModelResponse, S as ShowRequestOptions, a as ShowResponse, P as PullRequestOptions, b as ProgressResponse, c as PushRequestOptions, C as CreateRequestOptions, D as DeleteRequestOptions, d as StatusResponse, e as CopyRequestOptions, f as PsResponse, V as VersionResponse, R as RequestCancellationOptions, g as ToolParameters, h as ThinkValue, i as ToolCall, j as ChatRequestOptions, k as ModelOptions, l as ToolDefinition, F as FormatOption, m as ChatResponse, n as Message, G as GenerateRequestOptions, o as GenerateResponse, E as EmbedRequestOptions, p as EmbedResponse, q as EmbeddingsRequestOptions, r as EmbeddingsResponse, W as WebSearchRequestOptions, s as WebSearchResponse, t as WebFetchRequestOptions, u as WebFetchResponse, U as UsageRequestOptions, B as BalanceRequestOptions, v as VisionInput } from './types-BtQexFpT.cjs';
+export { L as ListResponse, w as Logprob, x as LogprobToken, y as ModelDetails, z as Role, A as ToolCallFunction, H as ToolFunctionDefinition, I as ToolProperty, J as UsageRange, K as UsageScope, N as WebSearchResult, O as imageStringNeedsResolution, Q as resolveImageInput } from './types-BtQexFpT.cjs';
 import { z } from 'zod';
-import { R as RequestRunner, S as SystemOneChoiceAnswer, o as SystemOneNoulAnswer, p as SystemOneScoreAnswer, q as SystemOneChoiceQuestion, r as SystemOneNoulQuestion, s as SystemOneScoreQuestion, t as SystemOneRequest$1, u as SystemOneResponse$1, v as SystemOneContent, N as NativeApi, w as OpenAICompatClient, x as AnthropicCompatClient } from './anthropic-BIh45UXj.cjs';
-export { y as AnthropicCacheControl, z as AnthropicContentBlock, B as AnthropicImageContentBlock, C as AnthropicMessage, n as AnthropicMessageStreamEvent, A as AnthropicMessagesRequest, m as AnthropicMessagesResponse, D as AnthropicMessagesStream, E as AnthropicMetadata, F as AnthropicOutputConfig, G as AnthropicRedactedThinkingContentBlock, H as AnthropicSystem, I as AnthropicSystemTextBlock, J as AnthropicTextContentBlock, K as AnthropicThinkingConfig, L as AnthropicThinkingContentBlock, M as AnthropicTool, P as AnthropicToolChoice, Q as AnthropicToolResultContentBlock, T as AnthropicToolUseContentBlock, U as AnthropicUnknownStreamEvent, V as OllamaAnthropicContentBlock, W as OllamaAnthropicMessage, X as OllamaAnthropicMessagesRequest, Y as OllamaAnthropicSystem, Z as OllamaAnthropicSystemTextBlock, _ as OllamaAnthropicTextContentBlock, $ as OllamaAnthropicThinkingConfig, a0 as OllamaAnthropicTool, a1 as OllamaAnthropicToolResultContentBlock, a2 as OllamaOpenAIChatCompletionRequest, a3 as OllamaOpenAICompletionRequest, a4 as OllamaOpenAIEmbeddingRequest, a5 as OllamaOpenAIReasoningEffort, a6 as OllamaOpenAIResponsesRequest, a7 as OpenAIChatCompletionChoice, b as OpenAIChatCompletionChunk, a8 as OpenAIChatCompletionChunkChoice, a9 as OpenAIChatCompletionDelta, O as OpenAIChatCompletionRequest, a as OpenAIChatCompletionResponse, aa as OpenAIChatCompletionStream, ab as OpenAICompletionChoice, e as OpenAICompletionChunk, c as OpenAICompletionRequest, d as OpenAICompletionResponse, ac as OpenAICompletionStream, ad as OpenAIContentPart, ae as OpenAIEmbeddingInput, af as OpenAIEmbeddingItem, f as OpenAIEmbeddingRequest, g as OpenAIEmbeddingResponse, ag as OpenAIFunctionDefinition, ah as OpenAIImageUrlContentPart, h as OpenAIListModelsResponse, ai as OpenAIMessage, i as OpenAIModelItem, aj as OpenAIReasoningEffort, ak as OpenAIResponseFormat, al as OpenAIResponsesCompletedEvent, am as OpenAIResponsesContentPartAddedEvent, an as OpenAIResponsesContentPartDoneEvent, ao as OpenAIResponsesCreatedEvent, ap as OpenAIResponsesError, aq as OpenAIResponsesEventBase, ar as OpenAIResponsesFailedEvent, as as OpenAIResponsesFunctionCallArgumentsDeltaEvent, at as OpenAIResponsesFunctionCallArgumentsDoneEvent, au as OpenAIResponsesInProgressEvent, av as OpenAIResponsesIncompleteDetails, aw as OpenAIResponsesIncompleteEvent, ax as OpenAIResponsesOutputContent, ay as OpenAIResponsesOutputFunctionCall, az as OpenAIResponsesOutputItem, aA as OpenAIResponsesOutputItemAddedEvent, aB as OpenAIResponsesOutputItemDoneEvent, aC as OpenAIResponsesOutputMessage, aD as OpenAIResponsesOutputReasoning, aE as OpenAIResponsesOutputRefusal, aF as OpenAIResponsesOutputTextContent, aG as OpenAIResponsesOutputTextDeltaEvent, aH as OpenAIResponsesOutputTextDoneEvent, aI as OpenAIResponsesQueuedEvent, aJ as OpenAIResponsesReasoningSummaryPartAddedEvent, aK as OpenAIResponsesReasoningSummaryPartDoneEvent, aL as OpenAIResponsesReasoningSummaryTextDeltaEvent, aM as OpenAIResponsesReasoningSummaryTextDoneEvent, aN as OpenAIResponsesReasoningTextDeltaEvent, aO as OpenAIResponsesReasoningTextDoneEvent, aP as OpenAIResponsesRefusalDeltaEvent, aQ as OpenAIResponsesRefusalDoneEvent, j as OpenAIResponsesRequest, k as OpenAIResponsesResponse, aR as OpenAIResponsesStatus, aS as OpenAIResponsesStream, aT as OpenAIResponsesStreamError, l as OpenAIResponsesStreamEvent, aU as OpenAIResponsesUsage, aV as OpenAIStreamOptions, aW as OpenAITextContentPart, aX as OpenAITool, aY as OpenAIToolCall, aZ as OpenAIToolCallDelta, a_ as OpenAIUsage, a$ as SystemOneConfidence, b0 as SystemOneProbabilities } from './anthropic-BIh45UXj.cjs';
-import { T as ToolHandler, a as Tool, A as AnyTool, b as ToolRegistry, c as ToolExecutionResult } from './compose-Dj6FdF2n.cjs';
-export { d as AppliedSkillResult, e as ApplySkillOptions, P as ParsedFrontmatter, S as Skill, f as SkillFrontmatter, g as SkillSummary, h as ToolExecutionContext, i as ToolExecutionFailure, j as ToolExecutionSuccess, k as ToolRegistryOptions, l as applySkill, p as parseFrontmatter } from './compose-Dj6FdF2n.cjs';
+import { R as RequestRunner, S as SystemOneChoiceAnswer, o as SystemOneNoulAnswer, p as SystemOneScoreAnswer, q as SystemOneChoiceQuestion, r as SystemOneNoulQuestion, s as SystemOneScoreQuestion, t as SystemOneRequest$1, u as SystemOneResponse$1, v as SystemOneContent, N as NativeApi, w as OpenAICompatClient, x as AnthropicCompatClient } from './anthropic-DuE8M6Tz.cjs';
+export { y as AnthropicCacheControl, z as AnthropicContentBlock, B as AnthropicImageContentBlock, C as AnthropicMessage, n as AnthropicMessageStreamEvent, A as AnthropicMessagesRequest, m as AnthropicMessagesResponse, D as AnthropicMessagesStream, E as AnthropicMetadata, F as AnthropicOutputConfig, G as AnthropicRedactedThinkingContentBlock, H as AnthropicSystem, I as AnthropicSystemTextBlock, J as AnthropicTextContentBlock, K as AnthropicThinkingConfig, L as AnthropicThinkingContentBlock, M as AnthropicTool, P as AnthropicToolChoice, Q as AnthropicToolResultContentBlock, T as AnthropicToolUseContentBlock, U as AnthropicUnknownStreamEvent, V as OllamaAnthropicContentBlock, W as OllamaAnthropicMessage, X as OllamaAnthropicMessagesRequest, Y as OllamaAnthropicSystem, Z as OllamaAnthropicSystemTextBlock, _ as OllamaAnthropicTextContentBlock, $ as OllamaAnthropicThinkingConfig, a0 as OllamaAnthropicTool, a1 as OllamaAnthropicToolResultContentBlock, a2 as OllamaOpenAIChatCompletionRequest, a3 as OllamaOpenAICompletionRequest, a4 as OllamaOpenAIEmbeddingRequest, a5 as OllamaOpenAIReasoningEffort, a6 as OllamaOpenAIResponsesRequest, a7 as OpenAIChatCompletionChoice, b as OpenAIChatCompletionChunk, a8 as OpenAIChatCompletionChunkChoice, a9 as OpenAIChatCompletionDelta, O as OpenAIChatCompletionRequest, a as OpenAIChatCompletionResponse, aa as OpenAIChatCompletionStream, ab as OpenAICompletionChoice, e as OpenAICompletionChunk, c as OpenAICompletionRequest, d as OpenAICompletionResponse, ac as OpenAICompletionStream, ad as OpenAIContentPart, ae as OpenAIEmbeddingInput, af as OpenAIEmbeddingItem, f as OpenAIEmbeddingRequest, g as OpenAIEmbeddingResponse, ag as OpenAIFunctionDefinition, ah as OpenAIImageUrlContentPart, h as OpenAIListModelsResponse, ai as OpenAIMessage, i as OpenAIModelItem, aj as OpenAIReasoningEffort, ak as OpenAIResponseFormat, al as OpenAIResponsesCompletedEvent, am as OpenAIResponsesContentPartAddedEvent, an as OpenAIResponsesContentPartDoneEvent, ao as OpenAIResponsesCreatedEvent, ap as OpenAIResponsesError, aq as OpenAIResponsesEventBase, ar as OpenAIResponsesFailedEvent, as as OpenAIResponsesFunctionCallArgumentsDeltaEvent, at as OpenAIResponsesFunctionCallArgumentsDoneEvent, au as OpenAIResponsesInProgressEvent, av as OpenAIResponsesIncompleteDetails, aw as OpenAIResponsesIncompleteEvent, ax as OpenAIResponsesOutputContent, ay as OpenAIResponsesOutputFunctionCall, az as OpenAIResponsesOutputItem, aA as OpenAIResponsesOutputItemAddedEvent, aB as OpenAIResponsesOutputItemDoneEvent, aC as OpenAIResponsesOutputMessage, aD as OpenAIResponsesOutputReasoning, aE as OpenAIResponsesOutputRefusal, aF as OpenAIResponsesOutputTextContent, aG as OpenAIResponsesOutputTextDeltaEvent, aH as OpenAIResponsesOutputTextDoneEvent, aI as OpenAIResponsesQueuedEvent, aJ as OpenAIResponsesReasoningSummaryPartAddedEvent, aK as OpenAIResponsesReasoningSummaryPartDoneEvent, aL as OpenAIResponsesReasoningSummaryTextDeltaEvent, aM as OpenAIResponsesReasoningSummaryTextDoneEvent, aN as OpenAIResponsesReasoningTextDeltaEvent, aO as OpenAIResponsesReasoningTextDoneEvent, aP as OpenAIResponsesRefusalDeltaEvent, aQ as OpenAIResponsesRefusalDoneEvent, j as OpenAIResponsesRequest, k as OpenAIResponsesResponse, aR as OpenAIResponsesStatus, aS as OpenAIResponsesStream, aT as OpenAIResponsesStreamError, l as OpenAIResponsesStreamEvent, aU as OpenAIResponsesUsage, aV as OpenAIStreamOptions, aW as OpenAITextContentPart, aX as OpenAITool, aY as OpenAIToolCall, aZ as OpenAIToolCallDelta, a_ as OpenAIUsage, a$ as SystemOneConfidence, b0 as SystemOneProbabilities } from './anthropic-DuE8M6Tz.cjs';
+import { T as ToolHandler, a as Tool, A as AnyTool, b as ToolRegistry, c as ToolExecutionResult } from './compose-lt-Qz3lI.cjs';
+export { d as AppliedSkillResult, e as ApplySkillOptions, P as ParsedFrontmatter, S as Skill, f as SkillFrontmatter, g as SkillSummary, h as ToolExecutionContext, i as ToolExecutionFailure, j as ToolExecutionSuccess, k as ToolRegistryOptions, l as applySkill, p as parseFrontmatter } from './compose-lt-Qz3lI.cjs';
 import { M as McpClientLike, a as McpToolDescriptor, b as McpTask, c as McpCallToolResult } from './types-CI-gM2u5.cjs';
 export { d as McpContentBlock, e as McpCreateTaskResult, f as McpElicitationFormRequest, g as McpElicitationHandlers, h as McpElicitationRequest, i as McpElicitationResult, j as McpElicitationUrlRequest, k as McpIcon, l as McpInputRequiredResult, m as McpListToolsParams, n as McpListToolsResult, o as McpRequestOptions, p as McpServerCapabilities, q as McpTaskMethod, r as McpTaskRequest, s as McpTaskStatus, t as McpToolAnnotations, u as McpToolCallResult, v as McpToolExecution } from './types-CI-gM2u5.cjs';
 import * as Otel from '@opentelemetry/api';
@@ -981,6 +981,131 @@ declare class ConversationSession {
     reset(): void;
 }
 
+/**
+ * Batch-constrained embedding generation for high-volume ingestion pipelines.
+ *
+ * Ollama's `/api/embed` accepts an array of inputs, but two independent limits
+ * make naive usage fragile at corpus scale:
+ *
+ * 1. **Concurrency.** `Promise.all(inputs.map(...))` fires every request at
+ *    once; against a local Ollama daemon this contends sockets, saturates the
+ *    request queue (`OLLAMA_MAX_QUEUE`, surfacing as 503 `overloaded`), and —
+ *    worst case — exhausts GPU VRAM as the model instance serves N prompts
+ *    simultaneously.
+ * 2. **Context window.** Ollama truncates any input string longer than the
+ *    model's context window (`num_ctx`) *silently* unless `truncate: false` is
+ *    set — degraded vectors with no error, no warning, nothing on the wire.
+ *
+ * {@link batchEmbed} addresses the first with a fixed-size worker pool over
+ * `client.embed()` (so every batch still rides the client's full pipeline:
+ * failover, retry, auth, logging, telemetry). The client-level
+ * `embedBatch()` wrapper addresses the second with a pre-flight per-string
+ * token estimate against the resolved `num_ctx` window (see
+ * {@link findOversizedEmbedInputs}).
+ *
+ * Semantics:
+ * - **Order-preserving** — result index `i` is always the embedding of
+ *   `input[i]`, regardless of which batch completed first.
+ * - **Fail-fast** — the first batch error aborts every sibling in-flight
+ *   batch (surfaces the *original* error, not the derived aborts) and rejects
+ *   the whole operation. Vector indexing that half-succeeds is a corrupted
+ *   index; callers re-running an ingestion prefer all-or-nothing.
+ * - **Abortable** — the caller's `signal` cancels queued and in-flight batches
+ *   alike and rejects with `code: 'aborted'`.
+ */
+
+/**
+ * Default number of input strings per `/api/embed` request. 32 keeps a single
+ * request's prompt comfortably inside typical embedding-model context windows
+ * (2048+) while amortizing HTTP overhead across the corpus.
+ */
+declare const DEFAULT_EMBED_BATCH_SIZE = 32;
+/**
+ * Default number of batches kept in flight concurrently. 3 is deliberately
+ * conservative: a local Ollama daemon serves one model instance per GPU, so
+ * 3 concurrent embedding prompts bound queue depth and VRAM pressure without
+ * leaving the hardware idle between batches. Raise for multi-endpoint
+ * clients or remote daemons; lower (to 1) for constrained devices.
+ */
+declare const DEFAULT_EMBED_CONCURRENCY = 3;
+interface EmbedBatchOptions {
+    /** Model to embed with (e.g. `'nomic-embed-text:latest'`). */
+    readonly model: string;
+    /** Input strings, in corpus order. Result index `i` matches `input[i]`. */
+    readonly input: readonly string[];
+    /**
+     * Inputs per `/api/embed` request. Default {@link DEFAULT_EMBED_BATCH_SIZE}
+     * (32). Must be a positive integer.
+     */
+    readonly batchSize?: number | undefined;
+    /**
+     * Batches kept in flight simultaneously. Default
+     * {@link DEFAULT_EMBED_CONCURRENCY} (3). Must be a positive integer.
+     */
+    readonly concurrency?: number | undefined;
+    /** Passed through to each batch request — `false` makes oversized inputs
+     * a server-side error instead of a silent truncation. */
+    readonly truncate?: boolean | undefined;
+    /** Passed through to each batch request (dimensionality truncation). */
+    readonly dimensions?: number | undefined;
+    /** Passed through to each batch request (model unload policy). */
+    readonly keep_alive?: string | number | undefined;
+    /** Ollama model options (`num_ctx`, …) passed through to each batch request. */
+    readonly options?: ModelOptions | undefined;
+    /** Aborts the whole batch operation — queued and in-flight alike. */
+    readonly signal?: AbortSignal | undefined;
+    /** Per-request timeout applied to each individual batch request. */
+    readonly timeoutMs?: number | undefined;
+    /**
+     * Progress callback invoked after each successfully embedded batch as
+     * `onBatchComplete(completedBatches, totalBatches)`. Never called after the
+     * operation rejects.
+     */
+    readonly onBatchComplete?: ((completedBatches: number, totalBatches: number) => void) | undefined;
+}
+interface EmbedBatchResult {
+    readonly model: string;
+    /**
+     * Embeddings in input order — `embeddings[i]` corresponds to
+     * `options.input[i]`, independent of batch completion order.
+     */
+    readonly embeddings: readonly (readonly number[])[];
+    /** Number of `/api/embed` requests the corpus was split into. */
+    readonly batchCount: number;
+}
+/** An input string whose estimated token count exceeds the context window. */
+interface OversizedEmbedInput {
+    /** Index into the input array. */
+    readonly index: number;
+    /** Estimated token count for `input[index]`. */
+    readonly estimate: number;
+}
+/**
+ * Finds input strings whose heuristic token estimate exceeds `contextLength`
+ * (typically the resolved `num_ctx`). Unlike chat/generate pre-flight there is
+ * no warning margin and no output-token headroom: an embedding request's
+ * prompt IS the whole input, so the exact window is the trip point, and
+ * crossing it means silent truncation (or, with `truncate: false`, a
+ * server-side rejection).
+ */
+declare function findOversizedEmbedInputs(input: readonly string[], contextLength: number): OversizedEmbedInput[];
+/**
+ * Builds the warn/throw message for oversized embed inputs (mirrors
+ * `contextWarningMessage` for chat/generate).
+ */
+declare function embedBatchOverflowMessage(oversized: readonly OversizedEmbedInput[], totalInputs: number, contextLength: number, truncateFalse: boolean): string;
+/**
+ * Splits `input` into `batchSize` slices and embeds them through `client.embed()`
+ * with at most `concurrency` batches in flight. Order-preserving and fail-fast —
+ * see the module doc for the full semantics.
+ *
+ * Prefer the {@link OllamaClient.embedBatch} wrapper, which additionally
+ * injects `defaultContextLength` as `num_ctx` and enforces the client's
+ * `onContextOverflow` policy against per-string estimates before any request
+ * is sent.
+ */
+declare function batchEmbed(client: OllamaClient, options: EmbedBatchOptions): Promise<EmbedBatchResult>;
+
 /**
  * Ergonomic System One types — key-safe generics on top of the generated
  * contract models.
@@ -1425,6 +1550,14 @@ declare class OllamaClient {
     private readonly contextWarningThreshold;
     /** Configured overflow policy — `'warn'` (default) or `'throw'`. */
     private readonly onContextOverflow;
+    /**
+     * Controllers for every in-flight request and active stream managed by this
+     * client — see {@link OllamaClient.destroy} and {@link createRequestScope}.
+     * Entries live for exactly as long as the underlying work does: until the
+     * response settles (requests) or the stream's `finalResult` settles
+     * (streams, via `executeWithFailover`'s `holdUntil` release path).
+     */
+    private readonly activeRequests;
     constructor(config?: OllamaClientConfig);
     get modelsClient(): ModelsClient;
     /**
@@ -1523,6 +1656,17 @@ declare class OllamaClient {
      * reference when there's nothing to inject (the common case).
      */
     private injectDefaultContextLength;
+    /**
+     * Creates the per-request abort scope registered with {@link destroy}:
+     * a controller that fires when the caller's `signal` aborts OR when
+     * `destroy()` runs, whichever comes first. `dispose()` removes the
+     * registration when the work it guards settles — for plain requests when
+     * the response settles, for streams when `finalResult` settles (the same
+     * moment `executeWithFailover`'s `holdUntil` releases the endpoint slot),
+     * so a destroy() can tear down a stream mid-consumption but never outlives
+     * one that already finished.
+     */
+    private createRequestScope;
     /**
      * Fail-fast guard for `format` (structured output) requests: throws before any network
      * call if the candidate endpoint is inferred as Ollama Cloud, which does not currently
@@ -1585,6 +1729,36 @@ declare class OllamaClient {
     generateWithSchema<T>(req: Omit<GenerateRequestOptions, 'stream' | 'format'>, schema: z.ZodType<T>): Promise<T>;
     embed(req: EmbedRequestOptions): Promise<EmbedResponse>;
     embedText(model: string, input: string | readonly string[]): Promise<readonly (readonly number[])[]>;
+    /**
+     * Batch-constrained, order-preserving embedding generation for large
+     * corpora — the safe building block for RAG/vector-index ingestion at
+     * scale. Splits `input` into `batchSize` slices and embeds them through
+     * {@link embed} with at most `concurrency` batches in flight (defaults:
+     * 32 × 3), so `Promise.all`-style floods that OOM local daemons and
+     * saturate `OLLAMA_MAX_QUEUE` can't happen by accident. Fail-fast: the
+     * first batch error aborts every sibling in-flight batch and rejects the
+     * whole operation with the original error.
+     *
+     * Before any request is sent, `defaultContextLength` is injected as
+     * `num_ctx` (same as chat/generate) and each input string's token estimate
+     * is checked against the resolved window: Ollama silently truncates
+     * oversized embedding inputs by default, so the client warns (or throws,
+     * under `onContextOverflow: 'throw'`) with the offending indexes — see
+     * `src/embed-batch.ts` for the exact trip point (no margin: an embedding
+     * prompt IS the whole input).
+     *
+     * ```ts
+     * const { embeddings, batchCount } = await client.embedBatch({
+     *   model: 'nomic-embed-text:latest',
+     *   input: corpus,                 // e.g. 50k chunks
+     *   batchSize: 32,                 // inputs per /api/embed request
+     *   concurrency: 3,                // batches in flight
+     *   onBatchComplete: (done, total) => progress.log(`${done}/${total}`),
+     * });
+     * // embeddings[i] always corresponds to corpus[i]
+     * ```
+     */
+    embedBatch(options: EmbedBatchOptions): Promise<EmbedBatchResult>;
     /**
      * @deprecated Ollama's `/api/embeddings` endpoint has been superseded by `/api/embed`
      * (exposed here as {@link OllamaClient.embed}), which additionally supports batch
@@ -1710,6 +1884,26 @@ declare class OllamaClient {
     runtimeMode(): RuntimeMode;
     healthCheck(): Promise<EndpointHealthCheckResult[]>;
     endpointStatus(): EndpointHealth[];
+    /**
+     * Immediately aborts every in-flight request and active stream managed by
+     * this client — the clean teardown path for short-lived processes, CLI
+     * runners, and `node:worker_threads` tasks where dangling fetch bodies,
+     * unconsumed stream readers, or queued capacity waiters would otherwise
+     * keep the event loop (and thus the thread/process) alive.
+     *
+     * Every aborted operation rejects with an {@link OllamaAbortError}
+     * (`code: 'aborted'`) carrying `reason` as its message — including requests
+     * still queued behind `maxConcurrentPerEndpoint`, not just dispatched ones.
+     * Streams reject their `finalResult` and release their endpoint slots.
+     *
+     * Idempotent: calling it again (or on a client with nothing in flight)
+     * is a no-op that returns `0`. The client remains usable afterward —
+     * `destroy()` is a drain, not a permanent disable; callers that want a
+     * permanently-dead client should discard the instance.
+     *
+     * @returns the number of in-flight operations that were aborted.
+     */
+    destroy(reason?: string): number;
     get openai(): OpenAICompatClient;
     get anthropic(): AnthropicCompatClient;
 }
@@ -2657,4 +2851,4 @@ declare const OLLAMA_FALLBACK_CONTEXT_LENGTH = 2048;
  */
 declare function contextWarningMessage(check: ContextCheck, kind: 'chat' | 'generate'): string;
 
-export { ATTR_GEN_AI_OPERATION_NAME, ATTR_GEN_AI_REQUEST_MODEL, ATTR_GEN_AI_RESPONSE_MODEL, ATTR_GEN_AI_SYSTEM, ATTR_GEN_AI_TOOL_CALL_ID, ATTR_GEN_AI_TOOL_NAME, ATTR_GEN_AI_USAGE_INPUT_TOKENS, ATTR_GEN_AI_USAGE_OUTPUT_TOKENS, ATTR_HTTP_REQUEST_METHOD, ATTR_HTTP_RESPONSE_STATUS_CODE, ATTR_OLLAMA_AGENT_ITERATION, ATTR_OLLAMA_AGENT_MAX_ITERATIONS, ATTR_OLLAMA_ENDPOINT_ATTEMPT, ATTR_OLLAMA_ENDPOINT_NAME, ATTR_SERVER_ADDRESS, ATTR_SERVER_PORT, ATTR_URL_FULL, AbortableAsyncIterable, Agent, type AgentChatClient, type AgentConfig, type AgentHooks, type AgentResult, type AgentRunInput, type AgentTurn, type AnswerFor, AnthropicCompatClient, AnyTool, type BackoffOptions, BalanceRequestOptions, type BalanceResponse, type BlobUploadResult, ChatRequestOptions, ChatResponse, ChatStreamResult, type ChoiceDecision, type ContextCheck, type ContextCheckOptions, type ContextWindowSource, type ConversationSendOptions, ConversationSession, type ConversationSessionOptions, CopyRequestOptions, CreateRequestOptions, type CumulativeCacheStats, DEFAULT_BACKOFF, DEFAULT_BASE_URL, DEFAULT_CONTEXT_WARNING_THRESHOLD, DEFAULT_FAILOVER_CODES, DEFAULT_FETCH_BACKOFF_CONFIG, DEFAULT_RETRY_CONFIG, DEFAULT_TIMEOUT_MS, type Decision, type DefineToolOptions, DeleteRequestOptions, EmbedRequestOptions, EmbedResponse, EmbeddingsRequestOptions, EmbeddingsResponse, type EndpointHealth, type EndpointHealthCheckResult, EndpointRegistry, type EndpointRegistryOptions, FetchLike, type FetchWithBackoffConfig, FormatOption, GEN_AI_SYSTEM_OLLAMA, GenerateRequestOptions, GenerateResponse, GenerateStreamResult, HttpClient, IMAGE_TOKEN_ESTIMATE, type IncludedBalance, KEEP_ALIVE_INDEFINITE, KEEP_ALIVE_UNLOAD, KNOWN_DONE_REASONS, type KeepAlive, type KnownDoneReason, type LegacyBalanceLimit, type LegacyIncludedBalance, type LoadMcpToolsOptions, Logger, MAX_SYSTEM_ONE_IMAGES_BYTES, MAX_SYSTEM_ONE_REQUEST_BYTES, McpBridge, type McpBridgeOptions, McpCallToolResult, McpClientLike, McpTask, McpToolDescriptor, Message, Middleware, type ModelCapabilities, ModelOptions, ModelResponse, ModelsClient, type NoulDecision, OLLAMA_CLOUD_BASE_URL, OLLAMA_FALLBACK_CONTEXT_LENGTH, OllamaClient, type OllamaClientConfig, type OllamaCloudFreeTierQuotaBudgets, type OllamaCredentialConfig, type OllamaEndpoint, OllamaStream, OllamaStreamEvent, OllamaStreamEventType, OpenAICompatClient, type ParallelToolCallBehavior, ProgressResponse, ProgressStreamResult, PsResponse, PullRequestOptions, PushRequestOptions, QuotaManager, type QuotaManagerOptions, type QuotaWindowConfig, type QuotaWindowStatus, RETRYABLE_STATUS_CODES, type RankResult, RawUsageSource, RequestCancellationOptions, RequestLifecycleHook, RequestRunner, type ResponsesCreateRequest, type ResponsesCreateResponse, ResponsesModule, type ResponsesReasoningEffort, type ResponsesStreamEvent, type ResponsesToolDefinition, type ResponsesUsage, type RetryConfig, type RuntimeMode, type ScoreDecision, type SessionTurn, ShowRequestOptions, ShowResponse, type SpanAttributeValue, type SpanAttributes, StatusResponse, type SupportedSchema, type SystemOneAnswer, type SystemOneAnswers, SystemOneChoiceAnswer, SystemOneChoiceQuestion, SystemOneContent, SystemOneNoulAnswer, SystemOneNoulQuestion, type SystemOneQuestion, type SystemOneQuestions, type SystemOneRequest, type SystemOneResponse, SystemOneScoreAnswer, SystemOneScoreQuestion, type SystemOneUsage, ThinkValue, ThinkingMetadata, type TimeoutSignal, TokenUsage, Tool, ToolCall, ToolCallAccumulator, ToolDefinition, ToolExecutionResult, ToolHandler, ToolParameters, ToolRegistry, type TurnCacheStats, type UsageBucket, type UsageMetrics, UsageRequestOptions, type UsageResponse, VersionResponse, VisionInput, WebFetchRequestOptions, WebFetchResponse, WebSearchRequestOptions, WebSearchResponse, addSpanEvent, calculateBackoff, checkChatContext, checkEndpointHealth, checkGenerateContext, contextWarningMessage, createDecision, createOllamaCloudFreeTierQuota, createTimeoutSignal, defineTool, detectModelCapabilities, encodeImage, estimateChatRequestTokens, estimateGenerateRequestTokens, estimateSystemOneRequestBytes, estimateTokens, fetchWithBackoff, inferRuntimeMode, isKeepAliveSugar, isKnownDoneReason, isSameToolCall, listAllMcpTools, listAvailableModels, loadMcpTools, mergeToolCall, mergeToolCallArgumentsString, mergeToolCallArrays, normalizeChatStream, normalizeGenerateStream, normalizeKeepAlive, normalizeProgressStream, parseNdjsonStream, parseStructuredOutput, registerMcpTools, resolveCredentialEndpoints, setSpanError, toDataStream, toResponse, toTextStream, withRetry, withSpan, zodToJsonSchema };
+export { ATTR_GEN_AI_OPERATION_NAME, ATTR_GEN_AI_REQUEST_MODEL, ATTR_GEN_AI_RESPONSE_MODEL, ATTR_GEN_AI_SYSTEM, ATTR_GEN_AI_TOOL_CALL_ID, ATTR_GEN_AI_TOOL_NAME, ATTR_GEN_AI_USAGE_INPUT_TOKENS, ATTR_GEN_AI_USAGE_OUTPUT_TOKENS, ATTR_HTTP_REQUEST_METHOD, ATTR_HTTP_RESPONSE_STATUS_CODE, ATTR_OLLAMA_AGENT_ITERATION, ATTR_OLLAMA_AGENT_MAX_ITERATIONS, ATTR_OLLAMA_ENDPOINT_ATTEMPT, ATTR_OLLAMA_ENDPOINT_NAME, ATTR_SERVER_ADDRESS, ATTR_SERVER_PORT, ATTR_URL_FULL, AbortableAsyncIterable, Agent, type AgentChatClient, type AgentConfig, type AgentHooks, type AgentResult, type AgentRunInput, type AgentTurn, type AnswerFor, AnthropicCompatClient, AnyTool, type BackoffOptions, BalanceRequestOptions, type BalanceResponse, type BlobUploadResult, ChatRequestOptions, ChatResponse, ChatStreamResult, type ChoiceDecision, type ContextCheck, type ContextCheckOptions, type ContextWindowSource, type ConversationSendOptions, ConversationSession, type ConversationSessionOptions, CopyRequestOptions, CreateRequestOptions, type CumulativeCacheStats, DEFAULT_BACKOFF, DEFAULT_BASE_URL, DEFAULT_CONTEXT_WARNING_THRESHOLD, DEFAULT_EMBED_BATCH_SIZE, DEFAULT_EMBED_CONCURRENCY, DEFAULT_FAILOVER_CODES, DEFAULT_FETCH_BACKOFF_CONFIG, DEFAULT_RETRY_CONFIG, DEFAULT_TIMEOUT_MS, type Decision, type DefineToolOptions, DeleteRequestOptions, type EmbedBatchOptions, type EmbedBatchResult, EmbedRequestOptions, EmbedResponse, EmbeddingsRequestOptions, EmbeddingsResponse, type EndpointHealth, type EndpointHealthCheckResult, EndpointRegistry, type EndpointRegistryOptions, FetchLike, type FetchWithBackoffConfig, FormatOption, GEN_AI_SYSTEM_OLLAMA, GenerateRequestOptions, GenerateResponse, GenerateStreamResult, HttpClient, IMAGE_TOKEN_ESTIMATE, type IncludedBalance, KEEP_ALIVE_INDEFINITE, KEEP_ALIVE_UNLOAD, KNOWN_DONE_REASONS, type KeepAlive, type KnownDoneReason, type LegacyBalanceLimit, type LegacyIncludedBalance, type LoadMcpToolsOptions, Logger, MAX_SYSTEM_ONE_IMAGES_BYTES, MAX_SYSTEM_ONE_REQUEST_BYTES, McpBridge, type McpBridgeOptions, McpCallToolResult, McpClientLike, McpTask, McpToolDescriptor, Message, Middleware, type ModelCapabilities, ModelOptions, ModelResponse, ModelsClient, type NoulDecision, OLLAMA_CLOUD_BASE_URL, OLLAMA_FALLBACK_CONTEXT_LENGTH, OllamaClient, type OllamaClientConfig, type OllamaCloudFreeTierQuotaBudgets, type OllamaCredentialConfig, type OllamaEndpoint, OllamaStream, OllamaStreamEvent, OllamaStreamEventType, OpenAICompatClient, type OversizedEmbedInput, type ParallelToolCallBehavior, ProgressResponse, ProgressStreamResult, PsResponse, PullRequestOptions, PushRequestOptions, QuotaManager, type QuotaManagerOptions, type QuotaWindowConfig, type QuotaWindowStatus, RETRYABLE_STATUS_CODES, type RankResult, RawUsageSource, RequestCancellationOptions, RequestLifecycleHook, RequestRunner, type ResponsesCreateRequest, type ResponsesCreateResponse, ResponsesModule, type ResponsesReasoningEffort, type ResponsesStreamEvent, type ResponsesToolDefinition, type ResponsesUsage, type RetryConfig, type RuntimeMode, type ScoreDecision, type SessionTurn, ShowRequestOptions, ShowResponse, type SpanAttributeValue, type SpanAttributes, StatusResponse, type SupportedSchema, type SystemOneAnswer, type SystemOneAnswers, SystemOneChoiceAnswer, SystemOneChoiceQuestion, SystemOneContent, SystemOneNoulAnswer, SystemOneNoulQuestion, type SystemOneQuestion, type SystemOneQuestions, type SystemOneRequest, type SystemOneResponse, SystemOneScoreAnswer, SystemOneScoreQuestion, type SystemOneUsage, ThinkValue, ThinkingMetadata, type TimeoutSignal, TokenUsage, Tool, ToolCall, ToolCallAccumulator, ToolDefinition, ToolExecutionResult, ToolHandler, ToolParameters, ToolRegistry, type TurnCacheStats, type UsageBucket, type UsageMetrics, UsageRequestOptions, type UsageResponse, VersionResponse, VisionInput, WebFetchRequestOptions, WebFetchResponse, WebSearchRequestOptions, WebSearchResponse, addSpanEvent, batchEmbed, calculateBackoff, checkChatContext, checkEndpointHealth, checkGenerateContext, contextWarningMessage, createDecision, createOllamaCloudFreeTierQuota, createTimeoutSignal, defineTool, detectModelCapabilities, embedBatchOverflowMessage, encodeImage, estimateChatRequestTokens, estimateGenerateRequestTokens, estimateSystemOneRequestBytes, estimateTokens, fetchWithBackoff, findOversizedEmbedInputs, inferRuntimeMode, isKeepAliveSugar, isKnownDoneReason, isSameToolCall, listAllMcpTools, listAvailableModels, loadMcpTools, mergeToolCall, mergeToolCallArgumentsString, mergeToolCallArrays, normalizeChatStream, normalizeGenerateStream, normalizeKeepAlive, normalizeProgressStream, parseNdjsonStream, parseStructuredOutput, registerMcpTools, resolveCredentialEndpoints, setSpanError, toDataStream, toResponse, toTextStream, withRetry, withSpan, zodToJsonSchema };

package/dist/index.d.ts

@@ -1,12 +1,12 @@
-import { F as FetchLike, M as Middleware, L as Logger, R as RequestLifecycleHook, H as HttpClient, c as OllamaStreamEvent, A as AbortableAsyncIterable, d as OllamaStreamEventType, P as ProgressStreamResult, B as BinaryBody, O as OllamaRuntime, C as ChatStreamResult, G as GenerateStreamResult, e as RawUsageSource, T as TokenUsage } from './runtime-C3rfpkZw.js';
-export { D as DoneEvent, E as ErrorEvent, f as HttpBody, g as HttpClientOptions, h as HttpRequestOptions, i as LifecycleErrorEvent, j as LifecycleRetryEvent, k as LifecycleStartEvent, l as LifecycleSuccessEvent, m as LogFn, n as MessageEvent, o as MiddlewareContext, N as NANOS_PER_MS, p as NANOS_PER_SECOND, q as NOOP_LOGGER, r as NextFunction, s as OllamaAbortError, t as OllamaAgentMaxIterationsError, u as OllamaAgentMaxToolCallsError, v as OllamaAuthError, w as OllamaBadGatewayError, x as OllamaClientError, y as OllamaClientErrorOptions, z as OllamaErrorRequestContext, J as OllamaErrorResponseContext, K as OllamaGenericClientError, Q as OllamaIncompatibleModelError, U as OllamaMcpError, V as OllamaModelRoutingError, W as OllamaNetworkError, X as OllamaNotFoundError, Y as OllamaOverloadedError, Z as OllamaQuotaExceededError, _ as OllamaRateLimitError, $ as OllamaRequestTooLargeError, a0 as OllamaRequestValidationError, a1 as OllamaResponseValidationError, a2 as OllamaServerError, a3 as OllamaServerVersionUnknownError, a4 as OllamaSkillInvalidError, a5 as OllamaSkillNotFoundError, a6 as OllamaStreamError, a7 as OllamaTimeoutError, a8 as OllamaToolTimeoutError, a9 as OllamaToolValidationError, aa as OllamaUnsupportedCapabilityError, ab as RequestContext, ac as RequestLifecycleEvent, ad as ResponseContext, ae as SseEvent, af as ThinkingEvent, ag as TokenEvent, ah as ToolCallEvent, ai as composeMiddleware, aj as createConsoleLogger, ak as extractUsage, al as mapError, am as parseSseStream } from './runtime-C3rfpkZw.js';
-import { T as ThinkingMetadata, M as ModelResponse, S as ShowRequestOptions, a as ShowResponse, P as PullRequestOptions, b as ProgressResponse, c as PushRequestOptions, C as CreateRequestOptions, D as DeleteRequestOptions, d as StatusResponse, e as CopyRequestOptions, f as PsResponse, V as VersionResponse, R as RequestCancellationOptions, g as ToolParameters, h as ThinkValue, i as ToolCall, j as ChatRequestOptions, k as ModelOptions, l as ToolDefinition, F as FormatOption, m as ChatResponse, n as Message, G as GenerateRequestOptions, o as GenerateResponse, E as EmbedRequestOptions, p as EmbedResponse, q as EmbeddingsRequestOptions, r as EmbeddingsResponse, W as WebSearchRequestOptions, s as WebSearchResponse, t as WebFetchRequestOptions, u as WebFetchResponse, U as UsageRequestOptions, B as BalanceRequestOptions, v as VisionInput } from './types-DVcX_zMp.js';
-export { L as ListResponse, w as Logprob, x as LogprobToken, y as ModelDetails, z as Role, A as ToolCallFunction, H as ToolFunctionDefinition, I as ToolProperty, J as UsageRange, K as UsageScope, N as WebSearchResult, O as imageStringNeedsResolution, Q as resolveImageInput } from './types-DVcX_zMp.js';
+import { F as FetchLike, M as Middleware, L as Logger, R as RequestLifecycleHook, H as HttpClient, c as OllamaStreamEvent, A as AbortableAsyncIterable, d as OllamaStreamEventType, P as ProgressStreamResult, B as BinaryBody, O as OllamaRuntime, C as ChatStreamResult, G as GenerateStreamResult, e as RawUsageSource, T as TokenUsage } from './runtime-BV9SK_cf.js';
+export { D as DoneEvent, E as ErrorEvent, f as HttpBody, g as HttpClientOptions, h as HttpRequestOptions, i as LifecycleErrorEvent, j as LifecycleRetryEvent, k as LifecycleStartEvent, l as LifecycleSuccessEvent, m as LogFn, n as MessageEvent, o as MiddlewareContext, N as NANOS_PER_MS, p as NANOS_PER_SECOND, q as NOOP_LOGGER, r as NextFunction, s as OllamaAbortError, t as OllamaAgentMaxIterationsError, u as OllamaAgentMaxToolCallsError, v as OllamaAuthError, w as OllamaBadGatewayError, x as OllamaClientError, y as OllamaClientErrorOptions, z as OllamaErrorRequestContext, J as OllamaErrorResponseContext, K as OllamaGenericClientError, Q as OllamaIncompatibleModelError, U as OllamaMcpError, V as OllamaModelRoutingError, W as OllamaNetworkError, X as OllamaNotFoundError, Y as OllamaOverloadedError, Z as OllamaQuotaExceededError, _ as OllamaRateLimitError, $ as OllamaRequestTooLargeError, a0 as OllamaRequestValidationError, a1 as OllamaResponseValidationError, a2 as OllamaServerError, a3 as OllamaServerVersionUnknownError, a4 as OllamaSkillInvalidError, a5 as OllamaSkillNotFoundError, a6 as OllamaStreamError, a7 as OllamaTimeoutError, a8 as OllamaToolTimeoutError, a9 as OllamaToolValidationError, aa as OllamaUnsupportedCapabilityError, ab as RequestContext, ac as RequestLifecycleEvent, ad as ResponseContext, ae as SseEvent, af as ThinkingEvent, ag as TokenEvent, ah as ToolCallEvent, ai as composeMiddleware, aj as createConsoleLogger, ak as extractUsage, al as mapError, am as parseSseStream } from './runtime-BV9SK_cf.js';
+import { T as ThinkingMetadata, M as ModelResponse, S as ShowRequestOptions, a as ShowResponse, P as PullRequestOptions, b as ProgressResponse, c as PushRequestOptions, C as CreateRequestOptions, D as DeleteRequestOptions, d as StatusResponse, e as CopyRequestOptions, f as PsResponse, V as VersionResponse, R as RequestCancellationOptions, g as ToolParameters, h as ThinkValue, i as ToolCall, j as ChatRequestOptions, k as ModelOptions, l as ToolDefinition, F as FormatOption, m as ChatResponse, n as Message, G as GenerateRequestOptions, o as GenerateResponse, E as EmbedRequestOptions, p as EmbedResponse, q as EmbeddingsRequestOptions, r as EmbeddingsResponse, W as WebSearchRequestOptions, s as WebSearchResponse, t as WebFetchRequestOptions, u as WebFetchResponse, U as UsageRequestOptions, B as BalanceRequestOptions, v as VisionInput } from './types-BtQexFpT.js';
+export { L as ListResponse, w as Logprob, x as LogprobToken, y as ModelDetails, z as Role, A as ToolCallFunction, H as ToolFunctionDefinition, I as ToolProperty, J as UsageRange, K as UsageScope, N as WebSearchResult, O as imageStringNeedsResolution, Q as resolveImageInput } from './types-BtQexFpT.js';
 import { z } from 'zod';
-import { R as RequestRunner, S as SystemOneChoiceAnswer, o as SystemOneNoulAnswer, p as SystemOneScoreAnswer, q as SystemOneChoiceQuestion, r as SystemOneNoulQuestion, s as SystemOneScoreQuestion, t as SystemOneRequest$1, u as SystemOneResponse$1, v as SystemOneContent, N as NativeApi, w as OpenAICompatClient, x as AnthropicCompatClient } from './anthropic-5oYIRS6D.js';
-export { y as AnthropicCacheControl, z as AnthropicContentBlock, B as AnthropicImageContentBlock, C as AnthropicMessage, n as AnthropicMessageStreamEvent, A as AnthropicMessagesRequest, m as AnthropicMessagesResponse, D as AnthropicMessagesStream, E as AnthropicMetadata, F as AnthropicOutputConfig, G as AnthropicRedactedThinkingContentBlock, H as AnthropicSystem, I as AnthropicSystemTextBlock, J as AnthropicTextContentBlock, K as AnthropicThinkingConfig, L as AnthropicThinkingContentBlock, M as AnthropicTool, P as AnthropicToolChoice, Q as AnthropicToolResultContentBlock, T as AnthropicToolUseContentBlock, U as AnthropicUnknownStreamEvent, V as OllamaAnthropicContentBlock, W as OllamaAnthropicMessage, X as OllamaAnthropicMessagesRequest, Y as OllamaAnthropicSystem, Z as OllamaAnthropicSystemTextBlock, _ as OllamaAnthropicTextContentBlock, $ as OllamaAnthropicThinkingConfig, a0 as OllamaAnthropicTool, a1 as OllamaAnthropicToolResultContentBlock, a2 as OllamaOpenAIChatCompletionRequest, a3 as OllamaOpenAICompletionRequest, a4 as OllamaOpenAIEmbeddingRequest, a5 as OllamaOpenAIReasoningEffort, a6 as OllamaOpenAIResponsesRequest, a7 as OpenAIChatCompletionChoice, b as OpenAIChatCompletionChunk, a8 as OpenAIChatCompletionChunkChoice, a9 as OpenAIChatCompletionDelta, O as OpenAIChatCompletionRequest, a as OpenAIChatCompletionResponse, aa as OpenAIChatCompletionStream, ab as OpenAICompletionChoice, e as OpenAICompletionChunk, c as OpenAICompletionRequest, d as OpenAICompletionResponse, ac as OpenAICompletionStream, ad as OpenAIContentPart, ae as OpenAIEmbeddingInput, af as OpenAIEmbeddingItem, f as OpenAIEmbeddingRequest, g as OpenAIEmbeddingResponse, ag as OpenAIFunctionDefinition, ah as OpenAIImageUrlContentPart, h as OpenAIListModelsResponse, ai as OpenAIMessage, i as OpenAIModelItem, aj as OpenAIReasoningEffort, ak as OpenAIResponseFormat, al as OpenAIResponsesCompletedEvent, am as OpenAIResponsesContentPartAddedEvent, an as OpenAIResponsesContentPartDoneEvent, ao as OpenAIResponsesCreatedEvent, ap as OpenAIResponsesError, aq as OpenAIResponsesEventBase, ar as OpenAIResponsesFailedEvent, as as OpenAIResponsesFunctionCallArgumentsDeltaEvent, at as OpenAIResponsesFunctionCallArgumentsDoneEvent, au as OpenAIResponsesInProgressEvent, av as OpenAIResponsesIncompleteDetails, aw as OpenAIResponsesIncompleteEvent, ax as OpenAIResponsesOutputContent, ay as OpenAIResponsesOutputFunctionCall, az as OpenAIResponsesOutputItem, aA as OpenAIResponsesOutputItemAddedEvent, aB as OpenAIResponsesOutputItemDoneEvent, aC as OpenAIResponsesOutputMessage, aD as OpenAIResponsesOutputReasoning, aE as OpenAIResponsesOutputRefusal, aF as OpenAIResponsesOutputTextContent, aG as OpenAIResponsesOutputTextDeltaEvent, aH as OpenAIResponsesOutputTextDoneEvent, aI as OpenAIResponsesQueuedEvent, aJ as OpenAIResponsesReasoningSummaryPartAddedEvent, aK as OpenAIResponsesReasoningSummaryPartDoneEvent, aL as OpenAIResponsesReasoningSummaryTextDeltaEvent, aM as OpenAIResponsesReasoningSummaryTextDoneEvent, aN as OpenAIResponsesReasoningTextDeltaEvent, aO as OpenAIResponsesReasoningTextDoneEvent, aP as OpenAIResponsesRefusalDeltaEvent, aQ as OpenAIResponsesRefusalDoneEvent, j as OpenAIResponsesRequest, k as OpenAIResponsesResponse, aR as OpenAIResponsesStatus, aS as OpenAIResponsesStream, aT as OpenAIResponsesStreamError, l as OpenAIResponsesStreamEvent, aU as OpenAIResponsesUsage, aV as OpenAIStreamOptions, aW as OpenAITextContentPart, aX as OpenAITool, aY as OpenAIToolCall, aZ as OpenAIToolCallDelta, a_ as OpenAIUsage, a$ as SystemOneConfidence, b0 as SystemOneProbabilities } from './anthropic-5oYIRS6D.js';
-import { T as ToolHandler, a as Tool, A as AnyTool, b as ToolRegistry, c as ToolExecutionResult } from './compose-DCVYMvOL.js';
-export { d as AppliedSkillResult, e as ApplySkillOptions, P as ParsedFrontmatter, S as Skill, f as SkillFrontmatter, g as SkillSummary, h as ToolExecutionContext, i as ToolExecutionFailure, j as ToolExecutionSuccess, k as ToolRegistryOptions, l as applySkill, p as parseFrontmatter } from './compose-DCVYMvOL.js';
+import { R as RequestRunner, S as SystemOneChoiceAnswer, o as SystemOneNoulAnswer, p as SystemOneScoreAnswer, q as SystemOneChoiceQuestion, r as SystemOneNoulQuestion, s as SystemOneScoreQuestion, t as SystemOneRequest$1, u as SystemOneResponse$1, v as SystemOneContent, N as NativeApi, w as OpenAICompatClient, x as AnthropicCompatClient } from './anthropic-CPaoxeLX.js';
+export { y as AnthropicCacheControl, z as AnthropicContentBlock, B as AnthropicImageContentBlock, C as AnthropicMessage, n as AnthropicMessageStreamEvent, A as AnthropicMessagesRequest, m as AnthropicMessagesResponse, D as AnthropicMessagesStream, E as AnthropicMetadata, F as AnthropicOutputConfig, G as AnthropicRedactedThinkingContentBlock, H as AnthropicSystem, I as AnthropicSystemTextBlock, J as AnthropicTextContentBlock, K as AnthropicThinkingConfig, L as AnthropicThinkingContentBlock, M as AnthropicTool, P as AnthropicToolChoice, Q as AnthropicToolResultContentBlock, T as AnthropicToolUseContentBlock, U as AnthropicUnknownStreamEvent, V as OllamaAnthropicContentBlock, W as OllamaAnthropicMessage, X as OllamaAnthropicMessagesRequest, Y as OllamaAnthropicSystem, Z as OllamaAnthropicSystemTextBlock, _ as OllamaAnthropicTextContentBlock, $ as OllamaAnthropicThinkingConfig, a0 as OllamaAnthropicTool, a1 as OllamaAnthropicToolResultContentBlock, a2 as OllamaOpenAIChatCompletionRequest, a3 as OllamaOpenAICompletionRequest, a4 as OllamaOpenAIEmbeddingRequest, a5 as OllamaOpenAIReasoningEffort, a6 as OllamaOpenAIResponsesRequest, a7 as OpenAIChatCompletionChoice, b as OpenAIChatCompletionChunk, a8 as OpenAIChatCompletionChunkChoice, a9 as OpenAIChatCompletionDelta, O as OpenAIChatCompletionRequest, a as OpenAIChatCompletionResponse, aa as OpenAIChatCompletionStream, ab as OpenAICompletionChoice, e as OpenAICompletionChunk, c as OpenAICompletionRequest, d as OpenAICompletionResponse, ac as OpenAICompletionStream, ad as OpenAIContentPart, ae as OpenAIEmbeddingInput, af as OpenAIEmbeddingItem, f as OpenAIEmbeddingRequest, g as OpenAIEmbeddingResponse, ag as OpenAIFunctionDefinition, ah as OpenAIImageUrlContentPart, h as OpenAIListModelsResponse, ai as OpenAIMessage, i as OpenAIModelItem, aj as OpenAIReasoningEffort, ak as OpenAIResponseFormat, al as OpenAIResponsesCompletedEvent, am as OpenAIResponsesContentPartAddedEvent, an as OpenAIResponsesContentPartDoneEvent, ao as OpenAIResponsesCreatedEvent, ap as OpenAIResponsesError, aq as OpenAIResponsesEventBase, ar as OpenAIResponsesFailedEvent, as as OpenAIResponsesFunctionCallArgumentsDeltaEvent, at as OpenAIResponsesFunctionCallArgumentsDoneEvent, au as OpenAIResponsesInProgressEvent, av as OpenAIResponsesIncompleteDetails, aw as OpenAIResponsesIncompleteEvent, ax as OpenAIResponsesOutputContent, ay as OpenAIResponsesOutputFunctionCall, az as OpenAIResponsesOutputItem, aA as OpenAIResponsesOutputItemAddedEvent, aB as OpenAIResponsesOutputItemDoneEvent, aC as OpenAIResponsesOutputMessage, aD as OpenAIResponsesOutputReasoning, aE as OpenAIResponsesOutputRefusal, aF as OpenAIResponsesOutputTextContent, aG as OpenAIResponsesOutputTextDeltaEvent, aH as OpenAIResponsesOutputTextDoneEvent, aI as OpenAIResponsesQueuedEvent, aJ as OpenAIResponsesReasoningSummaryPartAddedEvent, aK as OpenAIResponsesReasoningSummaryPartDoneEvent, aL as OpenAIResponsesReasoningSummaryTextDeltaEvent, aM as OpenAIResponsesReasoningSummaryTextDoneEvent, aN as OpenAIResponsesReasoningTextDeltaEvent, aO as OpenAIResponsesReasoningTextDoneEvent, aP as OpenAIResponsesRefusalDeltaEvent, aQ as OpenAIResponsesRefusalDoneEvent, j as OpenAIResponsesRequest, k as OpenAIResponsesResponse, aR as OpenAIResponsesStatus, aS as OpenAIResponsesStream, aT as OpenAIResponsesStreamError, l as OpenAIResponsesStreamEvent, aU as OpenAIResponsesUsage, aV as OpenAIStreamOptions, aW as OpenAITextContentPart, aX as OpenAITool, aY as OpenAIToolCall, aZ as OpenAIToolCallDelta, a_ as OpenAIUsage, a$ as SystemOneConfidence, b0 as SystemOneProbabilities } from './anthropic-CPaoxeLX.js';
+import { T as ToolHandler, a as Tool, A as AnyTool, b as ToolRegistry, c as ToolExecutionResult } from './compose-BQYXdpar.js';
+export { d as AppliedSkillResult, e as ApplySkillOptions, P as ParsedFrontmatter, S as Skill, f as SkillFrontmatter, g as SkillSummary, h as ToolExecutionContext, i as ToolExecutionFailure, j as ToolExecutionSuccess, k as ToolRegistryOptions, l as applySkill, p as parseFrontmatter } from './compose-BQYXdpar.js';
 import { M as McpClientLike, a as McpToolDescriptor, b as McpTask, c as McpCallToolResult } from './types-CI-gM2u5.js';
 export { d as McpContentBlock, e as McpCreateTaskResult, f as McpElicitationFormRequest, g as McpElicitationHandlers, h as McpElicitationRequest, i as McpElicitationResult, j as McpElicitationUrlRequest, k as McpIcon, l as McpInputRequiredResult, m as McpListToolsParams, n as McpListToolsResult, o as McpRequestOptions, p as McpServerCapabilities, q as McpTaskMethod, r as McpTaskRequest, s as McpTaskStatus, t as McpToolAnnotations, u as McpToolCallResult, v as McpToolExecution } from './types-CI-gM2u5.js';
 import * as Otel from '@opentelemetry/api';
@@ -981,6 +981,131 @@ declare class ConversationSession {
     reset(): void;
 }
 
+/**
+ * Batch-constrained embedding generation for high-volume ingestion pipelines.
+ *
+ * Ollama's `/api/embed` accepts an array of inputs, but two independent limits
+ * make naive usage fragile at corpus scale:
+ *
+ * 1. **Concurrency.** `Promise.all(inputs.map(...))` fires every request at
+ *    once; against a local Ollama daemon this contends sockets, saturates the
+ *    request queue (`OLLAMA_MAX_QUEUE`, surfacing as 503 `overloaded`), and —
+ *    worst case — exhausts GPU VRAM as the model instance serves N prompts
+ *    simultaneously.
+ * 2. **Context window.** Ollama truncates any input string longer than the
+ *    model's context window (`num_ctx`) *silently* unless `truncate: false` is
+ *    set — degraded vectors with no error, no warning, nothing on the wire.
+ *
+ * {@link batchEmbed} addresses the first with a fixed-size worker pool over
+ * `client.embed()` (so every batch still rides the client's full pipeline:
+ * failover, retry, auth, logging, telemetry). The client-level
+ * `embedBatch()` wrapper addresses the second with a pre-flight per-string
+ * token estimate against the resolved `num_ctx` window (see
+ * {@link findOversizedEmbedInputs}).
+ *
+ * Semantics:
+ * - **Order-preserving** — result index `i` is always the embedding of
+ *   `input[i]`, regardless of which batch completed first.
+ * - **Fail-fast** — the first batch error aborts every sibling in-flight
+ *   batch (surfaces the *original* error, not the derived aborts) and rejects
+ *   the whole operation. Vector indexing that half-succeeds is a corrupted
+ *   index; callers re-running an ingestion prefer all-or-nothing.
+ * - **Abortable** — the caller's `signal` cancels queued and in-flight batches
+ *   alike and rejects with `code: 'aborted'`.
+ */
+
+/**
+ * Default number of input strings per `/api/embed` request. 32 keeps a single
+ * request's prompt comfortably inside typical embedding-model context windows
+ * (2048+) while amortizing HTTP overhead across the corpus.
+ */
+declare const DEFAULT_EMBED_BATCH_SIZE = 32;
+/**
+ * Default number of batches kept in flight concurrently. 3 is deliberately
+ * conservative: a local Ollama daemon serves one model instance per GPU, so
+ * 3 concurrent embedding prompts bound queue depth and VRAM pressure without
+ * leaving the hardware idle between batches. Raise for multi-endpoint
+ * clients or remote daemons; lower (to 1) for constrained devices.
+ */
+declare const DEFAULT_EMBED_CONCURRENCY = 3;
+interface EmbedBatchOptions {
+    /** Model to embed with (e.g. `'nomic-embed-text:latest'`). */
+    readonly model: string;
+    /** Input strings, in corpus order. Result index `i` matches `input[i]`. */
+    readonly input: readonly string[];
+    /**
+     * Inputs per `/api/embed` request. Default {@link DEFAULT_EMBED_BATCH_SIZE}
+     * (32). Must be a positive integer.
+     */
+    readonly batchSize?: number | undefined;
+    /**
+     * Batches kept in flight simultaneously. Default
+     * {@link DEFAULT_EMBED_CONCURRENCY} (3). Must be a positive integer.
+     */
+    readonly concurrency?: number | undefined;
+    /** Passed through to each batch request — `false` makes oversized inputs
+     * a server-side error instead of a silent truncation. */
+    readonly truncate?: boolean | undefined;
+    /** Passed through to each batch request (dimensionality truncation). */
+    readonly dimensions?: number | undefined;
+    /** Passed through to each batch request (model unload policy). */
+    readonly keep_alive?: string | number | undefined;
+    /** Ollama model options (`num_ctx`, …) passed through to each batch request. */
+    readonly options?: ModelOptions | undefined;
+    /** Aborts the whole batch operation — queued and in-flight alike. */
+    readonly signal?: AbortSignal | undefined;
+    /** Per-request timeout applied to each individual batch request. */
+    readonly timeoutMs?: number | undefined;
+    /**
+     * Progress callback invoked after each successfully embedded batch as
+     * `onBatchComplete(completedBatches, totalBatches)`. Never called after the
+     * operation rejects.
+     */
+    readonly onBatchComplete?: ((completedBatches: number, totalBatches: number) => void) | undefined;
+}
+interface EmbedBatchResult {
+    readonly model: string;
+    /**
+     * Embeddings in input order — `embeddings[i]` corresponds to
+     * `options.input[i]`, independent of batch completion order.
+     */
+    readonly embeddings: readonly (readonly number[])[];
+    /** Number of `/api/embed` requests the corpus was split into. */
+    readonly batchCount: number;
+}
+/** An input string whose estimated token count exceeds the context window. */
+interface OversizedEmbedInput {
+    /** Index into the input array. */
+    readonly index: number;
+    /** Estimated token count for `input[index]`. */
+    readonly estimate: number;
+}
+/**
+ * Finds input strings whose heuristic token estimate exceeds `contextLength`
+ * (typically the resolved `num_ctx`). Unlike chat/generate pre-flight there is
+ * no warning margin and no output-token headroom: an embedding request's
+ * prompt IS the whole input, so the exact window is the trip point, and
+ * crossing it means silent truncation (or, with `truncate: false`, a
+ * server-side rejection).
+ */
+declare function findOversizedEmbedInputs(input: readonly string[], contextLength: number): OversizedEmbedInput[];
+/**
+ * Builds the warn/throw message for oversized embed inputs (mirrors
+ * `contextWarningMessage` for chat/generate).
+ */
+declare function embedBatchOverflowMessage(oversized: readonly OversizedEmbedInput[], totalInputs: number, contextLength: number, truncateFalse: boolean): string;
+/**
+ * Splits `input` into `batchSize` slices and embeds them through `client.embed()`
+ * with at most `concurrency` batches in flight. Order-preserving and fail-fast —
+ * see the module doc for the full semantics.
+ *
+ * Prefer the {@link OllamaClient.embedBatch} wrapper, which additionally
+ * injects `defaultContextLength` as `num_ctx` and enforces the client's
+ * `onContextOverflow` policy against per-string estimates before any request
+ * is sent.
+ */
+declare function batchEmbed(client: OllamaClient, options: EmbedBatchOptions): Promise<EmbedBatchResult>;
+
 /**
  * Ergonomic System One types — key-safe generics on top of the generated
  * contract models.
@@ -1425,6 +1550,14 @@ declare class OllamaClient {
     private readonly contextWarningThreshold;
     /** Configured overflow policy — `'warn'` (default) or `'throw'`. */
     private readonly onContextOverflow;
+    /**
+     * Controllers for every in-flight request and active stream managed by this
+     * client — see {@link OllamaClient.destroy} and {@link createRequestScope}.
+     * Entries live for exactly as long as the underlying work does: until the
+     * response settles (requests) or the stream's `finalResult` settles
+     * (streams, via `executeWithFailover`'s `holdUntil` release path).
+     */
+    private readonly activeRequests;
     constructor(config?: OllamaClientConfig);
     get modelsClient(): ModelsClient;
     /**
@@ -1523,6 +1656,17 @@ declare class OllamaClient {
      * reference when there's nothing to inject (the common case).
      */
     private injectDefaultContextLength;
+    /**
+     * Creates the per-request abort scope registered with {@link destroy}:
+     * a controller that fires when the caller's `signal` aborts OR when
+     * `destroy()` runs, whichever comes first. `dispose()` removes the
+     * registration when the work it guards settles — for plain requests when
+     * the response settles, for streams when `finalResult` settles (the same
+     * moment `executeWithFailover`'s `holdUntil` releases the endpoint slot),
+     * so a destroy() can tear down a stream mid-consumption but never outlives
+     * one that already finished.
+     */
+    private createRequestScope;
     /**
      * Fail-fast guard for `format` (structured output) requests: throws before any network
      * call if the candidate endpoint is inferred as Ollama Cloud, which does not currently
@@ -1585,6 +1729,36 @@ declare class OllamaClient {
     generateWithSchema<T>(req: Omit<GenerateRequestOptions, 'stream' | 'format'>, schema: z.ZodType<T>): Promise<T>;
     embed(req: EmbedRequestOptions): Promise<EmbedResponse>;
     embedText(model: string, input: string | readonly string[]): Promise<readonly (readonly number[])[]>;
+    /**
+     * Batch-constrained, order-preserving embedding generation for large
+     * corpora — the safe building block for RAG/vector-index ingestion at
+     * scale. Splits `input` into `batchSize` slices and embeds them through
+     * {@link embed} with at most `concurrency` batches in flight (defaults:
+     * 32 × 3), so `Promise.all`-style floods that OOM local daemons and
+     * saturate `OLLAMA_MAX_QUEUE` can't happen by accident. Fail-fast: the
+     * first batch error aborts every sibling in-flight batch and rejects the
+     * whole operation with the original error.
+     *
+     * Before any request is sent, `defaultContextLength` is injected as
+     * `num_ctx` (same as chat/generate) and each input string's token estimate
+     * is checked against the resolved window: Ollama silently truncates
+     * oversized embedding inputs by default, so the client warns (or throws,
+     * under `onContextOverflow: 'throw'`) with the offending indexes — see
+     * `src/embed-batch.ts` for the exact trip point (no margin: an embedding
+     * prompt IS the whole input).
+     *
+     * ```ts
+     * const { embeddings, batchCount } = await client.embedBatch({
+     *   model: 'nomic-embed-text:latest',
+     *   input: corpus,                 // e.g. 50k chunks
+     *   batchSize: 32,                 // inputs per /api/embed request
+     *   concurrency: 3,                // batches in flight
+     *   onBatchComplete: (done, total) => progress.log(`${done}/${total}`),
+     * });
+     * // embeddings[i] always corresponds to corpus[i]
+     * ```
+     */
+    embedBatch(options: EmbedBatchOptions): Promise<EmbedBatchResult>;
     /**
      * @deprecated Ollama's `/api/embeddings` endpoint has been superseded by `/api/embed`
      * (exposed here as {@link OllamaClient.embed}), which additionally supports batch
@@ -1710,6 +1884,26 @@ declare class OllamaClient {
     runtimeMode(): RuntimeMode;
     healthCheck(): Promise<EndpointHealthCheckResult[]>;
     endpointStatus(): EndpointHealth[];
+    /**
+     * Immediately aborts every in-flight request and active stream managed by
+     * this client — the clean teardown path for short-lived processes, CLI
+     * runners, and `node:worker_threads` tasks where dangling fetch bodies,
+     * unconsumed stream readers, or queued capacity waiters would otherwise
+     * keep the event loop (and thus the thread/process) alive.
+     *
+     * Every aborted operation rejects with an {@link OllamaAbortError}
+     * (`code: 'aborted'`) carrying `reason` as its message — including requests
+     * still queued behind `maxConcurrentPerEndpoint`, not just dispatched ones.
+     * Streams reject their `finalResult` and release their endpoint slots.
+     *
+     * Idempotent: calling it again (or on a client with nothing in flight)
+     * is a no-op that returns `0`. The client remains usable afterward —
+     * `destroy()` is a drain, not a permanent disable; callers that want a
+     * permanently-dead client should discard the instance.
+     *
+     * @returns the number of in-flight operations that were aborted.
+     */
+    destroy(reason?: string): number;
     get openai(): OpenAICompatClient;
     get anthropic(): AnthropicCompatClient;
 }
@@ -2657,4 +2851,4 @@ declare const OLLAMA_FALLBACK_CONTEXT_LENGTH = 2048;
  */
 declare function contextWarningMessage(check: ContextCheck, kind: 'chat' | 'generate'): string;
 
-export { ATTR_GEN_AI_OPERATION_NAME, ATTR_GEN_AI_REQUEST_MODEL, ATTR_GEN_AI_RESPONSE_MODEL, ATTR_GEN_AI_SYSTEM, ATTR_GEN_AI_TOOL_CALL_ID, ATTR_GEN_AI_TOOL_NAME, ATTR_GEN_AI_USAGE_INPUT_TOKENS, ATTR_GEN_AI_USAGE_OUTPUT_TOKENS, ATTR_HTTP_REQUEST_METHOD, ATTR_HTTP_RESPONSE_STATUS_CODE, ATTR_OLLAMA_AGENT_ITERATION, ATTR_OLLAMA_AGENT_MAX_ITERATIONS, ATTR_OLLAMA_ENDPOINT_ATTEMPT, ATTR_OLLAMA_ENDPOINT_NAME, ATTR_SERVER_ADDRESS, ATTR_SERVER_PORT, ATTR_URL_FULL, AbortableAsyncIterable, Agent, type AgentChatClient, type AgentConfig, type AgentHooks, type AgentResult, type AgentRunInput, type AgentTurn, type AnswerFor, AnthropicCompatClient, AnyTool, type BackoffOptions, BalanceRequestOptions, type BalanceResponse, type BlobUploadResult, ChatRequestOptions, ChatResponse, ChatStreamResult, type ChoiceDecision, type ContextCheck, type ContextCheckOptions, type ContextWindowSource, type ConversationSendOptions, ConversationSession, type ConversationSessionOptions, CopyRequestOptions, CreateRequestOptions, type CumulativeCacheStats, DEFAULT_BACKOFF, DEFAULT_BASE_URL, DEFAULT_CONTEXT_WARNING_THRESHOLD, DEFAULT_FAILOVER_CODES, DEFAULT_FETCH_BACKOFF_CONFIG, DEFAULT_RETRY_CONFIG, DEFAULT_TIMEOUT_MS, type Decision, type DefineToolOptions, DeleteRequestOptions, EmbedRequestOptions, EmbedResponse, EmbeddingsRequestOptions, EmbeddingsResponse, type EndpointHealth, type EndpointHealthCheckResult, EndpointRegistry, type EndpointRegistryOptions, FetchLike, type FetchWithBackoffConfig, FormatOption, GEN_AI_SYSTEM_OLLAMA, GenerateRequestOptions, GenerateResponse, GenerateStreamResult, HttpClient, IMAGE_TOKEN_ESTIMATE, type IncludedBalance, KEEP_ALIVE_INDEFINITE, KEEP_ALIVE_UNLOAD, KNOWN_DONE_REASONS, type KeepAlive, type KnownDoneReason, type LegacyBalanceLimit, type LegacyIncludedBalance, type LoadMcpToolsOptions, Logger, MAX_SYSTEM_ONE_IMAGES_BYTES, MAX_SYSTEM_ONE_REQUEST_BYTES, McpBridge, type McpBridgeOptions, McpCallToolResult, McpClientLike, McpTask, McpToolDescriptor, Message, Middleware, type ModelCapabilities, ModelOptions, ModelResponse, ModelsClient, type NoulDecision, OLLAMA_CLOUD_BASE_URL, OLLAMA_FALLBACK_CONTEXT_LENGTH, OllamaClient, type OllamaClientConfig, type OllamaCloudFreeTierQuotaBudgets, type OllamaCredentialConfig, type OllamaEndpoint, OllamaStream, OllamaStreamEvent, OllamaStreamEventType, OpenAICompatClient, type ParallelToolCallBehavior, ProgressResponse, ProgressStreamResult, PsResponse, PullRequestOptions, PushRequestOptions, QuotaManager, type QuotaManagerOptions, type QuotaWindowConfig, type QuotaWindowStatus, RETRYABLE_STATUS_CODES, type RankResult, RawUsageSource, RequestCancellationOptions, RequestLifecycleHook, RequestRunner, type ResponsesCreateRequest, type ResponsesCreateResponse, ResponsesModule, type ResponsesReasoningEffort, type ResponsesStreamEvent, type ResponsesToolDefinition, type ResponsesUsage, type RetryConfig, type RuntimeMode, type ScoreDecision, type SessionTurn, ShowRequestOptions, ShowResponse, type SpanAttributeValue, type SpanAttributes, StatusResponse, type SupportedSchema, type SystemOneAnswer, type SystemOneAnswers, SystemOneChoiceAnswer, SystemOneChoiceQuestion, SystemOneContent, SystemOneNoulAnswer, SystemOneNoulQuestion, type SystemOneQuestion, type SystemOneQuestions, type SystemOneRequest, type SystemOneResponse, SystemOneScoreAnswer, SystemOneScoreQuestion, type SystemOneUsage, ThinkValue, ThinkingMetadata, type TimeoutSignal, TokenUsage, Tool, ToolCall, ToolCallAccumulator, ToolDefinition, ToolExecutionResult, ToolHandler, ToolParameters, ToolRegistry, type TurnCacheStats, type UsageBucket, type UsageMetrics, UsageRequestOptions, type UsageResponse, VersionResponse, VisionInput, WebFetchRequestOptions, WebFetchResponse, WebSearchRequestOptions, WebSearchResponse, addSpanEvent, calculateBackoff, checkChatContext, checkEndpointHealth, checkGenerateContext, contextWarningMessage, createDecision, createOllamaCloudFreeTierQuota, createTimeoutSignal, defineTool, detectModelCapabilities, encodeImage, estimateChatRequestTokens, estimateGenerateRequestTokens, estimateSystemOneRequestBytes, estimateTokens, fetchWithBackoff, inferRuntimeMode, isKeepAliveSugar, isKnownDoneReason, isSameToolCall, listAllMcpTools, listAvailableModels, loadMcpTools, mergeToolCall, mergeToolCallArgumentsString, mergeToolCallArrays, normalizeChatStream, normalizeGenerateStream, normalizeKeepAlive, normalizeProgressStream, parseNdjsonStream, parseStructuredOutput, registerMcpTools, resolveCredentialEndpoints, setSpanError, toDataStream, toResponse, toTextStream, withRetry, withSpan, zodToJsonSchema };
+export { ATTR_GEN_AI_OPERATION_NAME, ATTR_GEN_AI_REQUEST_MODEL, ATTR_GEN_AI_RESPONSE_MODEL, ATTR_GEN_AI_SYSTEM, ATTR_GEN_AI_TOOL_CALL_ID, ATTR_GEN_AI_TOOL_NAME, ATTR_GEN_AI_USAGE_INPUT_TOKENS, ATTR_GEN_AI_USAGE_OUTPUT_TOKENS, ATTR_HTTP_REQUEST_METHOD, ATTR_HTTP_RESPONSE_STATUS_CODE, ATTR_OLLAMA_AGENT_ITERATION, ATTR_OLLAMA_AGENT_MAX_ITERATIONS, ATTR_OLLAMA_ENDPOINT_ATTEMPT, ATTR_OLLAMA_ENDPOINT_NAME, ATTR_SERVER_ADDRESS, ATTR_SERVER_PORT, ATTR_URL_FULL, AbortableAsyncIterable, Agent, type AgentChatClient, type AgentConfig, type AgentHooks, type AgentResult, type AgentRunInput, type AgentTurn, type AnswerFor, AnthropicCompatClient, AnyTool, type BackoffOptions, BalanceRequestOptions, type BalanceResponse, type BlobUploadResult, ChatRequestOptions, ChatResponse, ChatStreamResult, type ChoiceDecision, type ContextCheck, type ContextCheckOptions, type ContextWindowSource, type ConversationSendOptions, ConversationSession, type ConversationSessionOptions, CopyRequestOptions, CreateRequestOptions, type CumulativeCacheStats, DEFAULT_BACKOFF, DEFAULT_BASE_URL, DEFAULT_CONTEXT_WARNING_THRESHOLD, DEFAULT_EMBED_BATCH_SIZE, DEFAULT_EMBED_CONCURRENCY, DEFAULT_FAILOVER_CODES, DEFAULT_FETCH_BACKOFF_CONFIG, DEFAULT_RETRY_CONFIG, DEFAULT_TIMEOUT_MS, type Decision, type DefineToolOptions, DeleteRequestOptions, type EmbedBatchOptions, type EmbedBatchResult, EmbedRequestOptions, EmbedResponse, EmbeddingsRequestOptions, EmbeddingsResponse, type EndpointHealth, type EndpointHealthCheckResult, EndpointRegistry, type EndpointRegistryOptions, FetchLike, type FetchWithBackoffConfig, FormatOption, GEN_AI_SYSTEM_OLLAMA, GenerateRequestOptions, GenerateResponse, GenerateStreamResult, HttpClient, IMAGE_TOKEN_ESTIMATE, type IncludedBalance, KEEP_ALIVE_INDEFINITE, KEEP_ALIVE_UNLOAD, KNOWN_DONE_REASONS, type KeepAlive, type KnownDoneReason, type LegacyBalanceLimit, type LegacyIncludedBalance, type LoadMcpToolsOptions, Logger, MAX_SYSTEM_ONE_IMAGES_BYTES, MAX_SYSTEM_ONE_REQUEST_BYTES, McpBridge, type McpBridgeOptions, McpCallToolResult, McpClientLike, McpTask, McpToolDescriptor, Message, Middleware, type ModelCapabilities, ModelOptions, ModelResponse, ModelsClient, type NoulDecision, OLLAMA_CLOUD_BASE_URL, OLLAMA_FALLBACK_CONTEXT_LENGTH, OllamaClient, type OllamaClientConfig, type OllamaCloudFreeTierQuotaBudgets, type OllamaCredentialConfig, type OllamaEndpoint, OllamaStream, OllamaStreamEvent, OllamaStreamEventType, OpenAICompatClient, type OversizedEmbedInput, type ParallelToolCallBehavior, ProgressResponse, ProgressStreamResult, PsResponse, PullRequestOptions, PushRequestOptions, QuotaManager, type QuotaManagerOptions, type QuotaWindowConfig, type QuotaWindowStatus, RETRYABLE_STATUS_CODES, type RankResult, RawUsageSource, RequestCancellationOptions, RequestLifecycleHook, RequestRunner, type ResponsesCreateRequest, type ResponsesCreateResponse, ResponsesModule, type ResponsesReasoningEffort, type ResponsesStreamEvent, type ResponsesToolDefinition, type ResponsesUsage, type RetryConfig, type RuntimeMode, type ScoreDecision, type SessionTurn, ShowRequestOptions, ShowResponse, type SpanAttributeValue, type SpanAttributes, StatusResponse, type SupportedSchema, type SystemOneAnswer, type SystemOneAnswers, SystemOneChoiceAnswer, SystemOneChoiceQuestion, SystemOneContent, SystemOneNoulAnswer, SystemOneNoulQuestion, type SystemOneQuestion, type SystemOneQuestions, type SystemOneRequest, type SystemOneResponse, SystemOneScoreAnswer, SystemOneScoreQuestion, type SystemOneUsage, ThinkValue, ThinkingMetadata, type TimeoutSignal, TokenUsage, Tool, ToolCall, ToolCallAccumulator, ToolDefinition, ToolExecutionResult, ToolHandler, ToolParameters, ToolRegistry, type TurnCacheStats, type UsageBucket, type UsageMetrics, UsageRequestOptions, type UsageResponse, VersionResponse, VisionInput, WebFetchRequestOptions, WebFetchResponse, WebSearchRequestOptions, WebSearchResponse, addSpanEvent, batchEmbed, calculateBackoff, checkChatContext, checkEndpointHealth, checkGenerateContext, contextWarningMessage, createDecision, createOllamaCloudFreeTierQuota, createTimeoutSignal, defineTool, detectModelCapabilities, embedBatchOverflowMessage, encodeImage, estimateChatRequestTokens, estimateGenerateRequestTokens, estimateSystemOneRequestBytes, estimateTokens, fetchWithBackoff, findOversizedEmbedInputs, inferRuntimeMode, isKeepAliveSugar, isKnownDoneReason, isSameToolCall, listAllMcpTools, listAvailableModels, loadMcpTools, mergeToolCall, mergeToolCallArgumentsString, mergeToolCallArrays, normalizeChatStream, normalizeGenerateStream, normalizeKeepAlive, normalizeProgressStream, parseNdjsonStream, parseStructuredOutput, registerMcpTools, resolveCredentialEndpoints, setSpanError, toDataStream, toResponse, toTextStream, withRetry, withSpan, zodToJsonSchema };