@nemesis-oss/ollama-sdk 1.18.0 → 1.19.0

package/dist/index.d.cts

@@ -1,12 +1,12 @@
-import { T as ThinkingMetadata, M as ModelResponse, R as RequestCancellationOptions, 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, 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 PingResult, w as VisionInput } from './types-DbPnfVh-.cjs';
-export { L as ListResponse, x as Logprob, y as LogprobToken, z as ModelDetails, A as Role, H as ToolCallFunction, I as ToolFunctionDefinition, J as ToolProperty, K as UsageRange, N as UsageScope, O as WebSearchResult, Q as imageStringNeedsResolution, X as resolveImageInput } from './types-DbPnfVh-.cjs';
-import { H as HttpClient, F as FetchLike, M as Middleware, L as Logger, R as RequestLifecycleHook, 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-CdZP-N7j.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 NETWORK_ERROR_CODES, r as NOOP_LOGGER, s as NextFunction, t as OllamaAbortError, u as OllamaAgentMaxIterationsError, v as OllamaAgentMaxToolCallsError, w as OllamaAgentToolLoopError, x as OllamaAuthError, y as OllamaBadGatewayError, z as OllamaClientError, J as OllamaClientErrorOptions, K as OllamaErrorRequestContext, Q as OllamaErrorResponseContext, U as OllamaGenericClientError, V as OllamaIncompatibleModelError, W as OllamaMcpError, X as OllamaModelRoutingError, Y as OllamaNetworkError, Z as OllamaNotFoundError, _ as OllamaOverloadedError, $ as OllamaQuotaExceededError, a0 as OllamaRateLimitError, a1 as OllamaRequestTooLargeError, a2 as OllamaRequestValidationError, a3 as OllamaResponseValidationError, a4 as OllamaServerError, a5 as OllamaServerVersionUnknownError, a6 as OllamaSkillInvalidError, a7 as OllamaSkillNotFoundError, a8 as OllamaStreamError, a9 as OllamaTimeoutError, aa as OllamaToolTimeoutError, ab as OllamaToolValidationError, ac as OllamaUnsupportedCapabilityError, ad as RequestContext, ae as RequestLifecycleEvent, af as ResponseContext, ag as SseEvent, ah as ThinkingEvent, ai as TokenEvent, aj as ToolCallEvent, ak as composeMiddleware, al as createConsoleLogger, am as extractUsage, an as isNetworkLevelError, ao as mapError, ap as parseSseStream } from './runtime-CdZP-N7j.cjs';
+import { T as ThinkingMetadata, M as ModelResponse, R as RequestCancellationOptions, 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, 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 PingResult, w as VisionInput } from './types-Y3u0FtdN.cjs';
+export { L as ListResponse, x as Logprob, y as LogprobToken, z as ModelDetails, A as Role, H as ToolCallFunction, I as ToolFunctionDefinition, J as ToolProperty, K as UsageRange, N as UsageScope, O as WebSearchResult, Q as imageStringNeedsResolution, X as resolveImageInput } from './types-Y3u0FtdN.cjs';
+import { H as HttpClient, F as FetchLike, M as Middleware, L as Logger, R as RequestLifecycleHook, 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-C28YEAHC.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 NETWORK_ERROR_CODES, r as NOOP_LOGGER, s as NextFunction, t as OllamaAbortError, u as OllamaAgentMaxIterationsError, v as OllamaAgentMaxToolCallsError, w as OllamaAgentToolLoopError, x as OllamaAuthError, y as OllamaBadGatewayError, z as OllamaClientError, J as OllamaClientErrorOptions, K as OllamaErrorRequestContext, Q as OllamaErrorResponseContext, U as OllamaGenericClientError, V as OllamaIncompatibleModelError, W as OllamaMcpError, X as OllamaModelRoutingError, Y as OllamaNetworkError, Z as OllamaNotFoundError, _ as OllamaOverloadedError, $ as OllamaPaymentRequiredError, a0 as OllamaQuotaExceededError, a1 as OllamaRateLimitError, a2 as OllamaRequestTooLargeError, a3 as OllamaRequestValidationError, a4 as OllamaResponseValidationError, a5 as OllamaServerError, a6 as OllamaServerVersionUnknownError, a7 as OllamaSkillInvalidError, a8 as OllamaSkillNotFoundError, a9 as OllamaStreamError, aa as OllamaTimeoutError, ab as OllamaToolTimeoutError, ac as OllamaToolValidationError, ad as OllamaUnsupportedCapabilityError, ae as RequestContext, af as RequestLifecycleEvent, ag as ResponseContext, ah as SseEvent, ai as ThinkingEvent, aj as TokenEvent, ak as ToolCallEvent, al as composeMiddleware, am as createConsoleLogger, an as extractUsage, ao as isNetworkLevelError, ap as mapError, aq as parseSseStream } from './runtime-C28YEAHC.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-BvHOsidk.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 OllamaOpenAIChatContentPart, a4 as OllamaOpenAIChatMessage, a5 as OllamaOpenAICompletionRequest, a6 as OllamaOpenAIEmbeddingRequest, a7 as OllamaOpenAIImageDataUrl, a8 as OllamaOpenAIReasoningEffort, a9 as OllamaOpenAIResponsesRequest, aa as OpenAIChatCompletionChoice, b as OpenAIChatCompletionChunk, ab as OpenAIChatCompletionChunkChoice, ac as OpenAIChatCompletionDelta, O as OpenAIChatCompletionRequest, a as OpenAIChatCompletionResponse, ad as OpenAIChatCompletionStream, ae as OpenAICompletionChoice, e as OpenAICompletionChunk, c as OpenAICompletionRequest, d as OpenAICompletionResponse, af as OpenAICompletionStream, ag as OpenAIContentPart, ah as OpenAIEmbeddingInput, ai as OpenAIEmbeddingItem, f as OpenAIEmbeddingRequest, g as OpenAIEmbeddingResponse, aj as OpenAIFunctionDefinition, ak as OpenAIImageUrlContentPart, h as OpenAIListModelsResponse, al as OpenAIMessage, i as OpenAIModelItem, am as OpenAIReasoningEffort, an as OpenAIResponseFormat, ao as OpenAIResponsesCompletedEvent, ap as OpenAIResponsesContentPartAddedEvent, aq as OpenAIResponsesContentPartDoneEvent, ar as OpenAIResponsesCreatedEvent, as as OpenAIResponsesError, at as OpenAIResponsesEventBase, au as OpenAIResponsesFailedEvent, av as OpenAIResponsesFunctionCallArgumentsDeltaEvent, aw as OpenAIResponsesFunctionCallArgumentsDoneEvent, ax as OpenAIResponsesInProgressEvent, ay as OpenAIResponsesIncompleteDetails, az as OpenAIResponsesIncompleteEvent, aA as OpenAIResponsesOutputContent, aB as OpenAIResponsesOutputFunctionCall, aC as OpenAIResponsesOutputItem, aD as OpenAIResponsesOutputItemAddedEvent, aE as OpenAIResponsesOutputItemDoneEvent, aF as OpenAIResponsesOutputMessage, aG as OpenAIResponsesOutputReasoning, aH as OpenAIResponsesOutputRefusal, aI as OpenAIResponsesOutputTextContent, aJ as OpenAIResponsesOutputTextDeltaEvent, aK as OpenAIResponsesOutputTextDoneEvent, aL as OpenAIResponsesQueuedEvent, aM as OpenAIResponsesReasoningSummaryPartAddedEvent, aN as OpenAIResponsesReasoningSummaryPartDoneEvent, aO as OpenAIResponsesReasoningSummaryTextDeltaEvent, aP as OpenAIResponsesReasoningSummaryTextDoneEvent, aQ as OpenAIResponsesReasoningTextDeltaEvent, aR as OpenAIResponsesReasoningTextDoneEvent, aS as OpenAIResponsesRefusalDeltaEvent, aT as OpenAIResponsesRefusalDoneEvent, j as OpenAIResponsesRequest, k as OpenAIResponsesResponse, aU as OpenAIResponsesStatus, aV as OpenAIResponsesStream, aW as OpenAIResponsesStreamError, l as OpenAIResponsesStreamEvent, aX as OpenAIResponsesUsage, aY as OpenAIStreamOptions, aZ as OpenAITextContentPart, a_ as OpenAITool, a$ as OpenAIToolCall, b0 as OpenAIToolCallDelta, b1 as OpenAIUsage, b2 as SystemOneConfidence, b3 as SystemOneProbabilities } from './anthropic-BvHOsidk.cjs';
-import { T as ToolHandler, a as Tool, A as AnyTool, b as ToolRegistry, c as ToolExecutionResult } from './compose-DFkmtUOP.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-DFkmtUOP.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-DIPnBwlf.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 OllamaOpenAIChatContentPart, a4 as OllamaOpenAIChatMessage, a5 as OllamaOpenAICompletionRequest, a6 as OllamaOpenAIEmbeddingRequest, a7 as OllamaOpenAIImageDataUrl, a8 as OllamaOpenAIReasoningEffort, a9 as OllamaOpenAIResponsesRequest, aa as OpenAIChatCompletionChoice, b as OpenAIChatCompletionChunk, ab as OpenAIChatCompletionChunkChoice, ac as OpenAIChatCompletionDelta, O as OpenAIChatCompletionRequest, a as OpenAIChatCompletionResponse, ad as OpenAIChatCompletionStream, ae as OpenAICompletionChoice, e as OpenAICompletionChunk, c as OpenAICompletionRequest, d as OpenAICompletionResponse, af as OpenAICompletionStream, ag as OpenAIContentPart, ah as OpenAIEmbeddingInput, ai as OpenAIEmbeddingItem, f as OpenAIEmbeddingRequest, g as OpenAIEmbeddingResponse, aj as OpenAIFunctionDefinition, ak as OpenAIImageUrlContentPart, h as OpenAIListModelsResponse, al as OpenAIMessage, i as OpenAIModelItem, am as OpenAIReasoningEffort, an as OpenAIResponseFormat, ao as OpenAIResponsesCompletedEvent, ap as OpenAIResponsesContentPartAddedEvent, aq as OpenAIResponsesContentPartDoneEvent, ar as OpenAIResponsesCreatedEvent, as as OpenAIResponsesError, at as OpenAIResponsesEventBase, au as OpenAIResponsesFailedEvent, av as OpenAIResponsesFunctionCallArgumentsDeltaEvent, aw as OpenAIResponsesFunctionCallArgumentsDoneEvent, ax as OpenAIResponsesInProgressEvent, ay as OpenAIResponsesIncompleteDetails, az as OpenAIResponsesIncompleteEvent, aA as OpenAIResponsesOutputContent, aB as OpenAIResponsesOutputFunctionCall, aC as OpenAIResponsesOutputItem, aD as OpenAIResponsesOutputItemAddedEvent, aE as OpenAIResponsesOutputItemDoneEvent, aF as OpenAIResponsesOutputMessage, aG as OpenAIResponsesOutputReasoning, aH as OpenAIResponsesOutputRefusal, aI as OpenAIResponsesOutputTextContent, aJ as OpenAIResponsesOutputTextDeltaEvent, aK as OpenAIResponsesOutputTextDoneEvent, aL as OpenAIResponsesQueuedEvent, aM as OpenAIResponsesReasoningSummaryPartAddedEvent, aN as OpenAIResponsesReasoningSummaryPartDoneEvent, aO as OpenAIResponsesReasoningSummaryTextDeltaEvent, aP as OpenAIResponsesReasoningSummaryTextDoneEvent, aQ as OpenAIResponsesReasoningTextDeltaEvent, aR as OpenAIResponsesReasoningTextDoneEvent, aS as OpenAIResponsesRefusalDeltaEvent, aT as OpenAIResponsesRefusalDoneEvent, j as OpenAIResponsesRequest, k as OpenAIResponsesResponse, aU as OpenAIResponsesStatus, aV as OpenAIResponsesStream, aW as OpenAIResponsesStreamError, l as OpenAIResponsesStreamEvent, aX as OpenAIResponsesUsage, aY as OpenAIStreamOptions, aZ as OpenAITextContentPart, a_ as OpenAITool, a$ as OpenAIToolCall, b0 as OpenAIToolCallDelta, b1 as OpenAIUsage, b2 as SystemOneConfidence, b3 as SystemOneProbabilities } from './anthropic-DIPnBwlf.cjs';
+import { T as ToolHandler, a as Tool, A as AnyTool, b as ToolRegistry, c as ToolExecutionResult } from './compose-C4QxEZOS.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-C4QxEZOS.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';
@@ -1655,25 +1655,34 @@ declare function batchEmbed(client: OllamaClient, options: EmbedBatchOptions): P
  */
 
 /**
- * Maximum total request size, in bytes, accepted by `/v1/systemone`
- * when the request body carries a non-empty `images` array.
+ * Whole-request size cap, in bytes, for `/v1/systemone` — the maximum
+ * serialized size of the complete request, base64 image bytes included.
  *
- * The Ollama server enforces this limit server-side and rejects
+ * Part of the dual cap clarified upstream on 2026-10-16: every request
+ * must satisfy BOTH bounds simultaneously — this whole-request cap AND
+ * the text-and-schema cap of {@link MAX_SYSTEM_ONE_REQUEST_BYTES}
+ * (64 KiB, images excluded). The bumped whole-request cap exists
+ * specifically to accommodate base64-encoded image bytes shared across
+ * all questions in a single System One call; the text and schema
+ * portion may never exceed 64 KiB regardless of whether images are
+ * attached.
+ *
+ * The Ollama server enforces both bounds server-side and rejects
  * oversized payloads with HTTP 413. The SDK's runtime also enforces
- * it client-side (see `OllamaRuntime`'s `maxRequestBytesWithImages`
+ * them client-side (see `OllamaRuntime`'s `maxRequestBytesWithImages`
  * handling) so the round-trip is avoided.
- *
- * Without images, the limit is the much smaller
- * {@link MAX_SYSTEM_ONE_REQUEST_BYTES} (64 KiB) — the bumped
- * images-cap exists specifically to accommodate base64-encoded image
- * bytes shared across all questions in a single System One call.
  */
 declare const MAX_SYSTEM_ONE_IMAGES_BYTES: number;
 /**
- * Maximum total request size, in bytes, for `/v1/systemone` when no
- * images are attached. The Ollama server caps the JSON-only payload
- * at 64 KiB; the runtime enforces this client-side (see
- * `OllamaRuntime`'s `maxRequestBytes`).
+ * Text-and-schema size cap, in bytes, for `/v1/systemone` — the maximum
+ * serialized size of the request with its `images` array excluded.
+ *
+ * Part of the dual cap clarified upstream on 2026-10-16: this bound
+ * applies to EVERY request, images or not — attaching images raises
+ * the whole-request ceiling to {@link MAX_SYSTEM_ONE_IMAGES_BYTES}
+ * (32 MiB) but never relaxes this 64 KiB text-and-schema ceiling. The
+ * runtime enforces both bounds client-side (see `OllamaRuntime`'s
+ * `maxRequestBytes` handling).
  */
 declare const MAX_SYSTEM_ONE_REQUEST_BYTES: number;
 /**
@@ -1740,16 +1749,33 @@ type SystemOneAnswers<Q extends SystemOneQuestions = SystemOneQuestions> = {
  * ## Image support
  *
  * The `images` field accepts up to N base64-encoded images shared by
- * all questions in the request. Total request payload (JSON + images)
- * must remain within {@link MAX_SYSTEM_ONE_IMAGES_BYTES} (32 MiB);
- * requests without images are capped at
- * {@link MAX_SYSTEM_ONE_REQUEST_BYTES} (64 KiB).
+ * all questions in the request. The dual cap (2026-10-16 upstream
+ * clarification) applies to every request simultaneously: the text and
+ * schema portion (images excluded) must stay within
+ * {@link MAX_SYSTEM_ONE_REQUEST_BYTES} (64 KiB), and the whole payload
+ * (JSON + images) within {@link MAX_SYSTEM_ONE_IMAGES_BYTES} (32 MiB).
  *
  * Use {@link estimateSystemOneRequestBytes} for a client-side pre-flight
  * check before sending a large multi-image batch — the runtime also
- * enforces the limit, but the early check avoids the round-trip when
+ * enforces both caps, but the early check avoids the round-trip when
  * the caller already knows the payload is too large.
  *
+ * ## Context window
+ *
+ * Beyond the byte caps, upstream requires the complete input (state,
+ * questions, and images) to fit the model's loaded context window, and
+ * System One never truncates input — oversized prompts are rejected by
+ * the server with HTTP 400 locally ("a rendered prompt that exceeds the
+ * loaded context"), not 413 (which is reserved for the byte caps
+ * above); for cloud requests, upstream documents 413 for input that
+ * exceeds the model's context window (65,536 tokens) as well as for
+ * requests over 13 MiB.
+ * Decision models (Nimble, Tev) additionally reserve two token
+ * positions at the end of the window for scoring. The SDK deliberately
+ * does not estimate this client-side (upstream documents no
+ * tokenization formula); callers who need the real limit can use
+ * `client.models.getContextLength()`.
+ *
  * Requires Clef or Clef Flash with vision weights. URLs and data URLs
  * are NOT supported — pre-encode the bytes to base64 before adding to
  * the array.
@@ -1768,8 +1794,12 @@ interface SystemOneResponse<Q extends SystemOneQuestions = SystemOneQuestions> e
 }
 /**
  * Estimate the on-the-wire byte size of a System One request body, for
- * pre-flight size validation against {@link MAX_SYSTEM_ONE_REQUEST_BYTES}
- * (no images) or {@link MAX_SYSTEM_ONE_IMAGES_BYTES} (with images).
+ * pre-flight validation against the dual cap (2026-10-16 upstream
+ * clarification): the whole body against
+ * {@link MAX_SYSTEM_ONE_IMAGES_BYTES} (32 MiB) and, when `images` is
+ * attached, the images-excluded portion against
+ * {@link MAX_SYSTEM_ONE_REQUEST_BYTES} (64 KiB) — pass
+ * `{ excludeImages: true }` to measure the text-and-schema portion.
  *
  * The estimate is conservative: it uses `JSON.stringify` length and
  * assumes UTF-8 encoding. The actual wire size may differ slightly due
@@ -1781,12 +1811,18 @@ interface SystemOneResponse<Q extends SystemOneQuestions = SystemOneQuestions> e
  *   ```ts
  *   const request: SystemOneRequest = { ... };
  *   const bytes = estimateSystemOneRequestBytes(request);
- *   const cap = request.images?.length ? MAX_SYSTEM_ONE_IMAGES_BYTES : MAX_SYSTEM_ONE_REQUEST_BYTES;
- *   if (bytes > cap) throw new Error(`payload ${bytes}B exceeds cap ${cap}B`);
+ *   const textBytes = request.images?.length
+ *     ? estimateSystemOneRequestBytes(request, { excludeImages: true })
+ *     : bytes;
+ *   if (textBytes > MAX_SYSTEM_ONE_REQUEST_BYTES || bytes > MAX_SYSTEM_ONE_IMAGES_BYTES) {
+ *     throw new Error(`payload ${bytes}B (text ${textBytes}B) exceeds the dual cap`);
+ *   }
  *   await client.systemOne(request);
  *   ```
  */
-declare function estimateSystemOneRequestBytes(request: SystemOneRequest | SystemOneRequest$1): number;
+declare function estimateSystemOneRequestBytes(request: SystemOneRequest | SystemOneRequest$1, options?: {
+    readonly excludeImages?: boolean;
+}): number;
 
 /**
  * Higher-level System One decision helpers.
@@ -2363,11 +2399,15 @@ declare class OllamaClient {
      * 1–64 typed questions (choice, noul, or score) against a provided state
      * and returns typed answers with probabilities and confidence.
      *
-     * Local-only; requires Ollama >= 0.35.0. The runtime enforces the
-     * version constraint automatically (lazily fetching `/api/version` when
-     * `enforceVersion` is `'auto'` or `'strict'`) and rejects cloud-mode
-     * calls. Request size is checked client-side: 64 KiB without images,
-     * 32 MiB with images.
+     * Local and cloud (since the 2026-10-16 upstream System One cloud
+     * launch): `model` also accepts cloud models such as `clef:cloud` —
+     * cloud models require Ollama >= 0.40.5 or the hosted
+     * `https://ollama.com` endpoint. Requires Ollama >= 0.35.0 locally.
+     * The runtime enforces the version constraint automatically (lazily
+     * fetching `/api/version` when `enforceVersion` is `'auto'` or
+     * `'strict'`). Request size is checked client-side as a dual cap:
+     * text and schema (images excluded) within 64 KiB AND the whole
+     * request within 32 MiB.
      *
      * The generic `Q` parameter captures the caller's question map at
      * compile time, giving key-safe answer access:

package/dist/index.d.ts

@@ -1,12 +1,12 @@
-import { T as ThinkingMetadata, M as ModelResponse, R as RequestCancellationOptions, 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, 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 PingResult, w as VisionInput } from './types-DbPnfVh-.js';
-export { L as ListResponse, x as Logprob, y as LogprobToken, z as ModelDetails, A as Role, H as ToolCallFunction, I as ToolFunctionDefinition, J as ToolProperty, K as UsageRange, N as UsageScope, O as WebSearchResult, Q as imageStringNeedsResolution, X as resolveImageInput } from './types-DbPnfVh-.js';
-import { H as HttpClient, F as FetchLike, M as Middleware, L as Logger, R as RequestLifecycleHook, 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-CFuv7Ifh.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 NETWORK_ERROR_CODES, r as NOOP_LOGGER, s as NextFunction, t as OllamaAbortError, u as OllamaAgentMaxIterationsError, v as OllamaAgentMaxToolCallsError, w as OllamaAgentToolLoopError, x as OllamaAuthError, y as OllamaBadGatewayError, z as OllamaClientError, J as OllamaClientErrorOptions, K as OllamaErrorRequestContext, Q as OllamaErrorResponseContext, U as OllamaGenericClientError, V as OllamaIncompatibleModelError, W as OllamaMcpError, X as OllamaModelRoutingError, Y as OllamaNetworkError, Z as OllamaNotFoundError, _ as OllamaOverloadedError, $ as OllamaQuotaExceededError, a0 as OllamaRateLimitError, a1 as OllamaRequestTooLargeError, a2 as OllamaRequestValidationError, a3 as OllamaResponseValidationError, a4 as OllamaServerError, a5 as OllamaServerVersionUnknownError, a6 as OllamaSkillInvalidError, a7 as OllamaSkillNotFoundError, a8 as OllamaStreamError, a9 as OllamaTimeoutError, aa as OllamaToolTimeoutError, ab as OllamaToolValidationError, ac as OllamaUnsupportedCapabilityError, ad as RequestContext, ae as RequestLifecycleEvent, af as ResponseContext, ag as SseEvent, ah as ThinkingEvent, ai as TokenEvent, aj as ToolCallEvent, ak as composeMiddleware, al as createConsoleLogger, am as extractUsage, an as isNetworkLevelError, ao as mapError, ap as parseSseStream } from './runtime-CFuv7Ifh.js';
+import { T as ThinkingMetadata, M as ModelResponse, R as RequestCancellationOptions, 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, 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 PingResult, w as VisionInput } from './types-Y3u0FtdN.js';
+export { L as ListResponse, x as Logprob, y as LogprobToken, z as ModelDetails, A as Role, H as ToolCallFunction, I as ToolFunctionDefinition, J as ToolProperty, K as UsageRange, N as UsageScope, O as WebSearchResult, Q as imageStringNeedsResolution, X as resolveImageInput } from './types-Y3u0FtdN.js';
+import { H as HttpClient, F as FetchLike, M as Middleware, L as Logger, R as RequestLifecycleHook, 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-DYCxYGG-.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 NETWORK_ERROR_CODES, r as NOOP_LOGGER, s as NextFunction, t as OllamaAbortError, u as OllamaAgentMaxIterationsError, v as OllamaAgentMaxToolCallsError, w as OllamaAgentToolLoopError, x as OllamaAuthError, y as OllamaBadGatewayError, z as OllamaClientError, J as OllamaClientErrorOptions, K as OllamaErrorRequestContext, Q as OllamaErrorResponseContext, U as OllamaGenericClientError, V as OllamaIncompatibleModelError, W as OllamaMcpError, X as OllamaModelRoutingError, Y as OllamaNetworkError, Z as OllamaNotFoundError, _ as OllamaOverloadedError, $ as OllamaPaymentRequiredError, a0 as OllamaQuotaExceededError, a1 as OllamaRateLimitError, a2 as OllamaRequestTooLargeError, a3 as OllamaRequestValidationError, a4 as OllamaResponseValidationError, a5 as OllamaServerError, a6 as OllamaServerVersionUnknownError, a7 as OllamaSkillInvalidError, a8 as OllamaSkillNotFoundError, a9 as OllamaStreamError, aa as OllamaTimeoutError, ab as OllamaToolTimeoutError, ac as OllamaToolValidationError, ad as OllamaUnsupportedCapabilityError, ae as RequestContext, af as RequestLifecycleEvent, ag as ResponseContext, ah as SseEvent, ai as ThinkingEvent, aj as TokenEvent, ak as ToolCallEvent, al as composeMiddleware, am as createConsoleLogger, an as extractUsage, ao as isNetworkLevelError, ap as mapError, aq as parseSseStream } from './runtime-DYCxYGG-.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-CvGVSoVP.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 OllamaOpenAIChatContentPart, a4 as OllamaOpenAIChatMessage, a5 as OllamaOpenAICompletionRequest, a6 as OllamaOpenAIEmbeddingRequest, a7 as OllamaOpenAIImageDataUrl, a8 as OllamaOpenAIReasoningEffort, a9 as OllamaOpenAIResponsesRequest, aa as OpenAIChatCompletionChoice, b as OpenAIChatCompletionChunk, ab as OpenAIChatCompletionChunkChoice, ac as OpenAIChatCompletionDelta, O as OpenAIChatCompletionRequest, a as OpenAIChatCompletionResponse, ad as OpenAIChatCompletionStream, ae as OpenAICompletionChoice, e as OpenAICompletionChunk, c as OpenAICompletionRequest, d as OpenAICompletionResponse, af as OpenAICompletionStream, ag as OpenAIContentPart, ah as OpenAIEmbeddingInput, ai as OpenAIEmbeddingItem, f as OpenAIEmbeddingRequest, g as OpenAIEmbeddingResponse, aj as OpenAIFunctionDefinition, ak as OpenAIImageUrlContentPart, h as OpenAIListModelsResponse, al as OpenAIMessage, i as OpenAIModelItem, am as OpenAIReasoningEffort, an as OpenAIResponseFormat, ao as OpenAIResponsesCompletedEvent, ap as OpenAIResponsesContentPartAddedEvent, aq as OpenAIResponsesContentPartDoneEvent, ar as OpenAIResponsesCreatedEvent, as as OpenAIResponsesError, at as OpenAIResponsesEventBase, au as OpenAIResponsesFailedEvent, av as OpenAIResponsesFunctionCallArgumentsDeltaEvent, aw as OpenAIResponsesFunctionCallArgumentsDoneEvent, ax as OpenAIResponsesInProgressEvent, ay as OpenAIResponsesIncompleteDetails, az as OpenAIResponsesIncompleteEvent, aA as OpenAIResponsesOutputContent, aB as OpenAIResponsesOutputFunctionCall, aC as OpenAIResponsesOutputItem, aD as OpenAIResponsesOutputItemAddedEvent, aE as OpenAIResponsesOutputItemDoneEvent, aF as OpenAIResponsesOutputMessage, aG as OpenAIResponsesOutputReasoning, aH as OpenAIResponsesOutputRefusal, aI as OpenAIResponsesOutputTextContent, aJ as OpenAIResponsesOutputTextDeltaEvent, aK as OpenAIResponsesOutputTextDoneEvent, aL as OpenAIResponsesQueuedEvent, aM as OpenAIResponsesReasoningSummaryPartAddedEvent, aN as OpenAIResponsesReasoningSummaryPartDoneEvent, aO as OpenAIResponsesReasoningSummaryTextDeltaEvent, aP as OpenAIResponsesReasoningSummaryTextDoneEvent, aQ as OpenAIResponsesReasoningTextDeltaEvent, aR as OpenAIResponsesReasoningTextDoneEvent, aS as OpenAIResponsesRefusalDeltaEvent, aT as OpenAIResponsesRefusalDoneEvent, j as OpenAIResponsesRequest, k as OpenAIResponsesResponse, aU as OpenAIResponsesStatus, aV as OpenAIResponsesStream, aW as OpenAIResponsesStreamError, l as OpenAIResponsesStreamEvent, aX as OpenAIResponsesUsage, aY as OpenAIStreamOptions, aZ as OpenAITextContentPart, a_ as OpenAITool, a$ as OpenAIToolCall, b0 as OpenAIToolCallDelta, b1 as OpenAIUsage, b2 as SystemOneConfidence, b3 as SystemOneProbabilities } from './anthropic-CvGVSoVP.js';
-import { T as ToolHandler, a as Tool, A as AnyTool, b as ToolRegistry, c as ToolExecutionResult } from './compose-CeFbHJeO.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-CeFbHJeO.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-CAvwqIjx.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 OllamaOpenAIChatContentPart, a4 as OllamaOpenAIChatMessage, a5 as OllamaOpenAICompletionRequest, a6 as OllamaOpenAIEmbeddingRequest, a7 as OllamaOpenAIImageDataUrl, a8 as OllamaOpenAIReasoningEffort, a9 as OllamaOpenAIResponsesRequest, aa as OpenAIChatCompletionChoice, b as OpenAIChatCompletionChunk, ab as OpenAIChatCompletionChunkChoice, ac as OpenAIChatCompletionDelta, O as OpenAIChatCompletionRequest, a as OpenAIChatCompletionResponse, ad as OpenAIChatCompletionStream, ae as OpenAICompletionChoice, e as OpenAICompletionChunk, c as OpenAICompletionRequest, d as OpenAICompletionResponse, af as OpenAICompletionStream, ag as OpenAIContentPart, ah as OpenAIEmbeddingInput, ai as OpenAIEmbeddingItem, f as OpenAIEmbeddingRequest, g as OpenAIEmbeddingResponse, aj as OpenAIFunctionDefinition, ak as OpenAIImageUrlContentPart, h as OpenAIListModelsResponse, al as OpenAIMessage, i as OpenAIModelItem, am as OpenAIReasoningEffort, an as OpenAIResponseFormat, ao as OpenAIResponsesCompletedEvent, ap as OpenAIResponsesContentPartAddedEvent, aq as OpenAIResponsesContentPartDoneEvent, ar as OpenAIResponsesCreatedEvent, as as OpenAIResponsesError, at as OpenAIResponsesEventBase, au as OpenAIResponsesFailedEvent, av as OpenAIResponsesFunctionCallArgumentsDeltaEvent, aw as OpenAIResponsesFunctionCallArgumentsDoneEvent, ax as OpenAIResponsesInProgressEvent, ay as OpenAIResponsesIncompleteDetails, az as OpenAIResponsesIncompleteEvent, aA as OpenAIResponsesOutputContent, aB as OpenAIResponsesOutputFunctionCall, aC as OpenAIResponsesOutputItem, aD as OpenAIResponsesOutputItemAddedEvent, aE as OpenAIResponsesOutputItemDoneEvent, aF as OpenAIResponsesOutputMessage, aG as OpenAIResponsesOutputReasoning, aH as OpenAIResponsesOutputRefusal, aI as OpenAIResponsesOutputTextContent, aJ as OpenAIResponsesOutputTextDeltaEvent, aK as OpenAIResponsesOutputTextDoneEvent, aL as OpenAIResponsesQueuedEvent, aM as OpenAIResponsesReasoningSummaryPartAddedEvent, aN as OpenAIResponsesReasoningSummaryPartDoneEvent, aO as OpenAIResponsesReasoningSummaryTextDeltaEvent, aP as OpenAIResponsesReasoningSummaryTextDoneEvent, aQ as OpenAIResponsesReasoningTextDeltaEvent, aR as OpenAIResponsesReasoningTextDoneEvent, aS as OpenAIResponsesRefusalDeltaEvent, aT as OpenAIResponsesRefusalDoneEvent, j as OpenAIResponsesRequest, k as OpenAIResponsesResponse, aU as OpenAIResponsesStatus, aV as OpenAIResponsesStream, aW as OpenAIResponsesStreamError, l as OpenAIResponsesStreamEvent, aX as OpenAIResponsesUsage, aY as OpenAIStreamOptions, aZ as OpenAITextContentPart, a_ as OpenAITool, a$ as OpenAIToolCall, b0 as OpenAIToolCallDelta, b1 as OpenAIUsage, b2 as SystemOneConfidence, b3 as SystemOneProbabilities } from './anthropic-CAvwqIjx.js';
+import { T as ToolHandler, a as Tool, A as AnyTool, b as ToolRegistry, c as ToolExecutionResult } from './compose-75UFDzOO.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-75UFDzOO.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';
@@ -1655,25 +1655,34 @@ declare function batchEmbed(client: OllamaClient, options: EmbedBatchOptions): P
  */
 
 /**
- * Maximum total request size, in bytes, accepted by `/v1/systemone`
- * when the request body carries a non-empty `images` array.
+ * Whole-request size cap, in bytes, for `/v1/systemone` — the maximum
+ * serialized size of the complete request, base64 image bytes included.
  *
- * The Ollama server enforces this limit server-side and rejects
+ * Part of the dual cap clarified upstream on 2026-10-16: every request
+ * must satisfy BOTH bounds simultaneously — this whole-request cap AND
+ * the text-and-schema cap of {@link MAX_SYSTEM_ONE_REQUEST_BYTES}
+ * (64 KiB, images excluded). The bumped whole-request cap exists
+ * specifically to accommodate base64-encoded image bytes shared across
+ * all questions in a single System One call; the text and schema
+ * portion may never exceed 64 KiB regardless of whether images are
+ * attached.
+ *
+ * The Ollama server enforces both bounds server-side and rejects
  * oversized payloads with HTTP 413. The SDK's runtime also enforces
- * it client-side (see `OllamaRuntime`'s `maxRequestBytesWithImages`
+ * them client-side (see `OllamaRuntime`'s `maxRequestBytesWithImages`
  * handling) so the round-trip is avoided.
- *
- * Without images, the limit is the much smaller
- * {@link MAX_SYSTEM_ONE_REQUEST_BYTES} (64 KiB) — the bumped
- * images-cap exists specifically to accommodate base64-encoded image
- * bytes shared across all questions in a single System One call.
  */
 declare const MAX_SYSTEM_ONE_IMAGES_BYTES: number;
 /**
- * Maximum total request size, in bytes, for `/v1/systemone` when no
- * images are attached. The Ollama server caps the JSON-only payload
- * at 64 KiB; the runtime enforces this client-side (see
- * `OllamaRuntime`'s `maxRequestBytes`).
+ * Text-and-schema size cap, in bytes, for `/v1/systemone` — the maximum
+ * serialized size of the request with its `images` array excluded.
+ *
+ * Part of the dual cap clarified upstream on 2026-10-16: this bound
+ * applies to EVERY request, images or not — attaching images raises
+ * the whole-request ceiling to {@link MAX_SYSTEM_ONE_IMAGES_BYTES}
+ * (32 MiB) but never relaxes this 64 KiB text-and-schema ceiling. The
+ * runtime enforces both bounds client-side (see `OllamaRuntime`'s
+ * `maxRequestBytes` handling).
  */
 declare const MAX_SYSTEM_ONE_REQUEST_BYTES: number;
 /**
@@ -1740,16 +1749,33 @@ type SystemOneAnswers<Q extends SystemOneQuestions = SystemOneQuestions> = {
  * ## Image support
  *
  * The `images` field accepts up to N base64-encoded images shared by
- * all questions in the request. Total request payload (JSON + images)
- * must remain within {@link MAX_SYSTEM_ONE_IMAGES_BYTES} (32 MiB);
- * requests without images are capped at
- * {@link MAX_SYSTEM_ONE_REQUEST_BYTES} (64 KiB).
+ * all questions in the request. The dual cap (2026-10-16 upstream
+ * clarification) applies to every request simultaneously: the text and
+ * schema portion (images excluded) must stay within
+ * {@link MAX_SYSTEM_ONE_REQUEST_BYTES} (64 KiB), and the whole payload
+ * (JSON + images) within {@link MAX_SYSTEM_ONE_IMAGES_BYTES} (32 MiB).
  *
  * Use {@link estimateSystemOneRequestBytes} for a client-side pre-flight
  * check before sending a large multi-image batch — the runtime also
- * enforces the limit, but the early check avoids the round-trip when
+ * enforces both caps, but the early check avoids the round-trip when
  * the caller already knows the payload is too large.
  *
+ * ## Context window
+ *
+ * Beyond the byte caps, upstream requires the complete input (state,
+ * questions, and images) to fit the model's loaded context window, and
+ * System One never truncates input — oversized prompts are rejected by
+ * the server with HTTP 400 locally ("a rendered prompt that exceeds the
+ * loaded context"), not 413 (which is reserved for the byte caps
+ * above); for cloud requests, upstream documents 413 for input that
+ * exceeds the model's context window (65,536 tokens) as well as for
+ * requests over 13 MiB.
+ * Decision models (Nimble, Tev) additionally reserve two token
+ * positions at the end of the window for scoring. The SDK deliberately
+ * does not estimate this client-side (upstream documents no
+ * tokenization formula); callers who need the real limit can use
+ * `client.models.getContextLength()`.
+ *
  * Requires Clef or Clef Flash with vision weights. URLs and data URLs
  * are NOT supported — pre-encode the bytes to base64 before adding to
  * the array.
@@ -1768,8 +1794,12 @@ interface SystemOneResponse<Q extends SystemOneQuestions = SystemOneQuestions> e
 }
 /**
  * Estimate the on-the-wire byte size of a System One request body, for
- * pre-flight size validation against {@link MAX_SYSTEM_ONE_REQUEST_BYTES}
- * (no images) or {@link MAX_SYSTEM_ONE_IMAGES_BYTES} (with images).
+ * pre-flight validation against the dual cap (2026-10-16 upstream
+ * clarification): the whole body against
+ * {@link MAX_SYSTEM_ONE_IMAGES_BYTES} (32 MiB) and, when `images` is
+ * attached, the images-excluded portion against
+ * {@link MAX_SYSTEM_ONE_REQUEST_BYTES} (64 KiB) — pass
+ * `{ excludeImages: true }` to measure the text-and-schema portion.
  *
  * The estimate is conservative: it uses `JSON.stringify` length and
  * assumes UTF-8 encoding. The actual wire size may differ slightly due
@@ -1781,12 +1811,18 @@ interface SystemOneResponse<Q extends SystemOneQuestions = SystemOneQuestions> e
  *   ```ts
  *   const request: SystemOneRequest = { ... };
  *   const bytes = estimateSystemOneRequestBytes(request);
- *   const cap = request.images?.length ? MAX_SYSTEM_ONE_IMAGES_BYTES : MAX_SYSTEM_ONE_REQUEST_BYTES;
- *   if (bytes > cap) throw new Error(`payload ${bytes}B exceeds cap ${cap}B`);
+ *   const textBytes = request.images?.length
+ *     ? estimateSystemOneRequestBytes(request, { excludeImages: true })
+ *     : bytes;
+ *   if (textBytes > MAX_SYSTEM_ONE_REQUEST_BYTES || bytes > MAX_SYSTEM_ONE_IMAGES_BYTES) {
+ *     throw new Error(`payload ${bytes}B (text ${textBytes}B) exceeds the dual cap`);
+ *   }
  *   await client.systemOne(request);
  *   ```
  */
-declare function estimateSystemOneRequestBytes(request: SystemOneRequest | SystemOneRequest$1): number;
+declare function estimateSystemOneRequestBytes(request: SystemOneRequest | SystemOneRequest$1, options?: {
+    readonly excludeImages?: boolean;
+}): number;
 
 /**
  * Higher-level System One decision helpers.
@@ -2363,11 +2399,15 @@ declare class OllamaClient {
      * 1–64 typed questions (choice, noul, or score) against a provided state
      * and returns typed answers with probabilities and confidence.
      *
-     * Local-only; requires Ollama >= 0.35.0. The runtime enforces the
-     * version constraint automatically (lazily fetching `/api/version` when
-     * `enforceVersion` is `'auto'` or `'strict'`) and rejects cloud-mode
-     * calls. Request size is checked client-side: 64 KiB without images,
-     * 32 MiB with images.
+     * Local and cloud (since the 2026-10-16 upstream System One cloud
+     * launch): `model` also accepts cloud models such as `clef:cloud` —
+     * cloud models require Ollama >= 0.40.5 or the hosted
+     * `https://ollama.com` endpoint. Requires Ollama >= 0.35.0 locally.
+     * The runtime enforces the version constraint automatically (lazily
+     * fetching `/api/version` when `enforceVersion` is `'auto'` or
+     * `'strict'`). Request size is checked client-side as a dual cap:
+     * text and schema (images excluded) within 64 KiB AND the whole
+     * request within 32 MiB.
      *
      * The generic `Q` parameter captures the caller's question map at
      * compile time, giving key-safe answer access:

package/dist/index.js

@@ -191,6 +191,11 @@ var OllamaAuthError = class extends OllamaClientError {
     super(message, { ...options, code: "auth_error", retryable: false });
   }
 };
+var OllamaPaymentRequiredError = class extends OllamaClientError {
+  constructor(message, options) {
+    super(message, { ...options, code: "payment_required", retryable: false, status: 402 });
+  }
+};
 var OllamaNotFoundError = class extends OllamaClientError {
   constructor(message, options) {
     super(message, { ...options, code: "not_found", retryable: false });
@@ -425,6 +430,9 @@ function statusToError(status, message, options) {
   if (status === 401 || status === 403) {
     return new OllamaAuthError(message, { ...options, status });
   }
+  if (status === 402) {
+    return new OllamaPaymentRequiredError(message, options);
+  }
   if (status === 404) {
     return new OllamaNotFoundError(message, { ...options, status });
   }
@@ -4423,14 +4431,25 @@ var OllamaRuntime = class {
       const hasImages = bodyHasImages(body);
       const maxBytesWithImages = req.operation.constraints?.maxRequestBytesWithImages;
       const maxBytesBase = req.operation.constraints?.maxRequestBytes;
-      const applicableLimit = hasImages && maxBytesWithImages !== void 0 ? maxBytesWithImages : maxBytesBase;
-      if (applicableLimit !== void 0 && serialized.byteLength > applicableLimit) {
+      const textAndSchemaBytes = hasImages ? serializeWithoutImages(body).byteLength : serialized.byteLength;
+      if (maxBytesBase !== void 0 && textAndSchemaBytes > maxBytesBase) {
         throw new OllamaRequestTooLargeError(
-          `Operation ${req.operation.operationId} (${req.operation.method} ${req.operation.path}) request body is ${serialized.byteLength} bytes, exceeding the contract limit of ${applicableLimit} bytes` + (hasImages ? " (images limit)." : "."),
+          `Operation ${req.operation.operationId} (${req.operation.method} ${req.operation.path}) text and schema (images excluded) is ${textAndSchemaBytes} bytes, exceeding the contract limit of ${maxBytesBase} bytes.`,
+          {
+            operationId: req.operation.operationId,
+            actualBytes: textAndSchemaBytes,
+            maxBytes: maxBytesBase,
+            request: { method: req.operation.method, url: req.operation.path }
+          }
+        );
+      }
+      if (maxBytesWithImages !== void 0 && serialized.byteLength > maxBytesWithImages) {
+        throw new OllamaRequestTooLargeError(
+          `Operation ${req.operation.operationId} (${req.operation.method} ${req.operation.path}) request body is ${serialized.byteLength} bytes, exceeding the contract limit of ${maxBytesWithImages} bytes (whole request, images included).`,
           {
             operationId: req.operation.operationId,
             actualBytes: serialized.byteLength,
-            maxBytes: applicableLimit,
+            maxBytes: maxBytesWithImages,
             request: { method: req.operation.method, url: req.operation.path }
           }
         );
@@ -4572,6 +4591,13 @@ function bodyHasImages(body) {
   const images = obj.images;
   return Array.isArray(images) && images.length > 0;
 }
+function serializeWithoutImages(body) {
+  if (typeof body === "object" && body !== null && !Array.isArray(body)) {
+    const { images: _images, ...rest } = body;
+    return serializeForByteCount(rest);
+  }
+  return serializeForByteCount(body);
+}
 function resolvePathParams(pathTemplate, pathParams) {
   if (!pathTemplate.includes("{")) return pathTemplate;
   const params = pathParams ?? {};
@@ -5324,7 +5350,7 @@ var systemOneOp = {
   path: "/v1/systemone",
   request: { $ref: "#/schemas/SystemOneRequest" },
   response: { $ref: "#/schemas/SystemOneResponse" },
-  environment: { local: true, cloud: false },
+  environment: { local: true, cloud: true },
   transport: { mode: "json", streaming: false },
   capabilities: {
     thinking: "unsupported",
@@ -5342,8 +5368,10 @@ var systemOneOp = {
   status: { documented: true, experimental: true },
   domain: "native",
   notes: [
-    "Local-only; Ollama Cloud rejects /v1/systemone.",
-    "64 KiB request limit without images, 32 MiB with images (includes JSON/base64 overhead).",
+    "Local and cloud: model accepts cloud models (e.g. clef:cloud); cloud models require Ollama >= 0.40.5 or the hosted https://ollama.com endpoint.",
+    "Local dual cap: text and schema (excluding images) within 64 KiB AND the whole request within 32 MiB, including base64 and JSON overhead.",
+    "Cloud requests: paid plan or usage credits required; up to 64 questions named with up to 100 ASCII letters, digits, `_`, `.`, or `-`; up to 4 images of at most 4 MiB and 16 megapixels each (8 MiB total); requests within 13 MiB; input within the model's 65,536-token context window.",
+    "401 when a cloud model is requested without signing in or with an invalid API key (through a local server the response includes a signin_url); 402 when the cloud model is not included in the Free plan or a payment fails.",
     "No tools, no generation controls (temperature, top_p, etc.).",
     "Vision is model-dependent: images are accepted by vision-weighted models (Clef/Clef Flash) but not by text-only models.",
     "Experimental: behavioral surface may change between Ollama releases."
@@ -6299,11 +6327,15 @@ var OllamaClient = class _OllamaClient {
    * 1–64 typed questions (choice, noul, or score) against a provided state
    * and returns typed answers with probabilities and confidence.
    *
-   * Local-only; requires Ollama >= 0.35.0. The runtime enforces the
-   * version constraint automatically (lazily fetching `/api/version` when
-   * `enforceVersion` is `'auto'` or `'strict'`) and rejects cloud-mode
-   * calls. Request size is checked client-side: 64 KiB without images,
-   * 32 MiB with images.
+   * Local and cloud (since the 2026-10-16 upstream System One cloud
+   * launch): `model` also accepts cloud models such as `clef:cloud` —
+   * cloud models require Ollama >= 0.40.5 or the hosted
+   * `https://ollama.com` endpoint. Requires Ollama >= 0.35.0 locally.
+   * The runtime enforces the version constraint automatically (lazily
+   * fetching `/api/version` when `enforceVersion` is `'auto'` or
+   * `'strict'`). Request size is checked client-side as a dual cap:
+   * text and schema (images excluded) within 64 KiB AND the whole
+   * request within 32 MiB.
    *
    * The generic `Q` parameter captures the caller's question map at
    * compile time, giving key-safe answer access:
@@ -8254,10 +8286,11 @@ function createModelAffinityScheduler(client, options) {
 // src/system-one.ts
 var MAX_SYSTEM_ONE_IMAGES_BYTES = 32 * 1024 * 1024;
 var MAX_SYSTEM_ONE_REQUEST_BYTES = 64 * 1024;
-function estimateSystemOneRequestBytes(request) {
-  return new TextEncoder().encode(JSON.stringify(request)).byteLength;
+function estimateSystemOneRequestBytes(request, options) {
+  const body = options?.excludeImages && request.images !== void 0 ? { ...request, images: void 0 } : request;
+  return new TextEncoder().encode(JSON.stringify(body)).byteLength;
 }
 
-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, Agent, AnthropicCompatClient, AnthropicMessagesStream, ConversationSession, 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, EndpointRegistry, GEN_AI_SYSTEM_OLLAMA, HttpClient, IMAGE_TOKEN_ESTIMATE, KEEP_ALIVE_INDEFINITE, KEEP_ALIVE_UNLOAD, KNOWN_DONE_REASONS, MAX_SYSTEM_ONE_IMAGES_BYTES, MAX_SYSTEM_ONE_REQUEST_BYTES, MESSAGE_OVERHEAD_TOKENS, McpBridge, ModelAffinityRouter, ModelAffinityScheduler, ModelsClient, NANOS_PER_MS, NANOS_PER_SECOND, NETWORK_ERROR_CODES, NOOP_LOGGER, OLLAMA_CLOUD_BASE_URL, OLLAMA_FALLBACK_CONTEXT_LENGTH, OllamaAbortError, OllamaAgentMaxIterationsError, OllamaAgentMaxToolCallsError, OllamaAgentToolLoopError, OllamaAuthError, OllamaBadGatewayError, OllamaClient, OllamaClientError, OllamaGenericClientError, OllamaIncompatibleModelError, OllamaMcpError, OllamaModelRoutingError, OllamaNetworkError, OllamaNotFoundError, OllamaOverloadedError, OllamaQuotaExceededError, OllamaRateLimitError, OllamaRequestTooLargeError, OllamaRequestValidationError, OllamaResponseValidationError, OllamaServerError, OllamaServerVersionUnknownError, OllamaSkillInvalidError, OllamaSkillNotFoundError, OllamaStream, OllamaStreamError, OllamaTimeoutError, OllamaToolTimeoutError, OllamaToolValidationError, OllamaUnsupportedCapabilityError, OpenAIChatCompletionStream, OpenAICompatClient, OpenAICompletionStream, OpenAIResponsesStream, OpenAIResponsesStreamError, QuotaManager, RETRYABLE_STATUS_CODES, ResponsesModule, ToolCallAccumulator, ToolRegistry, WEB_FETCH_TOOL_NAME, WEB_SEARCH_TOOL_NAME, addSpanEvent, applySkill, batchEmbed, calculateBackoff, canonicalToolCallSignature, checkChatContext, checkEndpointHealth, checkGenerateContext, compactConversationHistory, compareDraftPerformance, composeMiddleware, computeProgressPercent, contextWarningMessage, createConsoleLogger, createDecision, createModelAffinityScheduler, createOllamaCloudFreeTierQuota, createProgressSpeedTracker, createTimeoutSignal, createWebFetchTool, createWebSearchTool, createWebTools, defineTool, detectModelCapabilities, embedBatchOverflowMessage, encodeImage, estimateChatRequestTokens, estimateGenerateRequestTokens, estimateMessageTokens, estimateSystemOneRequestBytes, estimateTokens, extractContextLength as extractNativeContextLength, extractParameterNumCtx, extractUsage, fetchWithBackoff, findOversizedEmbedInputs, findRunningModelContextLength, formatTelemetry, formatWebFetchResult, formatWebSearchResult, imageStringNeedsResolution, inferRuntimeMode, isKeepAliveSugar, isKnownDoneReason, isModelResident, isNetworkLevelError, isSameToolCall, listAllMcpTools, listAvailableModels, loadMcpTools, mapError, mergeToolCall, mergeToolCallArgumentsString, mergeToolCallArrays, normalizeChatStream, normalizeGenerateStream, normalizeKeepAlive, normalizeProgressStream, observeStream, onProgress, parseFrontmatter, parseNdjsonStream, parseSseStream, parseStructuredOutput, recommendContextWindow, registerMcpTools, resolveContextLength, resolveCredentialEndpoints, resolveImageInput, sanitizeHistoryForNextTurn, setSpanError, toDataStream, toPullProgressEvent, 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, Agent, AnthropicCompatClient, AnthropicMessagesStream, ConversationSession, 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, EndpointRegistry, GEN_AI_SYSTEM_OLLAMA, HttpClient, IMAGE_TOKEN_ESTIMATE, KEEP_ALIVE_INDEFINITE, KEEP_ALIVE_UNLOAD, KNOWN_DONE_REASONS, MAX_SYSTEM_ONE_IMAGES_BYTES, MAX_SYSTEM_ONE_REQUEST_BYTES, MESSAGE_OVERHEAD_TOKENS, McpBridge, ModelAffinityRouter, ModelAffinityScheduler, ModelsClient, NANOS_PER_MS, NANOS_PER_SECOND, NETWORK_ERROR_CODES, NOOP_LOGGER, OLLAMA_CLOUD_BASE_URL, OLLAMA_FALLBACK_CONTEXT_LENGTH, OllamaAbortError, OllamaAgentMaxIterationsError, OllamaAgentMaxToolCallsError, OllamaAgentToolLoopError, OllamaAuthError, OllamaBadGatewayError, OllamaClient, OllamaClientError, OllamaGenericClientError, OllamaIncompatibleModelError, OllamaMcpError, OllamaModelRoutingError, OllamaNetworkError, OllamaNotFoundError, OllamaOverloadedError, OllamaPaymentRequiredError, OllamaQuotaExceededError, OllamaRateLimitError, OllamaRequestTooLargeError, OllamaRequestValidationError, OllamaResponseValidationError, OllamaServerError, OllamaServerVersionUnknownError, OllamaSkillInvalidError, OllamaSkillNotFoundError, OllamaStream, OllamaStreamError, OllamaTimeoutError, OllamaToolTimeoutError, OllamaToolValidationError, OllamaUnsupportedCapabilityError, OpenAIChatCompletionStream, OpenAICompatClient, OpenAICompletionStream, OpenAIResponsesStream, OpenAIResponsesStreamError, QuotaManager, RETRYABLE_STATUS_CODES, ResponsesModule, ToolCallAccumulator, ToolRegistry, WEB_FETCH_TOOL_NAME, WEB_SEARCH_TOOL_NAME, addSpanEvent, applySkill, batchEmbed, calculateBackoff, canonicalToolCallSignature, checkChatContext, checkEndpointHealth, checkGenerateContext, compactConversationHistory, compareDraftPerformance, composeMiddleware, computeProgressPercent, contextWarningMessage, createConsoleLogger, createDecision, createModelAffinityScheduler, createOllamaCloudFreeTierQuota, createProgressSpeedTracker, createTimeoutSignal, createWebFetchTool, createWebSearchTool, createWebTools, defineTool, detectModelCapabilities, embedBatchOverflowMessage, encodeImage, estimateChatRequestTokens, estimateGenerateRequestTokens, estimateMessageTokens, estimateSystemOneRequestBytes, estimateTokens, extractContextLength as extractNativeContextLength, extractParameterNumCtx, extractUsage, fetchWithBackoff, findOversizedEmbedInputs, findRunningModelContextLength, formatTelemetry, formatWebFetchResult, formatWebSearchResult, imageStringNeedsResolution, inferRuntimeMode, isKeepAliveSugar, isKnownDoneReason, isModelResident, isNetworkLevelError, isSameToolCall, listAllMcpTools, listAvailableModels, loadMcpTools, mapError, mergeToolCall, mergeToolCallArgumentsString, mergeToolCallArrays, normalizeChatStream, normalizeGenerateStream, normalizeKeepAlive, normalizeProgressStream, observeStream, onProgress, parseFrontmatter, parseNdjsonStream, parseSseStream, parseStructuredOutput, recommendContextWindow, registerMcpTools, resolveContextLength, resolveCredentialEndpoints, resolveImageInput, sanitizeHistoryForNextTurn, setSpanError, toDataStream, toPullProgressEvent, toResponse, toTextStream, withRetry, withSpan, zodToJsonSchema };
 //# sourceMappingURL=index.js.map
 //# sourceMappingURL=index.js.map