@nemesis-oss/ollama-sdk 1.16.0 → 1.18.0

package/dist/index.d.ts

@@ -3,8 +3,8 @@ export { L as ListResponse, x as Logprob, y as LogprobToken, z as ModelDetails,
 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 { 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-CTv2acbj.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-CTv2acbj.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-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 { M as McpClientLike, a as McpToolDescriptor, b as McpTask, c as McpCallToolResult } from './types-CI-gM2u5.js';
@@ -970,8 +970,8 @@ declare class ModelsClient {
      * model-not-found (404), or VRAM-exhaustion failures.
      */
     pin(model: string): Promise<void>;
-    createBlob(digest: string, data: BinaryBody): Promise<void>;
-    checkBlob(digest: string): Promise<boolean>;
+    createBlob(digest: string, data: BinaryBody, request?: RequestCancellationOptions): Promise<void>;
+    checkBlob(digest: string, request?: RequestCancellationOptions): Promise<boolean>;
     /**
      * Computes a blob digest the way Ollama's content-addressed store expects:
      * SHA-256 over the raw bytes, formatted as `sha256:<64 lowercase hex>`.
@@ -992,17 +992,44 @@ declare class ModelsClient {
      */
     createBlobFromData(data: Uint8Array, opts?: {
         readonly digest?: string | undefined;
-        readonly signal?: AbortSignal | undefined;
-    }): Promise<BlobUploadResult>;
+    } & RequestCancellationOptions): Promise<BlobUploadResult>;
+    /**
+     * Streams a blob upload straight from a web `ReadableStream` — the
+     * multi-gigabyte-safe counterpart of {@link createBlobFromData}. The body
+     * is never materialized in memory: the transport hands the stream to
+     * `fetch` directly (setting `duplex: 'half'`, which Node 18+ requires for
+     * stream bodies), so peak heap usage is a transport-level chunk, not the
+     * blob size.
+     *
+     * The digest must be known up front — it *is* the upload URL
+     * (`/api/blobs/<digest>`), the same content-addressed protocol
+     * `curl -T` performs. Hash first when you don't have it: in-memory bytes
+     * via {@link computeBlobDigest}, files via {@link createBlobFromFile}
+     * (which hashes in a streaming pass). The server verifies the streamed
+     * bytes against the digest and rejects mismatches, so a corrupted or
+     * truncated stream fails the upload instead of poisoning the store.
+     */
+    createBlobFromStream(digest: string, stream: ReadableStream<Uint8Array>, request?: RequestCancellationOptions): Promise<BlobUploadResult>;
     /**
      * Reads a local file and uploads it as a blob (Node.js only — dynamic
      * `node:fs` import, so browser bundles never pull it in). The file name
      * (basename) is what `/api/create`'s `files` map expects as the key, so
      * this returns it alongside the digest.
+     *
+     * Streams both passes in fixed-size chunks — peak heap is O(chunk), not
+     * O(file), so multi-gigabyte GGUFs never enter the Node heap:
+     *
+     *  1. **Hash pass** (skipped when `opts.digest` is provided): the file is
+     *     streamed through `node:crypto`'s incremental SHA-256.
+     *  2. **Upload pass**: `node:fs` createReadStream → web ReadableStream →
+     *     `fetch` body with `duplex: 'half'` (see {@link createBlobFromStream}).
+     *
+     * A pre-existing blob (HEAD 200) skips the upload pass entirely, same as
+     * {@link createBlobFromData}.
      */
     createBlobFromFile(path: string, opts?: {
-        readonly signal?: AbortSignal | undefined;
-    }): Promise<BlobUploadResult & {
+        readonly digest?: string | undefined;
+    } & RequestCancellationOptions): Promise<BlobUploadResult & {
         readonly fileName: string;
     }>;
     /**
@@ -1790,6 +1817,14 @@ interface ChoiceDecision {
     readonly probabilities: Record<string, number>;
     /** Distribution concentration (0–1). NOT a guarantee of correctness. */
     readonly confidence: number;
+    /**
+     * Present only when `minConfidence` was configured: `true` when
+     * `confidence >= minConfidence`. Autonomous flows should treat `false` as
+     * "do not act on this decision" — the official System One docs warn that a
+     * high confidence does not guarantee correctness, and a barely-majority
+     * choice (e.g. 0.51) is rarely safe to act on unattended.
+     */
+    readonly isConfident?: boolean;
     /** Token usage for the call. */
     readonly usage: {
         readonly input_tokens: number;
@@ -1803,7 +1838,11 @@ interface ChoiceDecision {
 interface NoulDecision {
     /** Probability of true (0–1). */
     readonly noul: number;
-    /** Boolean decision: true if noul >= 0.5, false otherwise. */
+    /**
+     * Boolean decision: `true` if `noul >= minProbability` (default 0.5).
+     * Raise the bar for safety gates with `minProbability` — e.g. 0.8 for
+     * irreversible actions.
+     */
     readonly bool: boolean;
     /** Token usage for the call. */
     readonly usage: {
@@ -1825,6 +1864,8 @@ interface ScoreDecision {
     readonly probabilities: Record<string, number>;
     /** Distribution concentration (0–1). NOT a guarantee of correctness. */
     readonly confidence: number;
+    /** Present only when `minConfidence` was configured: `confidence >= minConfidence`. */
+    readonly isConfident?: boolean;
     /** Token usage for the call. */
     readonly usage: {
         readonly input_tokens: number;
@@ -1839,6 +1880,8 @@ interface RankResult<T extends string> {
     readonly candidate: T;
     readonly score: number;
     readonly confidence: number;
+    /** Present only when `minConfidence` was configured: `confidence >= minConfidence`. */
+    readonly isConfident?: boolean;
 }
 /**
  * Higher-level System One decision helpers.
@@ -1879,6 +1922,12 @@ interface Decision {
         criteria: Record<string, string | null>;
         images?: string[];
         keepAlive?: string | number;
+        /**
+         * Opt-in confidence gate in (0, 1]. When set, the result carries
+         * `isConfident: confidence >= minConfidence` so autonomous flows can
+         * refuse to act on barely-majority decisions. Off by default.
+         */
+        minConfidence?: number;
     }): Promise<ChoiceDecision>;
     /**
      * Ask a single yes/no (noul) question. Returns the probability of
@@ -1893,6 +1942,12 @@ interface Decision {
         instructions: SystemOneContent;
         images?: string[];
         keepAlive?: string | number;
+        /**
+         * Inclusive lower bound in (0, 1] the probability of `true` must meet
+         * for `bool` to be `true`. Default 0.5 (a coin-flip majority). Raise it
+         * (e.g. 0.8) for safety gates and irreversible actions.
+         */
+        minProbability?: number;
     }): Promise<NoulDecision>;
     /**
      * Ask a single score question. Returns the probability-weighted
@@ -1905,10 +1960,12 @@ interface Decision {
         model: string;
         state: SystemOneContent;
         instructions: SystemOneContent;
-        /** Ordered criteria from lowest (index 0) to highest. 2–26 items. */
+        /** Ordered criteria from lowest (index 0) to highest. 2–10 items. */
         criteria: string[];
         images?: string[];
         keepAlive?: string | number;
+        /** Opt-in confidence gate in (0, 1] — see {@link Decision.choice}. */
+        minConfidence?: number;
     }): Promise<ScoreDecision>;
     /**
      * Route a request to one of N named options. Convenience wrapper
@@ -1924,9 +1981,12 @@ interface Decision {
         criteria: Record<T, string | null>;
         images?: string[];
         keepAlive?: string | number;
+        /** Opt-in confidence gate in (0, 1] — see {@link Decision.choice}. */
+        minConfidence?: number;
     }): Promise<{
         readonly route: T;
         readonly confidence: number;
+        readonly isConfident?: boolean;
     }>;
     /**
      * Verify whether a claim is supported by evidence. Convenience
@@ -1945,6 +2005,11 @@ interface Decision {
         instructions?: SystemOneContent;
         images?: string[];
         keepAlive?: string | number;
+        /**
+         * Inclusive lower bound in (0, 1] the probability of `true` must meet
+         * for `verified`. Default 0.5. See {@link Decision.noul}.
+         */
+        minProbability?: number;
     }): Promise<{
         readonly verified: boolean;
         readonly probability: number;
@@ -1964,7 +2029,7 @@ interface Decision {
     rank<T extends string>(params: {
         model: string;
         instructions: SystemOneContent;
-        /** Ordered criteria from lowest (index 0) to highest. 2–26 items. */
+        /** Ordered criteria from lowest (index 0) to highest. 2–10 items. */
         criteria: string[];
         candidates: ReadonlyArray<{
             readonly id: T;
@@ -1972,6 +2037,8 @@ interface Decision {
         }>;
         images?: string[];
         keepAlive?: string | number;
+        /** Opt-in confidence gate in (0, 1] — flags each result with `isConfident`. */
+        minConfidence?: number;
     }): Promise<ReadonlyArray<RankResult<T>>>;
 }
 /** Create a Decision helper bound to an OllamaClient. */
@@ -2277,8 +2344,8 @@ declare class OllamaClient {
      */
     readonly getContextLength: (request: ContextDiscoveryRequestOptions) => Promise<DiscoveredContextLength>;
     readonly version: () => Promise<VersionResponse>;
-    readonly createBlob: (digest: string, data: BinaryBody) => Promise<void>;
-    readonly checkBlob: (digest: string) => Promise<boolean>;
+    readonly createBlob: (digest: string, data: BinaryBody, request?: RequestCancellationOptions) => Promise<void>;
+    readonly checkBlob: (digest: string, request?: RequestCancellationOptions) => Promise<boolean>;
     /**
      * Immediately unloads `model` from VRAM. Convenience alias for
      * {@link ModelsClient.unload} — see that method for the full
@@ -2780,6 +2847,180 @@ declare function toDataStream<TChunk, TFinal>(stream: OllamaStream<TChunk, TFina
  */
 declare function toResponse<TChunk, TFinal>(stream: OllamaStream<TChunk, TFinal>, init?: ResponseInit): Response;
 
+/**
+ * Typed progress callback helpers for `/api/pull` and `/api/push` streams.
+ *
+ * Ollama emits newline-delimited progress frames of the shape:
+ *
+ * ```json
+ * {"status":"pulling manifest"}
+ * {"status":"downloading","digest":"sha256:...","total":1234567,"completed":123}
+ * {"status":"verifying sha256 digest"}
+ * {"status":"writing manifest","digest":"sha256:...","total":42,"completed":42}
+ * {"status":"success"}
+ * ```
+ *
+ * The SDK already exposes these as a typed {@link OllamaStream} via
+ * {@link ModelsClient.pull} / {@link ModelsClient.push} when called
+ * with `{ stream: true }`. This file adds a higher-level helper that
+ * invokes a typed callback on every progress event with a pre-computed
+ * `percent` field — for callers who want a fire-and-forget pull/push
+ * without manually iterating the stream.
+ *
+ * ## When to use this vs. the raw `OllamaStream`
+ *
+ * - **`onProgress(stream, cb)`** — best for download/upload UIs that
+ *   just need to render a progress bar. Resolves on `success`,
+ *   rejects on stream error.
+ * - **Raw `for await ... of stream`** — best for callers who want
+ *   every chunk as a typed event (e.g. for structured logging, or
+ *   to react to specific `status` values like `"downloading"` vs
+ *   `"verifying sha256 digest"`).
+ */
+
+/**
+ * A single progress event delivered to {@link onProgress}'s callback.
+ *
+ * `percent` is `undefined` when the server hasn't reported byte
+ * totals (e.g. for `pulling manifest`, `verifying sha256 digest`,
+ * `success` — these status lines carry no `total`/`completed`).
+ * For `downloading` / `writing manifest` / `writing layer` style
+ * events, `percent` is the ratio of `completed / total`, clamped
+ * to `[0, 100]` and rounded to two decimals.
+ *
+ * `bytesPerSecond` and `etaSeconds` are populated by {@link onProgress}
+ * (via {@link createProgressSpeedTracker}) once at least two
+ * byte-bearing frames for the same layer have arrived. Callers using
+ * {@link toPullProgressEvent} directly get them from their own tracker
+ * — the pure wrapper never measures time.
+ */
+interface PullProgressEvent {
+    /** Ollama's status string (`"pulling manifest"`, `"downloading"`, `"success"`, …). */
+    readonly status: string;
+    /** SHA-256 digest of the layer being transferred, when applicable. */
+    readonly digest?: string | undefined;
+    /** Total bytes for this transfer, when the server has reported it. */
+    readonly total?: number | undefined;
+    /** Bytes transferred so far, when the server has reported it. */
+    readonly completed?: number | undefined;
+    /** Completion percentage in `[0, 100]`, or `undefined` when no byte total. */
+    readonly percent?: number | undefined;
+    /**
+     * Cumulative-average transfer speed for the current layer, in bytes per
+     * second (measured since the layer's first byte-bearing frame). `undefined`
+     * until a second byte-bearing frame of the same layer arrives.
+     */
+    readonly bytesPerSecond?: number | undefined;
+    /**
+     * Estimated seconds remaining for the current layer at the observed
+     * speed: `(total - completed) / bytesPerSecond`. `undefined` while the
+     * speed is unknown or the server hasn't reported a total; `0` once
+     * `completed >= total`.
+     */
+    readonly etaSeconds?: number | undefined;
+    /** The raw upstream frame, for callers who need fields this helper skips. */
+    readonly raw: ProgressResponse;
+}
+/**
+ * Callback shape accepted by {@link onProgress}.
+ */
+type ProgressCallback = (event: PullProgressEvent) => void;
+/**
+ * Compute the completion percentage from a progress chunk.
+ *
+ * Returns `undefined` when either `total` or `completed` is missing
+ * or zero (the server emits byte totals only for `downloading` /
+ * `writing manifest` / `writing layer` style events; for
+ * `pulling manifest`, `verifying sha256 digest`, `success`, etc. the
+ * totals are absent and percent is undefined).
+ *
+ * When both are present, returns `Math.round((completed / total) * 10000) / 100`
+ * — i.e. percent with two decimal places (e.g. `42.37`), clamped to
+ * `[0, 100]` to guard against upstream reporting inconsistencies
+ * (sometimes the server briefly reports `completed > total` during
+ * dedup / verification phases).
+ */
+declare function computeProgressPercent(chunk: ProgressResponse): number | undefined;
+/**
+ * Wrap a {@link ProgressResponse} into a {@link PullProgressEvent}
+ * with a pre-computed `percent` field.
+ */
+declare function toPullProgressEvent(chunk: ProgressResponse): PullProgressEvent;
+/**
+ * Speed/ETA readings produced by {@link createProgressSpeedTracker}.
+ */
+interface ProgressSpeed {
+    /** Cumulative-average bytes/second for the current layer, once measurable. */
+    readonly bytesPerSecond?: number | undefined;
+    /** Estimated seconds remaining for the current layer; `0` once complete. */
+    readonly etaSeconds?: number | undefined;
+}
+/**
+ * A stateful per-layer transfer-speed measurer. Feed it every raw
+ * `ProgressResponse` chunk, in arrival order; it returns the current
+ * speed/ETA readings for that chunk's layer.
+ *
+ * This is the building block {@link onProgress} uses internally, exported
+ * for consumers who iterate the raw `OllamaStream` themselves (structured
+ * logging, custom dashboards) and would otherwise have to hand-roll a
+ * stateful byte counter.
+ *
+ * Semantics:
+ * - **Per-layer epochs.** A pull/push transfers several layers; each has
+ *   its own `digest` and byte range, and `completed` restarts from a low
+ *   value when the next layer begins. The tracker starts a fresh epoch
+ *   whenever the digest changes or `completed` regresses, so the speed
+ *   never averages across layers.
+ * - **Cumulative average.** Speed is measured from the layer's first
+ *   byte-bearing frame — `(completed - firstCompleted) / elapsed` — which
+ *   is stable for progress bars (instantaneous frame-to-frame rates jitter
+ *   wildly).
+ * - **Monotonic clock.** `performance.now()`, immune to wall-clock
+ *   adjustments.
+ *
+ * @example
+ * ```ts
+ * const speed = createProgressSpeedTracker();
+ * const stream = await client.models.pull({ model: 'llama3.2', stream: true });
+ * for await (const event of stream) {
+ *   if (event.type === 'message') {
+ *     const { bytesPerSecond, etaSeconds } = speed(event.data.chunk);
+ *     // render: 42.37% at 8.1 MiB/s, ETA 3s
+ *   }
+ * }
+ * ```
+ */
+declare function createProgressSpeedTracker(): (chunk: ProgressResponse) => ProgressSpeed;
+/**
+ * Subscribe to a pull/push stream with a typed progress callback.
+ *
+ * Iterates the stream, invokes `callback` on every progress event
+ * with a pre-computed `percent` field, and resolves when the stream
+ * completes (`success` status). Rejects with {@link OllamaStreamError}
+ * if the stream emits an in-band `{"error": "..."}` frame, or with
+ * whatever error the underlying transport surfaces for HTTP-level
+ * failures.
+ *
+ * @example
+ *   ```ts
+ *   const stream = await client.models.pull({ model: 'llama3.2', stream: true });
+ *   await onProgress(stream, (event) => {
+ *     if (event.percent !== undefined) {
+ *       console.log(`${event.status}: ${event.percent}% (${event.digest?.slice(0, 12) ?? '-'})`);
+ *     } else {
+ *       console.log(event.status);
+ *     }
+ *   });
+ *   console.log('pull complete');
+ *   ```
+ *
+ * @param stream The `OllamaStream` returned by `models.pull({ stream: true })`.
+ * @param callback Invoked on every progress event.
+ * @returns A promise that resolves to the final {@link ProgressStreamResult}
+ *          (carrying `status: 'success'`) when the stream completes.
+ */
+declare function onProgress(stream: OllamaStream<ProgressResponse, ProgressStreamResult>, callback: ProgressCallback): Promise<ProgressStreamResult>;
+
 /**
  * Zero-copy stream observation — the safe alternative to `ReadableStream.tee()`
  * for multi-sink consumption (audit logging, token counting, telemetry taps).
@@ -4022,4 +4263,4 @@ declare class ModelAffinityScheduler {
 /** Convenience factory — `createModelAffinityScheduler(client, opts)` reads better in setups. */
 declare function createModelAffinityScheduler(client: ModelAffinityClient | OllamaClient, options?: ModelAffinitySchedulerOptions): ModelAffinityScheduler;
 
-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, type AffinityRunningModel, 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 CompactionOptions, type CompareDraftPerformanceOptions, type ContextCheck, type ContextCheckOptions, type ContextDiscoveryRequestOptions, type ContextLengthSignals, type ContextLengthSource, type ContextWindowSource, type ConversationSendOptions, ConversationSession, type ConversationSessionOptions, CopyRequestOptions, type CreateModelFromGgufOptions, 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 DiscoveredContextLength, type DraftBenchmarkClient, type DraftPerformanceComparison, type DraftRunStats, type EmbedBatchOptions, type EmbedBatchResult, EmbedRequestOptions, EmbedResponse, EmbeddingsRequestOptions, EmbeddingsResponse, type EndpointHealth, type EndpointHealthCheckResult, EndpointRegistry, type EndpointRegistryOptions, FetchLike, type FetchWithBackoffConfig, FormatOption, type FormattedTelemetry, GEN_AI_SYSTEM_OLLAMA, GenerateRequestOptions, GenerateResponse, GenerateStreamResult, type HistorySanitizeOptions, 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, MESSAGE_OVERHEAD_TOKENS, McpBridge, type McpBridgeOptions, McpCallToolResult, McpClientLike, McpTask, McpToolDescriptor, Message, Middleware, type ModelAffinityClient, ModelAffinityRouter, type ModelAffinityRoutingOptions, ModelAffinityScheduler, type ModelAffinitySchedulerOptions, type ModelAffinitySnapshot, type ModelAffinityStats, 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, PingResult, ProgressResponse, ProgressStreamResult, PsResponse, PullRequestOptions, PushRequestOptions, QuotaManager, type QuotaManagerOptions, type QuotaWindowConfig, type QuotaWindowStatus, RETRYABLE_STATUS_CODES, type RankResult, type RawOllamaMetrics, RawUsageSource, RequestCancellationOptions, RequestLifecycleHook, RequestRunner, type ResponsesCreateRequest, type ResponsesCreateResponse, ResponsesModule, type ResponsesReasoningEffort, type ResponsesStreamEvent, type ResponsesToolDefinition, type ResponsesUsage, type RetryConfig, type RuntimeMode, type ScoreDecision, type SessionCompactionOptions, type SessionCompactionResult, type SessionTurn, ShowRequestOptions, ShowResponse, type SpanAttributeValue, type SpanAttributes, StatusResponse, type StreamObserverCallbacks, type StreamObserverPhase, 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, WEB_FETCH_TOOL_NAME, WEB_SEARCH_TOOL_NAME, WebFetchRequestOptions, WebFetchResponse, type WebFetchToolParams, WebSearchRequestOptions, WebSearchResponse, type WebSearchToolParams, type WebToolsClient, type WebToolsOptions, addSpanEvent, batchEmbed, calculateBackoff, canonicalToolCallSignature, checkChatContext, checkEndpointHealth, checkGenerateContext, compactConversationHistory, compareDraftPerformance, contextWarningMessage, createDecision, createModelAffinityScheduler, createOllamaCloudFreeTierQuota, createTimeoutSignal, createWebFetchTool, createWebSearchTool, createWebTools, defineTool, detectModelCapabilities, embedBatchOverflowMessage, encodeImage, estimateChatRequestTokens, estimateGenerateRequestTokens, estimateMessageTokens, estimateSystemOneRequestBytes, estimateTokens, extractContextLength as extractNativeContextLength, extractParameterNumCtx, fetchWithBackoff, findOversizedEmbedInputs, findRunningModelContextLength, formatTelemetry, formatWebFetchResult, formatWebSearchResult, inferRuntimeMode, isKeepAliveSugar, isKnownDoneReason, isModelResident, isSameToolCall, listAllMcpTools, listAvailableModels, loadMcpTools, mergeToolCall, mergeToolCallArgumentsString, mergeToolCallArrays, normalizeChatStream, normalizeGenerateStream, normalizeKeepAlive, normalizeProgressStream, observeStream, parseNdjsonStream, parseStructuredOutput, recommendContextWindow, registerMcpTools, resolveContextLength, resolveCredentialEndpoints, sanitizeHistoryForNextTurn, 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, type AffinityRunningModel, 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 CompactionOptions, type CompareDraftPerformanceOptions, type ContextCheck, type ContextCheckOptions, type ContextDiscoveryRequestOptions, type ContextLengthSignals, type ContextLengthSource, type ContextWindowSource, type ConversationSendOptions, ConversationSession, type ConversationSessionOptions, CopyRequestOptions, type CreateModelFromGgufOptions, 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 DiscoveredContextLength, type DraftBenchmarkClient, type DraftPerformanceComparison, type DraftRunStats, type EmbedBatchOptions, type EmbedBatchResult, EmbedRequestOptions, EmbedResponse, EmbeddingsRequestOptions, EmbeddingsResponse, type EndpointHealth, type EndpointHealthCheckResult, EndpointRegistry, type EndpointRegistryOptions, FetchLike, type FetchWithBackoffConfig, FormatOption, type FormattedTelemetry, GEN_AI_SYSTEM_OLLAMA, GenerateRequestOptions, GenerateResponse, GenerateStreamResult, type HistorySanitizeOptions, 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, MESSAGE_OVERHEAD_TOKENS, McpBridge, type McpBridgeOptions, McpCallToolResult, McpClientLike, McpTask, McpToolDescriptor, Message, Middleware, type ModelAffinityClient, ModelAffinityRouter, type ModelAffinityRoutingOptions, ModelAffinityScheduler, type ModelAffinitySchedulerOptions, type ModelAffinitySnapshot, type ModelAffinityStats, 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, PingResult, type ProgressCallback, ProgressResponse, type ProgressSpeed, ProgressStreamResult, PsResponse, type PullProgressEvent, PullRequestOptions, PushRequestOptions, QuotaManager, type QuotaManagerOptions, type QuotaWindowConfig, type QuotaWindowStatus, RETRYABLE_STATUS_CODES, type RankResult, type RawOllamaMetrics, RawUsageSource, RequestCancellationOptions, RequestLifecycleHook, RequestRunner, type ResponsesCreateRequest, type ResponsesCreateResponse, ResponsesModule, type ResponsesReasoningEffort, type ResponsesStreamEvent, type ResponsesToolDefinition, type ResponsesUsage, type RetryConfig, type RuntimeMode, type ScoreDecision, type SessionCompactionOptions, type SessionCompactionResult, type SessionTurn, ShowRequestOptions, ShowResponse, type SpanAttributeValue, type SpanAttributes, StatusResponse, type StreamObserverCallbacks, type StreamObserverPhase, 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, WEB_FETCH_TOOL_NAME, WEB_SEARCH_TOOL_NAME, WebFetchRequestOptions, WebFetchResponse, type WebFetchToolParams, WebSearchRequestOptions, WebSearchResponse, type WebSearchToolParams, type WebToolsClient, type WebToolsOptions, addSpanEvent, batchEmbed, calculateBackoff, canonicalToolCallSignature, checkChatContext, checkEndpointHealth, checkGenerateContext, compactConversationHistory, compareDraftPerformance, computeProgressPercent, contextWarningMessage, createDecision, createModelAffinityScheduler, createOllamaCloudFreeTierQuota, createProgressSpeedTracker, createTimeoutSignal, createWebFetchTool, createWebSearchTool, createWebTools, defineTool, detectModelCapabilities, embedBatchOverflowMessage, encodeImage, estimateChatRequestTokens, estimateGenerateRequestTokens, estimateMessageTokens, estimateSystemOneRequestBytes, estimateTokens, extractContextLength as extractNativeContextLength, extractParameterNumCtx, fetchWithBackoff, findOversizedEmbedInputs, findRunningModelContextLength, formatTelemetry, formatWebFetchResult, formatWebSearchResult, inferRuntimeMode, isKeepAliveSugar, isKnownDoneReason, isModelResident, isSameToolCall, listAllMcpTools, listAvailableModels, loadMcpTools, mergeToolCall, mergeToolCallArgumentsString, mergeToolCallArrays, normalizeChatStream, normalizeGenerateStream, normalizeKeepAlive, normalizeProgressStream, observeStream, onProgress, parseNdjsonStream, parseStructuredOutput, recommendContextWindow, registerMcpTools, resolveContextLength, resolveCredentialEndpoints, sanitizeHistoryForNextTurn, setSpanError, toDataStream, toPullProgressEvent, toResponse, toTextStream, withRetry, withSpan, zodToJsonSchema };