@nemesis-oss/ollama-sdk 1.17.0 → 1.19.0

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 });
   }
@@ -4177,7 +4185,7 @@ var SystemOneNoulQuestionSchema = z.object({
 var SystemOneScoreQuestionSchema = z.object({
   type: z.enum(["score"]),
   instructions: SystemOneContentSchema,
-  criteria: z.array(z.string()).min(2).max(26)
+  criteria: z.array(z.string()).min(2).max(10)
 });
 
 // src/generated/models/SystemOneRequest.schema.ts
@@ -4212,7 +4220,7 @@ var SystemOneNoulAnswerSchema = z.object({
 });
 var SystemOneScoreAnswerSchema = z.object({
   type: z.enum(["score"]),
-  score: z.number().min(0).max(25),
+  score: z.number().min(0).max(9),
   legend: z.object({}).catchall(z.string()),
   probabilities: SystemOneProbabilitiesSchema,
   confidence: SystemOneConfidenceSchema
@@ -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}) 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 ${applicableLimit} bytes` + (hasImages ? " (images limit)." : "."),
+          `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."
@@ -5523,9 +5551,16 @@ var FailoverHttpClient = class {
 };
 
 // src/decision.ts
+function validateThreshold(value, name) {
+  if (value === void 0) return;
+  if (typeof value !== "number" || !Number.isFinite(value) || value <= 0 || value > 1) {
+    throw new RangeError(`${name} must be a number in (0, 1] \u2014 got ${String(value)}`);
+  }
+}
 function createDecision(client) {
   return {
     async choice(params) {
+      validateThreshold(params.minConfidence, "minConfidence");
       const response = await client.systemOne({
         model: params.model,
         state: params.state,
@@ -5544,10 +5579,12 @@ function createDecision(client) {
         choice: answer.choice,
         probabilities: answer.probabilities,
         confidence: answer.confidence,
+        ...params.minConfidence !== void 0 ? { isConfident: answer.confidence >= params.minConfidence } : {},
         usage: response.usage
       };
     },
     async noul(params) {
+      validateThreshold(params.minProbability, "minProbability");
       const response = await client.systemOne({
         model: params.model,
         state: params.state,
@@ -5563,11 +5600,12 @@ function createDecision(client) {
       const answer = response.answers.decision;
       return {
         noul: answer.noul,
-        bool: answer.noul >= 0.5,
+        bool: answer.noul >= (params.minProbability ?? 0.5),
         usage: response.usage
       };
     },
     async score(params) {
+      validateThreshold(params.minConfidence, "minConfidence");
       const response = await client.systemOne({
         model: params.model,
         state: params.state,
@@ -5587,10 +5625,12 @@ function createDecision(client) {
         legend: answer.legend,
         probabilities: answer.probabilities,
         confidence: answer.confidence,
+        ...params.minConfidence !== void 0 ? { isConfident: answer.confidence >= params.minConfidence } : {},
         usage: response.usage
       };
     },
     async route(params) {
+      validateThreshold(params.minConfidence, "minConfidence");
       const response = await client.systemOne({
         model: params.model,
         state: params.state,
@@ -5607,10 +5647,12 @@ function createDecision(client) {
       const answer = response.answers.decision;
       return {
         route: answer.choice,
-        confidence: answer.confidence
+        confidence: answer.confidence,
+        ...params.minConfidence !== void 0 ? { isConfident: answer.confidence >= params.minConfidence } : {}
       };
     },
     async verify(params) {
+      validateThreshold(params.minProbability, "minProbability");
       const state = {
         claim: params.claim,
         evidence: params.evidence
@@ -5629,11 +5671,12 @@ function createDecision(client) {
       });
       const answer = response.answers.decision;
       return {
-        verified: answer.noul >= 0.5,
+        verified: answer.noul >= (params.minProbability ?? 0.5),
         probability: answer.noul
       };
     },
     async rank(params) {
+      validateThreshold(params.minConfidence, "minConfidence");
       const promises = params.candidates.map(async (candidate) => {
         const response = await client.systemOne({
           model: params.model,
@@ -5652,7 +5695,8 @@ function createDecision(client) {
         return {
           candidate: candidate.id,
           score: answer.score,
-          confidence: answer.confidence
+          confidence: answer.confidence,
+          ...params.minConfidence !== void 0 ? { isConfident: answer.confidence >= params.minConfidence } : {}
         };
       });
       const results = await Promise.all(promises);
@@ -6283,11 +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:
@@ -6784,6 +6832,64 @@ function toResponse(stream, init = {}) {
   return new Response(body, { ...init, headers });
 }
 
+// src/streaming/progress.ts
+function computeProgressPercent(chunk) {
+  if (typeof chunk.total !== "number" || typeof chunk.completed !== "number") return void 0;
+  if (chunk.total <= 0) return void 0;
+  const ratio = chunk.completed / chunk.total;
+  const clamped = Math.max(0, Math.min(1, ratio));
+  return Math.round(clamped * 1e4) / 100;
+}
+function toPullProgressEvent(chunk) {
+  return {
+    status: chunk.status,
+    ...chunk.digest !== void 0 ? { digest: chunk.digest } : {},
+    ...chunk.total !== void 0 ? { total: chunk.total } : {},
+    ...chunk.completed !== void 0 ? { completed: chunk.completed } : {},
+    percent: computeProgressPercent(chunk),
+    raw: chunk
+  };
+}
+function createProgressSpeedTracker() {
+  let epochDigest;
+  let epochStartMs = 0;
+  let epochStartCompleted = 0;
+  return (chunk) => {
+    if (typeof chunk.completed !== "number") return {};
+    const digestChanged = chunk.digest !== void 0 && chunk.digest !== epochDigest;
+    const regressed = chunk.completed < epochStartCompleted;
+    if (digestChanged || regressed) {
+      epochDigest = chunk.digest;
+      epochStartMs = performance.now();
+      epochStartCompleted = chunk.completed;
+      return {};
+    }
+    const elapsedSeconds = (performance.now() - epochStartMs) / 1e3;
+    if (elapsedSeconds <= 0) return {};
+    const transferred = chunk.completed - epochStartCompleted;
+    if (transferred <= 0) return {};
+    const bytesPerSecond = transferred / elapsedSeconds;
+    const etaSeconds = typeof chunk.total === "number" ? Math.max(0, (chunk.total - chunk.completed) / bytesPerSecond) : void 0;
+    return { bytesPerSecond, etaSeconds };
+  };
+}
+async function onProgress(stream, callback) {
+  void stream.finalResult.catch(() => void 0);
+  const speed = createProgressSpeedTracker();
+  for await (const event of stream) {
+    if (event.type === "message") {
+      callback({ ...toPullProgressEvent(event.data.chunk), ...speed(event.data.chunk) });
+    }
+    if (event.type === "done") {
+      return event.data.result;
+    }
+    if (event.type === "error") {
+      throw event.data.error;
+    }
+  }
+  throw new OllamaStreamError("Pull/push stream ended before reporting success");
+}
+
 // src/streaming/observe.ts
 async function* observeStream(source, callbacks = {}) {
   const { onChunk, onError, onDone, onAbandon, onListenerError } = callbacks;
@@ -8180,10 +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, contextWarningMessage, createConsoleLogger, createDecision, createModelAffinityScheduler, createOllamaCloudFreeTierQuota, 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, parseFrontmatter, parseNdjsonStream, parseSseStream, parseStructuredOutput, recommendContextWindow, registerMcpTools, resolveContextLength, resolveCredentialEndpoints, resolveImageInput, 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, 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