@steipete/summarize 0.8.2 → 0.10.0

package/dist/esm/llm/generate-text.js

@@ -1,265 +1,169 @@
-import { completeSimple, getModel, streamSimple } from '@mariozechner/pi-ai';
+import { completeSimple, streamSimple } from '@mariozechner/pi-ai';
+import { createUnsupportedFunctionalityError } from './errors.js';
 import { parseGatewayStyleModelId } from './model-id.js';
-function parseAnthropicErrorPayload(responseBody) {
-    try {
-        const parsed = JSON.parse(responseBody);
-        if (parsed?.type !== 'error')
-            return null;
-        const error = parsed.error;
-        if (!error || typeof error !== 'object')
-            return null;
-        const errorType = typeof error.type === 'string' ? error.type : null;
-        const errorMessage = typeof error.message === 'string' ? error.message : null;
-        if (!errorType || !errorMessage)
-            return null;
-        return { type: errorType, message: errorMessage };
+import { userTextAndImageMessage } from './prompt.js';
+import { completeAnthropicDocument, completeAnthropicText, normalizeAnthropicModelAccessError, } from './providers/anthropic.js';
+import { completeGoogleDocument, completeGoogleText } from './providers/google.js';
+import { resolveAnthropicModel, resolveGoogleModel, resolveOpenAiModel, resolveXaiModel, resolveZaiModel, } from './providers/models.js';
+import { completeOpenAiDocument, completeOpenAiText, resolveOpenAiClientConfig, } from './providers/openai.js';
+import { extractText } from './providers/shared.js';
+import { normalizeTokenUsage } from './usage.js';
+function promptToContext(prompt) {
+    const attachments = prompt.attachments ?? [];
+    if (attachments.some((attachment) => attachment.kind === 'document')) {
+        throw new Error('Internal error: document prompt cannot be converted to context.');
     }
-    catch {
-        return null;
+    if (attachments.length === 0) {
+        return {
+            systemPrompt: prompt.system,
+            messages: [{ role: 'user', content: prompt.userText, timestamp: Date.now() }],
+        };
     }
-}
-function normalizeAnthropicModelAccessError(error, modelId) {
-    if (!error || typeof error !== 'object')
-        return null;
-    const maybe = error;
-    const statusCode = typeof maybe.statusCode === 'number' ? maybe.statusCode : null;
-    const responseBody = typeof maybe.responseBody === 'string' ? maybe.responseBody : null;
-    const payload = responseBody ? parseAnthropicErrorPayload(responseBody) : null;
-    const payloadType = payload?.type ?? null;
-    const payloadMessage = payload?.message ?? null;
-    const message = typeof maybe.message === 'string' ? maybe.message : '';
-    const combinedMessage = (payloadMessage ?? message).trim();
-    const hasModelMessage = /^model:\s*\S+/i.test(combinedMessage);
-    const isAccessStatus = statusCode === 401 || statusCode === 403 || statusCode === 404;
-    const isAccessType = payloadType === 'not_found_error' ||
-        payloadType === 'permission_error' ||
-        payloadType === 'authentication_error';
-    if (!hasModelMessage && !isAccessStatus && !isAccessType)
-        return null;
-    const modelLabel = hasModelMessage ? combinedMessage.replace(/^model:\s*/i, '').trim() : modelId;
-    const hint = `Anthropic API rejected model "${modelLabel}". Your ANTHROPIC_API_KEY likely lacks access to this model or it is unavailable for your account. Try another anthropic/... model or request access.`;
-    return new Error(hint, { cause: error instanceof Error ? error : undefined });
-}
-function normalizeTokenUsage(raw) {
-    if (!raw || typeof raw !== 'object')
-        return null;
-    const usage = raw;
-    const promptTokens = typeof usage.input === 'number' && Number.isFinite(usage.input) ? usage.input : null;
-    const completionTokens = typeof usage.output === 'number' && Number.isFinite(usage.output) ? usage.output : null;
-    const totalTokens = typeof usage.totalTokens === 'number' && Number.isFinite(usage.totalTokens)
-        ? usage.totalTokens
-        : null;
-    if (promptTokens === null && completionTokens === null && totalTokens === null)
-        return null;
-    return { promptTokens, completionTokens, totalTokens };
-}
-function resolveBaseUrlOverride(raw) {
-    const trimmed = typeof raw === 'string' ? raw.trim() : '';
-    return trimmed.length > 0 ? trimmed : null;
-}
-function resolveOpenAiClientConfig({ apiKeys, forceOpenRouter, openaiBaseUrlOverride, forceChatCompletions, }) {
-    const baseUrlRaw = openaiBaseUrlOverride ??
-        (typeof process !== 'undefined' ? process.env.OPENAI_BASE_URL : undefined);
-    const baseUrl = typeof baseUrlRaw === 'string' && baseUrlRaw.trim().length > 0 ? baseUrlRaw.trim() : null;
-    const isOpenRouterViaBaseUrl = baseUrl ? /openrouter\.ai/i.test(baseUrl) : false;
-    const hasOpenRouterKey = apiKeys.openrouterApiKey != null;
-    const hasOpenAiKey = apiKeys.openaiApiKey != null;
-    const isOpenRouter = Boolean(forceOpenRouter) ||
-        isOpenRouterViaBaseUrl ||
-        (hasOpenRouterKey && !baseUrl && !hasOpenAiKey);
-    const apiKey = isOpenRouter
-        ? (apiKeys.openrouterApiKey ?? apiKeys.openaiApiKey)
-        : apiKeys.openaiApiKey;
-    if (!apiKey) {
-        throw new Error(isOpenRouter
-            ? 'Missing OPENROUTER_API_KEY (or OPENAI_API_KEY) for OpenRouter'
-            : 'Missing OPENAI_API_KEY for openai/... model');
+    if (attachments.length !== 1 || attachments[0]?.kind !== 'image') {
+        throw new Error('Internal error: only single image attachments are supported for prompts.');
     }
-    const baseURL = forceOpenRouter
-        ? 'https://openrouter.ai/api/v1'
-        : (baseUrl ?? (isOpenRouter ? 'https://openrouter.ai/api/v1' : undefined));
-    const isCustomBaseURL = (() => {
-        if (!baseURL)
-            return false;
-        try {
-            const url = new URL(baseURL);
-            return url.host !== 'api.openai.com' && url.host !== 'openrouter.ai';
-        }
-        catch {
-            return false;
-        }
-    })();
-    const useChatCompletions = Boolean(forceChatCompletions) || isOpenRouter || isCustomBaseURL;
-    return {
-        apiKey,
-        baseURL: baseURL ?? undefined,
-        useChatCompletions,
-        isOpenRouter,
-    };
+    const attachment = attachments[0];
+    const messages = [
+        userTextAndImageMessage({
+            text: prompt.userText,
+            imageBytes: attachment.bytes,
+            mimeType: attachment.mediaType,
+        }),
+    ];
+    return { systemPrompt: prompt.system, messages };
 }
-function promptToContext({ system, prompt }) {
-    const messages = typeof prompt === 'string'
-        ? [{ role: 'user', content: prompt, timestamp: Date.now() }]
-        : prompt.map((msg) => typeof msg.timestamp === 'number'
-            ? msg
-            : { ...msg, timestamp: Date.now() });
-    return { systemPrompt: system, messages };
+function isRetryableTimeoutError(error) {
+    if (!error)
+        return false;
+    const message = typeof error === 'string'
+        ? error
+        : error instanceof Error
+            ? error.message
+            : typeof error.message === 'string'
+                ? String(error.message)
+                : '';
+    return /timed out/i.test(message) || /empty summary/i.test(message);
 }
-function extractText(message) {
-    const text = message.content
-        .filter((c) => c.type === 'text')
-        .map((c) => c.text)
-        .join('');
-    return text.trim();
+function computeRetryDelayMs(attempt) {
+    const base = 500;
+    const jitter = Math.floor(Math.random() * 200);
+    return Math.min(2000, base * (attempt + 1) + jitter);
 }
-function wantsImages(context) {
-    for (const msg of context.messages) {
-        if (msg.role === 'user' || msg.role === 'toolResult') {
-            if (Array.isArray(msg.content) && msg.content.some((c) => c.type === 'image'))
-                return true;
-        }
-    }
-    return false;
+function sleep(ms) {
+    return new Promise((resolve) => setTimeout(resolve, ms));
 }
-function tryGetModel(provider, modelId) {
-    try {
-        return getModel(provider, modelId);
-    }
-    catch {
-        return null;
-    }
+function isOpenaiGpt5Model(parsed) {
+    return parsed.provider === 'openai' && /^gpt-5([-.].+)?$/i.test(parsed.model);
 }
-function createSyntheticModel({ provider, modelId, api, baseUrl, allowImages, headers, }) {
-    return {
-        id: modelId,
-        name: `${provider}/${modelId}`,
-        api,
-        provider,
-        baseUrl,
-        reasoning: false,
-        input: allowImages ? ['text', 'image'] : ['text'],
-        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
-        contextWindow: 128_000,
-        maxTokens: 16_384,
-        ...(headers ? { headers } : {}),
-    };
+function resolveEffectiveTemperature({ parsed, temperature, }) {
+    if (typeof temperature !== 'number')
+        return undefined;
+    if (isOpenaiGpt5Model(parsed))
+        return undefined;
+    return temperature;
 }
-function resolveModelForCall({ modelId, parsedProvider, openaiConfig, context, openaiBaseUrlOverride, anthropicBaseUrlOverride, googleBaseUrlOverride, xaiBaseUrlOverride, }) {
-    const allowImages = wantsImages(context);
-    if (parsedProvider === 'openai') {
-        const base = tryGetModel('openai', modelId);
-        const api = openaiConfig?.useChatCompletions ? 'openai-completions' : 'openai-responses';
-        const baseUrl = openaiConfig?.baseURL ?? base?.baseUrl ?? 'https://api.openai.com/v1';
-        const headers = openaiConfig?.isOpenRouter
-            ? {
-                ...(base?.headers ?? {}),
-                'HTTP-Referer': 'https://github.com/steipete/summarize',
-                'X-Title': 'summarize',
+export async function generateTextWithModelId({ modelId, apiKeys, prompt, temperature, maxOutputTokens, timeoutMs, fetchImpl, forceOpenRouter, openaiBaseUrlOverride, anthropicBaseUrlOverride, googleBaseUrlOverride, xaiBaseUrlOverride, forceChatCompletions, retries = 0, onRetry, }) {
+    const parsed = parseGatewayStyleModelId(modelId);
+    const effectiveTemperature = resolveEffectiveTemperature({ parsed, temperature });
+    const attachments = prompt.attachments ?? [];
+    const documentAttachment = attachments.find((attachment) => attachment.kind === 'document') ?? null;
+    if (documentAttachment) {
+        if (attachments.length !== 1) {
+            throw new Error('Internal error: document attachments cannot be combined with other inputs.');
+        }
+        if (parsed.provider === 'anthropic') {
+            const apiKey = apiKeys.anthropicApiKey;
+            if (!apiKey)
+                throw new Error('Missing ANTHROPIC_API_KEY for anthropic/... model');
+            try {
+                const result = await completeAnthropicDocument({
+                    modelId: parsed.model,
+                    apiKey,
+                    promptText: prompt.userText,
+                    document: documentAttachment,
+                    system: prompt.system,
+                    maxOutputTokens,
+                    timeoutMs,
+                    fetchImpl,
+                    anthropicBaseUrlOverride,
+                });
+                return {
+                    text: result.text,
+                    canonicalModelId: parsed.canonical,
+                    provider: parsed.provider,
+                    usage: result.usage,
+                };
             }
-            : base?.headers;
-        return {
-            ...(base ?? createSyntheticModel({ provider: 'openai', modelId, api, baseUrl, allowImages })),
-            api,
-            baseUrl,
-            ...(headers ? { headers } : {}),
-        };
-    }
-    if (parsedProvider === 'zai') {
-        const base = tryGetModel('zai', modelId);
-        const api = 'openai-completions';
-        const baseUrl = openaiBaseUrlOverride ??
-            base?.baseUrl ??
-            openaiConfig?.baseURL ??
-            'https://api.z.ai/api/paas/v4';
-        return {
-            ...(base ?? createSyntheticModel({ provider: 'zai', modelId, api, baseUrl, allowImages })),
-            api,
-            baseUrl,
-            input: allowImages ? ['text', 'image'] : ['text'],
-        };
-    }
-    if (parsedProvider === 'xai') {
-        const base = tryGetModel('xai', modelId);
-        const override = resolveBaseUrlOverride(xaiBaseUrlOverride);
-        if (override) {
+            catch (error) {
+                const normalized = normalizeAnthropicModelAccessError(error, parsed.model);
+                if (normalized)
+                    throw normalized;
+                throw error;
+            }
+        }
+        if (parsed.provider === 'openai') {
+            const openaiConfig = resolveOpenAiClientConfig({
+                apiKeys: {
+                    openaiApiKey: apiKeys.openaiApiKey,
+                    openrouterApiKey: apiKeys.openrouterApiKey,
+                },
+                forceOpenRouter,
+                openaiBaseUrlOverride,
+                forceChatCompletions,
+            });
+            const result = await completeOpenAiDocument({
+                modelId: parsed.model,
+                openaiConfig,
+                promptText: prompt.userText,
+                document: documentAttachment,
+                maxOutputTokens,
+                temperature: effectiveTemperature,
+                timeoutMs,
+                fetchImpl,
+            });
             return {
-                ...(base ??
-                    createSyntheticModel({
-                        provider: 'xai',
-                        modelId,
-                        api: 'openai-completions',
-                        baseUrl: override,
-                        allowImages,
-                    })),
-                baseUrl: override,
+                text: result.text,
+                canonicalModelId: parsed.canonical,
+                provider: parsed.provider,
+                usage: result.usage,
             };
         }
-        return (base ??
-            createSyntheticModel({
-                provider: 'xai',
-                modelId,
-                api: 'openai-completions',
-                baseUrl: 'https://api.x.ai/v1',
-                allowImages,
-            }));
-    }
-    if (parsedProvider === 'google') {
-        const base = tryGetModel('google', modelId);
-        const override = resolveBaseUrlOverride(googleBaseUrlOverride);
-        if (override) {
+        if (parsed.provider === 'google') {
+            const apiKey = apiKeys.googleApiKey;
+            if (!apiKey)
+                throw new Error('Missing GEMINI_API_KEY (or GOOGLE_GENERATIVE_AI_API_KEY / GOOGLE_API_KEY) for google/... model');
+            const result = await completeGoogleDocument({
+                modelId: parsed.model,
+                apiKey,
+                promptText: prompt.userText,
+                document: documentAttachment,
+                maxOutputTokens,
+                temperature: effectiveTemperature,
+                timeoutMs,
+                fetchImpl,
+                googleBaseUrlOverride,
+            });
             return {
-                ...(base ??
-                    createSyntheticModel({
-                        provider: 'google',
-                        modelId,
-                        api: 'google-generative-ai',
-                        baseUrl: override,
-                        allowImages,
-                    })),
-                baseUrl: override,
+                text: result.text,
+                canonicalModelId: parsed.canonical,
+                provider: parsed.provider,
+                usage: result.usage,
             };
         }
-        return (base ??
-            createSyntheticModel({
-                provider: 'google',
-                modelId,
-                api: 'google-generative-ai',
-                baseUrl: 'https://generativelanguage.googleapis.com/v1beta',
-                allowImages,
-            }));
+        throw createUnsupportedFunctionalityError(`document attachments are not supported for ${parsed.provider}/... models`);
     }
-    const base = tryGetModel('anthropic', modelId);
-    const override = resolveBaseUrlOverride(anthropicBaseUrlOverride);
-    if (override) {
-        return {
-            ...(base ??
-                createSyntheticModel({
-                    provider: 'anthropic',
-                    modelId,
-                    api: 'anthropic-messages',
-                    baseUrl: override,
-                    allowImages,
-                })),
-            baseUrl: override,
-        };
-    }
-    return (base ??
-        createSyntheticModel({
-            provider: 'anthropic',
-            modelId,
-            api: 'anthropic-messages',
-            baseUrl: 'https://api.anthropic.com',
-            allowImages,
-        }));
-}
-export async function generateTextWithModelId({ modelId, apiKeys, system, prompt, temperature, maxOutputTokens, timeoutMs, fetchImpl: _fetchImpl, forceOpenRouter, openaiBaseUrlOverride, anthropicBaseUrlOverride, googleBaseUrlOverride, xaiBaseUrlOverride, forceChatCompletions, retries = 0, onRetry, }) {
-    void _fetchImpl;
-    const parsed = parseGatewayStyleModelId(modelId);
-    const context = promptToContext({ system, prompt });
-    const isOpenaiGpt5 = parsed.provider === 'openai' && /^gpt-5([-.].+)?$/i.test(parsed.model);
-    const effectiveTemperature = typeof temperature === 'number' && !(isOpenaiGpt5 && temperature === 0)
-        ? temperature
-        : undefined;
+    const context = promptToContext(prompt);
+    const openaiConfig = parsed.provider === 'openai'
+        ? resolveOpenAiClientConfig({
+            apiKeys: {
+                openaiApiKey: apiKeys.openaiApiKey,
+                openrouterApiKey: apiKeys.openrouterApiKey,
+            },
+            forceOpenRouter,
+            openaiBaseUrlOverride,
+            forceChatCompletions,
+        })
+        : null;
     const maxRetries = Math.max(0, retries);
     let attempt = 0;
     while (attempt <= maxRetries) {
@@ -270,10 +174,8 @@ export async function generateTextWithModelId({ modelId, apiKeys, system, prompt
                 const apiKey = apiKeys.xaiApiKey;
                 if (!apiKey)
                     throw new Error('Missing XAI_API_KEY for xai/... model');
-                const model = resolveModelForCall({
+                const model = resolveXaiModel({
                     modelId: parsed.model,
-                    parsedProvider: parsed.provider,
-                    openaiConfig: null,
                     context,
                     xaiBaseUrlOverride,
                 });
@@ -299,76 +201,48 @@ export async function generateTextWithModelId({ modelId, apiKeys, system, prompt
                 const apiKey = apiKeys.googleApiKey;
                 if (!apiKey)
                     throw new Error('Missing GEMINI_API_KEY (or GOOGLE_GENERATIVE_AI_API_KEY / GOOGLE_API_KEY) for google/... model');
-                const model = resolveModelForCall({
+                const result = await completeGoogleText({
                     modelId: parsed.model,
-                    parsedProvider: parsed.provider,
-                    openaiConfig: null,
-                    context,
-                    googleBaseUrlOverride,
-                });
-                const result = await completeSimple(model, context, {
-                    ...(typeof effectiveTemperature === 'number'
-                        ? { temperature: effectiveTemperature }
-                        : {}),
-                    ...(typeof maxOutputTokens === 'number' ? { maxTokens: maxOutputTokens } : {}),
                     apiKey,
+                    context,
+                    temperature: effectiveTemperature,
+                    maxOutputTokens,
                     signal: controller.signal,
+                    googleBaseUrlOverride,
                 });
-                const text = extractText(result);
-                if (!text)
-                    throw new Error(`LLM returned an empty summary (model ${parsed.canonical}).`);
                 return {
-                    text,
+                    text: result.text,
                     canonicalModelId: parsed.canonical,
                     provider: parsed.provider,
-                    usage: normalizeTokenUsage(result.usage),
+                    usage: result.usage,
                 };
             }
             if (parsed.provider === 'anthropic') {
                 const apiKey = apiKeys.anthropicApiKey;
                 if (!apiKey)
                     throw new Error('Missing ANTHROPIC_API_KEY for anthropic/... model');
-                const model = resolveModelForCall({
+                const result = await completeAnthropicText({
                     modelId: parsed.model,
-                    parsedProvider: parsed.provider,
-                    openaiConfig: null,
-                    context,
-                    anthropicBaseUrlOverride,
-                });
-                const result = await completeSimple(model, context, {
-                    ...(typeof effectiveTemperature === 'number'
-                        ? { temperature: effectiveTemperature }
-                        : {}),
-                    ...(typeof maxOutputTokens === 'number' ? { maxTokens: maxOutputTokens } : {}),
                     apiKey,
+                    context,
+                    temperature: effectiveTemperature,
+                    maxOutputTokens,
                     signal: controller.signal,
+                    anthropicBaseUrlOverride,
                 });
-                const text = extractText(result);
-                if (!text)
-                    throw new Error(`LLM returned an empty summary (model ${parsed.canonical}).`);
                 return {
-                    text,
+                    text: result.text,
                     canonicalModelId: parsed.canonical,
                     provider: parsed.provider,
-                    usage: normalizeTokenUsage(result.usage),
+                    usage: result.usage,
                 };
             }
-            const openaiConfig = parsed.provider === 'openai'
-                ? resolveOpenAiClientConfig({
-                    apiKeys,
-                    forceOpenRouter,
-                    openaiBaseUrlOverride,
-                    forceChatCompletions,
-                })
-                : null;
             if (parsed.provider === 'zai') {
                 const apiKey = apiKeys.openaiApiKey;
                 if (!apiKey)
                     throw new Error('Missing Z_AI_API_KEY for zai/... model');
-                const model = resolveModelForCall({
+                const model = resolveZaiModel({
                     modelId: parsed.model,
-                    parsedProvider: parsed.provider,
-                    openaiConfig: null,
                     context,
                     openaiBaseUrlOverride,
                 });
@@ -390,30 +264,22 @@ export async function generateTextWithModelId({ modelId, apiKeys, system, prompt
                     usage: normalizeTokenUsage(result.usage),
                 };
             }
-            const model = resolveModelForCall({
+            if (!openaiConfig) {
+                throw new Error('Missing OPENAI_API_KEY for openai/... model');
+            }
+            const result = await completeOpenAiText({
                 modelId: parsed.model,
-                parsedProvider: parsed.provider,
                 openaiConfig,
                 context,
-                openaiBaseUrlOverride,
-                anthropicBaseUrlOverride,
-                googleBaseUrlOverride,
-                xaiBaseUrlOverride,
-            });
-            const result = await completeSimple(model, context, {
-                ...(typeof effectiveTemperature === 'number' ? { temperature: effectiveTemperature } : {}),
-                ...(typeof maxOutputTokens === 'number' ? { maxTokens: maxOutputTokens } : {}),
-                apiKey: openaiConfig?.apiKey ?? apiKeys.openaiApiKey ?? undefined,
+                temperature: effectiveTemperature,
+                maxOutputTokens,
                 signal: controller.signal,
             });
-            const text = extractText(result);
-            if (!text)
-                throw new Error(`LLM returned an empty summary (model ${parsed.canonical}).`);
             return {
-                text,
+                text: result.text,
                 canonicalModelId: parsed.canonical,
                 provider: parsed.provider,
-                usage: normalizeTokenUsage(result.usage),
+                usage: result.usage,
             };
         }
         catch (error) {
@@ -440,30 +306,28 @@ export async function generateTextWithModelId({ modelId, apiKeys, system, prompt
     }
     throw new Error(`LLM request failed after ${maxRetries + 1} attempts.`);
 }
-function isRetryableTimeoutError(error) {
-    if (!error)
-        return false;
-    const message = typeof error === 'string'
-        ? error
-        : error instanceof Error
-            ? error.message
-            : typeof error.message === 'string'
-                ? String(error.message)
-                : '';
-    return /timed out/i.test(message) || /empty summary/i.test(message);
-}
-function computeRetryDelayMs(attempt) {
-    const base = 500;
-    const jitter = Math.floor(Math.random() * 200);
-    return Math.min(2000, base * (attempt + 1) + jitter);
-}
-function sleep(ms) {
-    return new Promise((resolve) => setTimeout(resolve, ms));
+export async function streamTextWithModelId({ modelId, apiKeys, prompt, temperature, maxOutputTokens, timeoutMs, fetchImpl, forceOpenRouter, openaiBaseUrlOverride, anthropicBaseUrlOverride, googleBaseUrlOverride, xaiBaseUrlOverride, forceChatCompletions, }) {
+    const context = promptToContext(prompt);
+    return streamTextWithContext({
+        modelId,
+        apiKeys,
+        context,
+        temperature,
+        maxOutputTokens,
+        timeoutMs,
+        fetchImpl,
+        forceOpenRouter,
+        openaiBaseUrlOverride,
+        anthropicBaseUrlOverride,
+        googleBaseUrlOverride,
+        xaiBaseUrlOverride,
+        forceChatCompletions,
+    });
 }
-export async function streamTextWithModelId({ modelId, apiKeys, system, prompt, temperature, maxOutputTokens, timeoutMs, fetchImpl: _fetchImpl, forceOpenRouter, openaiBaseUrlOverride, anthropicBaseUrlOverride, googleBaseUrlOverride, xaiBaseUrlOverride, forceChatCompletions, }) {
-    void _fetchImpl;
+export async function streamTextWithContext({ modelId, apiKeys, context, temperature, maxOutputTokens, timeoutMs, fetchImpl, forceOpenRouter, openaiBaseUrlOverride, anthropicBaseUrlOverride, googleBaseUrlOverride, xaiBaseUrlOverride, forceChatCompletions, }) {
     const parsed = parseGatewayStyleModelId(modelId);
-    const context = promptToContext({ system, prompt });
+    const effectiveTemperature = resolveEffectiveTemperature({ parsed, temperature });
+    void fetchImpl;
     const controller = new AbortController();
     let timeoutId = null;
     const startedAtMs = Date.now();
@@ -539,15 +403,13 @@ export async function streamTextWithModelId({ modelId, apiKeys, system, prompt,
             const apiKey = apiKeys.xaiApiKey;
             if (!apiKey)
                 throw new Error('Missing XAI_API_KEY for xai/... model');
-            const model = resolveModelForCall({
+            const model = resolveXaiModel({
                 modelId: parsed.model,
-                parsedProvider: parsed.provider,
-                openaiConfig: null,
                 context,
                 xaiBaseUrlOverride,
             });
             const stream = streamSimple(model, context, {
-                ...(typeof temperature === 'number' ? { temperature } : {}),
+                ...(typeof effectiveTemperature === 'number' ? { temperature: effectiveTemperature } : {}),
                 ...(typeof maxOutputTokens === 'number' ? { maxTokens: maxOutputTokens } : {}),
                 apiKey,
                 signal: controller.signal,
@@ -579,15 +441,13 @@ export async function streamTextWithModelId({ modelId, apiKeys, system, prompt,
             const apiKey = apiKeys.googleApiKey;
             if (!apiKey)
                 throw new Error('Missing GEMINI_API_KEY (or GOOGLE_GENERATIVE_AI_API_KEY / GOOGLE_API_KEY) for google/... model');
-            const model = resolveModelForCall({
+            const model = resolveGoogleModel({
                 modelId: parsed.model,
-                parsedProvider: parsed.provider,
-                openaiConfig: null,
                 context,
                 googleBaseUrlOverride,
             });
             const stream = streamSimple(model, context, {
-                ...(typeof temperature === 'number' ? { temperature } : {}),
+                ...(typeof effectiveTemperature === 'number' ? { temperature: effectiveTemperature } : {}),
                 ...(typeof maxOutputTokens === 'number' ? { maxTokens: maxOutputTokens } : {}),
                 apiKey,
                 signal: controller.signal,
@@ -619,15 +479,13 @@ export async function streamTextWithModelId({ modelId, apiKeys, system, prompt,
             const apiKey = apiKeys.anthropicApiKey;
             if (!apiKey)
                 throw new Error('Missing ANTHROPIC_API_KEY for anthropic/... model');
-            const model = resolveModelForCall({
+            const model = resolveAnthropicModel({
                 modelId: parsed.model,
-                parsedProvider: parsed.provider,
-                openaiConfig: null,
                 context,
                 anthropicBaseUrlOverride,
             });
             const stream = streamSimple(model, context, {
-                ...(typeof temperature === 'number' ? { temperature } : {}),
+                ...(typeof effectiveTemperature === 'number' ? { temperature: effectiveTemperature } : {}),
                 ...(typeof maxOutputTokens === 'number' ? { maxTokens: maxOutputTokens } : {}),
                 apiKey,
                 signal: controller.signal,
@@ -660,15 +518,13 @@ export async function streamTextWithModelId({ modelId, apiKeys, system, prompt,
             const apiKey = apiKeys.openaiApiKey;
             if (!apiKey)
                 throw new Error('Missing Z_AI_API_KEY for zai/... model');
-            const model = resolveModelForCall({
+            const model = resolveZaiModel({
                 modelId: parsed.model,
-                parsedProvider: parsed.provider,
-                openaiConfig: null,
                 context,
                 openaiBaseUrlOverride,
             });
             const stream = streamSimple(model, context, {
-                ...(typeof temperature === 'number' ? { temperature } : {}),
+                ...(typeof effectiveTemperature === 'number' ? { temperature: effectiveTemperature } : {}),
                 ...(typeof maxOutputTokens === 'number' ? { maxTokens: maxOutputTokens } : {}),
                 apiKey,
                 signal: controller.signal,
@@ -697,23 +553,17 @@ export async function streamTextWithModelId({ modelId, apiKeys, system, prompt,
             };
         }
         const openaiConfig = resolveOpenAiClientConfig({
-            apiKeys,
+            apiKeys: {
+                openaiApiKey: apiKeys.openaiApiKey,
+                openrouterApiKey: apiKeys.openrouterApiKey,
+            },
             forceOpenRouter,
             openaiBaseUrlOverride,
             forceChatCompletions,
         });
-        const model = resolveModelForCall({
-            modelId: parsed.model,
-            parsedProvider: parsed.provider,
-            openaiConfig,
-            context,
-            openaiBaseUrlOverride,
-            anthropicBaseUrlOverride,
-            googleBaseUrlOverride,
-            xaiBaseUrlOverride,
-        });
+        const model = resolveOpenAiModel({ modelId: parsed.model, context, openaiConfig });
         const stream = streamSimple(model, context, {
-            ...(typeof temperature === 'number' ? { temperature } : {}),
+            ...(typeof effectiveTemperature === 'number' ? { temperature: effectiveTemperature } : {}),
             ...(typeof maxOutputTokens === 'number' ? { maxTokens: maxOutputTokens } : {}),
             apiKey: openaiConfig.apiKey,
             signal: controller.signal,