chore(x): migrate to AI SDK 7

Bump apps/x from AI SDK v5 to v7 (single hop through the 6-0 and 7-0
migration guides). All deps upgraded via pnpm --filter:
ai@7, @ai-sdk/{anthropic@4,google@4,openai@4,openai-compatible@3,provider@4},
@openrouter/ai-sdk-provider@3, ollama-ai-provider-v2@4.

Changes (per the official migration guides):
- Provider spec bumped V2->V4: ProviderV2 -> ProviderV4 (models.ts, gateway.ts)
- streamText: result.fullStream -> result.stream (+ StreamTextInvoker seam,
  test fakes); stepCountIs -> isStepCount
- system -> instructions on SDK generateText/streamText/generateObject calls
  (left our own generateObjectSafe.system param untouched)
- allowSystemInMessages: true on calls passing stored messages (v7 rejects
  system-role messages in the array by default; convertFromMessages emits them)
- image content part {type:'image'} -> {type:'file'} (message-encoding.ts)
- tool() no longer accepts a `name` key (removed; tools keyed by ToolSet key)
- usage token relocation: cachedInputTokens -> inputTokenDetails.cacheReadTokens,
  reasoningTokens -> outputTokenDetails.reasoningTokens (mapUsage rewrite);
  renderer decoupled from LanguageModelUsage to a local flat UsageSummary
- providerMetadata relay cast to shared ProviderOptions (spec loosened)

No persisted-data changes: @x/shared is AI-SDK-free, so old chats
list/view/continue unchanged. Full typecheck clean; 637 tests pass
(shared 107, core 447, renderer 83); main.cjs bundles ESM-only ai into CJS.

Pending live check (needs creds): OpenAI strictJsonSchema now defaults true
in v7 and may reject zod schemas for openai-family flavors at runtime.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

apps/x/packages/core/package.json

@@ -17,19 +17,19 @@
     "@agentclientprotocol/claude-agent-acp": "^0.55.0",
     "@agentclientprotocol/codex-acp": "^1.1.0",
     "@agentclientprotocol/sdk": "^1.1.0",
-    "@ai-sdk/anthropic": "^2.0.63",
-    "@ai-sdk/google": "^2.0.53",
-    "@ai-sdk/openai": "^2.0.91",
-    "@ai-sdk/openai-compatible": "^1.0.33",
-    "@ai-sdk/provider": "^2.0.1",
+    "@ai-sdk/anthropic": "^4.0.12",
+    "@ai-sdk/google": "^4.0.12",
+    "@ai-sdk/openai": "^4.0.11",
+    "@ai-sdk/openai-compatible": "^3.0.7",
+    "@ai-sdk/provider": "^4.0.3",
     "@composio/core": "^0.6.0",
     "@google-cloud/local-auth": "^3.0.1",
     "@modelcontextprotocol/sdk": "^1.25.1",
-    "@openrouter/ai-sdk-provider": "^1.2.6",
+    "@openrouter/ai-sdk-provider": "^3.0.0",
     "@react-pdf/renderer": "^4.3.2",
     "@types/react": "^19.2.7",
     "@x/shared": "workspace:*",
-    "ai": "^5.0.133",
+    "ai": "^7.0.22",
     "awilix": "^12.0.5",
     "baileys": "7.0.0-rc13",
     "chokidar": "^4.0.3",
@@ -42,7 +42,7 @@
     "isomorphic-git": "^1.29.0",
     "mammoth": "^1.11.0",
     "node-html-markdown": "^2.0.0",
-    "ollama-ai-provider-v2": "^1.5.4",
+    "ollama-ai-provider-v2": "^4.0.1",
     "openid-client": "^6.8.1",
     "papaparse": "^5.5.3",
     "pdf-parse": "^2.4.5",

apps/x/packages/core/src/apps/host-api.ts

@@ -304,8 +304,8 @@ async function handleLlmGenerate(
         const model = createProvider(providerConfig).languageModel(resolved.model);
         const result = await withUseCase({ useCase: 'app_llm_generate', subUseCase: slug }, () => generateText({
             model,
-            ...(system ? { system } : {}),
-            ...(rawMessages ? { messages: rawMessages as ModelMessage[] } : { prompt: prompt as string }),
+            ...(system ? { instructions: system } : {}),
+            ...(rawMessages ? { messages: rawMessages as ModelMessage[], allowSystemInMessages: true } : { prompt: prompt as string }),
             ...(temperature !== undefined ? { temperature } : {}),
             maxOutputTokens,
         }));

apps/x/packages/core/src/knowledge/inline_tasks.ts

@@ -653,7 +653,7 @@ Respond with ONLY valid JSON: either a schedule object or null. No other text.`;
     try {
         const result = await withUseCase({ useCase: 'knowledge_sync', subUseCase: 'inline_task_classify' }, () => generateText({
             model,
-            system: systemPrompt,
+            instructions: systemPrompt,
             prompt: instruction,
         }));
 

apps/x/packages/core/src/knowledge/meeting_prep_brief.ts

@@ -181,7 +181,7 @@ async function generateBrief(event: CalendarEvent, ctx: Awaited<ReturnType<typeo
 
     const result = await withUseCase({ useCase: 'meeting_prep' }, () => generateText({
         model,
-        system: BRIEF_SYSTEM,
+        instructions: BRIEF_SYSTEM,
         prompt: parts.join('\n\n'),
     }));
     captureLlmUsage({ useCase: 'meeting_prep', model: modelId, provider: providerName, usage: result.usage });

apps/x/packages/core/src/knowledge/summarize_meeting.ts

@@ -154,7 +154,7 @@ export async function summarizeMeeting(transcript: string, meetingStartTime?: st
 
     const result = await withUseCase({ useCase: 'meeting_note' }, () => generateText({
         model,
-        system: SYSTEM_PROMPT,
+        instructions: SYSTEM_PROMPT,
         prompt,
     }));
 

apps/x/packages/core/src/models/gateway.ts

@@ -1,4 +1,4 @@
-import { ProviderV2 } from '@ai-sdk/provider';
+import { ProviderV4 } from '@ai-sdk/provider';
 import { createOpenRouter } from '@openrouter/ai-sdk-provider';
 import { getAccessToken } from '../auth/tokens.js';
 import { getCurrentUseCase } from '../analytics/use_case.js';
@@ -16,7 +16,7 @@ const authedFetch: typeof fetch = async (input, init) => {
     return fetch(input, { ...init, headers });
 };
 
-export function getGatewayProvider(): ProviderV2 {
+export function getGatewayProvider(): ProviderV4 {
     return createOpenRouter({
         baseURL: `${API_URL}/v1/llm`,
         apiKey: 'managed-by-rowboat',

apps/x/packages/core/src/models/models.ts

@@ -1,4 +1,4 @@
-import { ProviderV2 } from "@ai-sdk/provider";
+import { ProviderV4 } from "@ai-sdk/provider";
 import { createGateway, generateText, type LanguageModel } from "ai";
 import { createOpenAI } from "@ai-sdk/openai";
 import { createGoogleGenerativeAI } from "@ai-sdk/google";
@@ -22,7 +22,7 @@ import {
 export const Provider = LlmProvider;
 export const ModelConfig = LlmModelConfig;
 
-export function createProvider(config: z.infer<typeof Provider>): ProviderV2 {
+export function createProvider(config: z.infer<typeof Provider>): ProviderV4 {
     const { apiKey, baseURL, headers } = config;
     switch (config.flavor) {
         case "openai":
@@ -78,7 +78,7 @@ export function createProvider(config: z.infer<typeof Provider>): ProviderV2 {
                 apiKey,
                 baseURL,
                 headers,
-            }) as unknown as ProviderV2;
+            }) as unknown as ProviderV4;
         case "rowboat":
             return getGatewayProvider();
         default:
@@ -342,7 +342,7 @@ export async function generateOneShot(opts: GenerateTextOptions): Promise<Genera
             { useCase: "copilot_chat", subUseCase: "email_compose" },
             () => generateText({
                 model: languageModel,
-                ...(opts.system ? { system: opts.system } : {}),
+                ...(opts.system ? { instructions: opts.system } : {}),
                 prompt: opts.prompt,
             }),
         );

apps/x/packages/core/src/models/structured.ts

@@ -44,7 +44,7 @@ export async function generateObjectSafe<T>(
     try {
         const result = await generateObject({
             model: options.model,
-            ...(options.system ? { system: options.system } : {}),
+            ...(options.system ? { instructions: options.system } : {}),
             prompt: options.prompt,
             schema: options.schema,
         });
@@ -68,7 +68,7 @@ export async function generateObjectSafe<T>(
             ].join("\n\n").trim();
             const result = await generateObject({
                 model: options.model,
-                system,
+                instructions: system,
                 prompt: options.prompt,
                 schema: options.schema,
             });

apps/x/packages/core/src/runtime/assembly/message-encoding.ts

@@ -101,7 +101,9 @@ export function convertFromMessages(messages: z.infer<typeof Message>[]): ModelM
                     // model knows which images show the user vs their screen.
                     const textSegments: string[] = userMessageContextPrefix ? [userMessageContextPrefix] : [];
                     const attachmentLines: string[] = [];
-                    type EncodedImagePart = { type: "image"; image: string; mediaType: string };
+                    // AI SDK 7 collapsed the `image` content part into `file`
+                    // (data + mediaType); image parts are deprecated.
+                    type EncodedImagePart = { type: "file"; data: string; mediaType: string };
                     const cameraParts: EncodedImagePart[] = [];
                     const screenParts: EncodedImagePart[] = [];
                     const frameTimes: string[] = [];
@@ -113,7 +115,7 @@ export function convertFromMessages(messages: z.infer<typeof Message>[]): ModelM
                             attachmentLines.push(`- ${part.filename} (${part.mimeType}${sizeStr}) at ${part.path}${lineStr}`);
                         } else if (part.type === "image") {
                             const target = part.source === "screen" ? screenParts : cameraParts;
-                            target.push({ type: "image", image: part.data, mediaType: part.mediaType });
+                            target.push({ type: "file", data: part.data, mediaType: part.mediaType });
                             if (part.capturedAt) frameTimes.push(part.capturedAt);
                         } else {
                             textSegments.push(part.text);

apps/x/packages/core/src/runtime/legacy/engine.ts

@@ -1,7 +1,7 @@
 import { jsonSchema } from "ai";
 import { Agent, ToolAttachment } from "@x/shared/dist/agent.js";
 import { AssistantContentPart, AssistantMessage, MessageList, ProviderOptions, ToolCallPart, ToolMessage, UserMessageContext } from "@x/shared/dist/message.js";
-import { LanguageModel, stepCountIs, streamText, tool, Tool, ToolSet } from "ai";
+import { LanguageModel, isStepCount, streamText, tool, Tool, ToolSet } from "ai";
 import { z } from "zod";
 import { LlmStepStreamEvent } from "@x/shared/dist/llm-step-events.js";
 import { execTool } from "../tools/exec-tool.js";
@@ -255,7 +255,6 @@ async function mapAgentTool(t: z.infer<typeof ToolAttachment>): Promise<Tool> {
     switch (t.type) {
         case "mcp":
             return tool({
-                name: t.name,
                 description: t.description,
                 inputSchema: jsonSchema(t.inputSchema),
             });
@@ -265,7 +264,6 @@ async function mapAgentTool(t: z.infer<typeof ToolAttachment>): Promise<Tool> {
                 throw new Error(`Agent ${t.name} not found`);
             }
             return tool({
-                name: t.name,
                 description: agent.description,
                 inputSchema: z.object({
                     message: z.string().describe("The message to send to the workflow"),
@@ -1151,21 +1149,23 @@ async function* streamLlm(
         }, () => streamText({
             model,
             messages: converted,
-            system: instructions,
+            instructions,
+            allowSystemInMessages: true,
             tools,
-            stopWhen: stepCountIs(1),
+            stopWhen: isStepCount(1),
             abortSignal: signal,
         }))
         : streamText({
             model,
             messages: converted,
-            system: instructions,
+            instructions,
+            allowSystemInMessages: true,
             tools,
-            stopWhen: stepCountIs(1),
+            stopWhen: isStepCount(1),
             abortSignal: signal,
         });
-    const { fullStream } = streamResult;
-    for await (const event of fullStream) {
+    const { stream } = streamResult;
+    for await (const event of stream) {
         // Check abort on every chunk for responsiveness
         signal?.throwIfAborted();
         console.log("-> \t\tstream event", JSON.stringify(event));
@@ -1179,39 +1179,39 @@ async function* streamLlm(
             case "reasoning-start":
                 yield {
                     type: "reasoning-start",
-                    providerOptions: event.providerMetadata,
+                    providerOptions: event.providerMetadata as z.infer<typeof ProviderOptions> | undefined,
                 };
                 break;
             case "reasoning-delta":
                 yield {
                     type: "reasoning-delta",
                     delta: event.text,
-                    providerOptions: event.providerMetadata,
+                    providerOptions: event.providerMetadata as z.infer<typeof ProviderOptions> | undefined,
                 };
                 break;
             case "reasoning-end":
                 yield {
                     type: "reasoning-end",
-                    providerOptions: event.providerMetadata,
+                    providerOptions: event.providerMetadata as z.infer<typeof ProviderOptions> | undefined,
                 };
                 break;
             case "text-start":
                 yield {
                     type: "text-start",
-                    providerOptions: event.providerMetadata,
+                    providerOptions: event.providerMetadata as z.infer<typeof ProviderOptions> | undefined,
                 };
                 break;
             case "text-end":
                 yield {
                     type: "text-end",
-                    providerOptions: event.providerMetadata,
+                    providerOptions: event.providerMetadata as z.infer<typeof ProviderOptions> | undefined,
                 };
                 break;
             case "text-delta":
                 yield {
                     type: "text-delta",
                     delta: event.text,
-                    providerOptions: event.providerMetadata,
+                    providerOptions: event.providerMetadata as z.infer<typeof ProviderOptions> | undefined,
                 };
                 break;
             case "tool-call":
@@ -1220,7 +1220,7 @@ async function* streamLlm(
                     toolCallId: event.toolCallId,
                     toolName: event.toolName,
                     input: event.input,
-                    providerOptions: event.providerMetadata,
+                    providerOptions: event.providerMetadata as z.infer<typeof ProviderOptions> | undefined,
                 };
                 break;
             case "finish-step":
@@ -1238,7 +1238,7 @@ async function* streamLlm(
                     type: "finish-step",
                     usage: event.usage,
                     finishReason: event.finishReason,
-                    providerOptions: event.providerMetadata,
+                    providerOptions: event.providerMetadata as z.infer<typeof ProviderOptions> | undefined,
                 };
                 break;
             default:

apps/x/packages/core/src/runtime/turns/bridges/real-model-registry.test.ts

@@ -15,7 +15,7 @@ function makeRegistry(parts: Array<Record<string, unknown>>, capture: InvokerOpt
         invoke: (options) => {
             capture.push(options);
             return {
-                fullStream: (async function* () {
+                stream: (async function* () {
                     yield* parts;
                 })(),
             };
@@ -378,7 +378,7 @@ describe("RealModelRegistry", () => {
                 invoke: (options) => {
                     capture.push(options);
                     return {
-                        fullStream: (async function* () {
+                        stream: (async function* () {
                             yield { type: "finish-step", finishReason: "stop", usage: {} };
                         })(),
                     };
@@ -528,7 +528,7 @@ describe("RealModelRegistry", () => {
             invoke: (options) => {
                 capture.push(options);
                 return {
-                    fullStream: (async function* () {
+                    stream: (async function* () {
                         yield { type: "finish-step", finishReason: "stop", usage: {} };
                     })(),
                 };

apps/x/packages/core/src/runtime/turns/bridges/real-model-registry.ts

@@ -1,6 +1,6 @@
 import {
     jsonSchema,
-    stepCountIs,
+    isStepCount,
     streamText,
     tool,
     type LanguageModel,
@@ -39,10 +39,18 @@ export type StreamTextInvoker = (options: {
     topP?: number;
     maxOutputTokens?: number;
     providerOptions?: Record<string, Record<string, JsonValue>>;
-}) => { fullStream: AsyncIterable<unknown> };
+}) => { stream: AsyncIterable<unknown> };
 
-const defaultInvoker: StreamTextInvoker = (options) =>
-    streamText({ ...options, stopWhen: stepCountIs(1) });
+const defaultInvoker: StreamTextInvoker = ({ system, ...options }) =>
+    // Our seam keeps a `system` field; the SDK's top-level option is now
+    // `instructions`. `allowSystemInMessages` opts into system-role messages
+    // in the array (v7 rejects them by default) — persisted chats rely on it.
+    streamText({
+        ...options,
+        instructions: system,
+        allowSystemInMessages: true,
+        stopWhen: isStepCount(1),
+    });
 
 export interface RealModelRegistryDeps {
     resolveProvider?: (name: string) => Promise<z.infer<typeof LlmProvider>>;
@@ -217,7 +225,7 @@ export class RealModelRegistry implements IModelRegistry {
             ...generationParams,
         });
 
-        for await (const raw of result.fullStream) {
+        for await (const raw of result.stream) {
             request.signal.throwIfAborted();
             const event = raw as {
                 type: string;
@@ -226,7 +234,13 @@ export class RealModelRegistry implements IModelRegistry {
                 toolName?: string;
                 input?: unknown;
                 finishReason?: string;
-                usage?: Record<string, number | undefined>;
+                usage?: {
+                    inputTokens?: number;
+                    outputTokens?: number;
+                    totalTokens?: number;
+                    inputTokenDetails?: { cacheReadTokens?: number | undefined };
+                    outputTokenDetails?: { reasoningTokens?: number | undefined };
+                };
                 providerMetadata?: unknown;
                 error?: unknown;
             };
@@ -390,24 +404,33 @@ function mergeProviderOptions(
 }
 
 function mapUsage(
-    usage: Record<string, number | undefined> | undefined,
+    usage:
+        | {
+              inputTokens?: number;
+              outputTokens?: number;
+              totalTokens?: number;
+              inputTokenDetails?: { cacheReadTokens?: number | undefined };
+              outputTokenDetails?: { reasoningTokens?: number | undefined };
+          }
+        | undefined,
 ): z.infer<typeof TurnUsage> {
     const mapped: z.infer<typeof TurnUsage> = {};
     if (!usage) {
         return mapped;
     }
-    for (const key of [
-        "inputTokens",
-        "outputTokens",
-        "totalTokens",
-        "reasoningTokens",
-        "cachedInputTokens",
-    ] as const) {
-        const value = usage[key];
+    // AI SDK 7 relocated cached/reasoning tokens into nested detail objects
+    // (usage.inputTokenDetails.cacheReadTokens, usage.outputTokenDetails.reasoningTokens);
+    // our persisted TurnUsage keeps the flat names.
+    const put = (key: keyof z.infer<typeof TurnUsage>, value: number | undefined) => {
         if (typeof value === "number" && Number.isFinite(value)) {
             mapped[key] = value;
         }
-    }
+    };
+    put("inputTokens", usage.inputTokens);
+    put("outputTokens", usage.outputTokens);
+    put("totalTokens", usage.totalTokens);
+    put("reasoningTokens", usage.outputTokenDetails?.reasoningTokens);
+    put("cachedInputTokens", usage.inputTokenDetails?.cacheReadTokens);
     return mapped;
 }