@earendil-works/pi-ai 1.4.0 → 1.5.0

package/README.md

@@ -103,15 +103,16 @@ Unified LLM API with provider collections, automatic auth resolution, token and
 npm install @earendil-works/pi-ai
 ```
 
-TypeBox exports are re-exported from `@earendil-works/pi-ai`: `Type`, `Static`, and `TSchema`.
+Tool parameters are JSON Schema. The examples build them with `@earendil-works/pi-schema`, which also types the arguments (`npm install @earendil-works/pi-schema`); TypeBox schemas and plain JSON Schema objects work as well.
 
 ## Quick Start
 
 You build a `Models` collection of providers and stream through it. The quickest start registers every built-in provider; apps that care about bundle size register individual providers instead (see [Provider Factories](#provider-factories) and [Bundling and Tree Shaking](#bundling-and-tree-shaking)).
 
 ```typescript
-import { Type, type Context, type Tool } from '@earendil-works/pi-ai';
+import { type Context, type Tool } from '@earendil-works/pi-ai';
 import { builtinModels } from '@earendil-works/pi-ai/providers/all';
+import { obj, opt, str } from '@earendil-works/pi-schema';
 
 // A Models collection with every built-in provider registered
 const models = builtinModels();
@@ -119,12 +120,12 @@ const models = builtinModels();
 // Sync lookup against the collection
 const model = models.getModel('openai', 'gpt-4o-mini')!;
 
-// Define tools with TypeBox schemas for type safety and validation
+// Define tools with JSON Schema; pi-schema's builders also type the arguments
 const tools: Tool[] = [{
   name: 'get_time',
   description: 'Get the current time',
-  parameters: Type.Object({
-    timezone: Type.Optional(Type.String({ description: 'Optional timezone (e.g., America/New_York)' }))
+  parameters: obj({
+    timezone: opt(str({ description: 'Optional timezone (e.g., America/New_York)' }))
   })
 }];
 
@@ -233,8 +234,12 @@ for (const block of response.content) {
 }
 ```
 
+`usage.cost.totalFromProvider` is `true` when `usage.cost.total` is the billed cost reported by the provider and `false` when it is estimated from the model's catalog rates. A missing value, as in sessions recorded before this field existed, also means `false`.
+
 Snippets in the rest of this README assume a `models` collection set up like this (with the relevant providers registered).
 
+`usage.cost.total` is the best available total cost. For OpenRouter chat responses, it uses the cost reported by OpenRouter, including both the OpenRouter charge and upstream inference cost for BYOK requests. When provider accounting is unavailable or invalid, it falls back to the model's catalog rates. The `input`, `output`, `cacheRead`, and `cacheWrite` cost components remain catalog estimates and may not sum to a provider-reported total.
+
 ## Providers and Models
 
 A **provider** is the runtime unit: it owns its model catalog, its auth (API key resolution, OAuth flows), and its stream behavior. A `Models` collection holds providers and routes every request to the provider that owns the model.
@@ -534,34 +539,35 @@ Amazon Bedrock resolves ambient AWS credentials (`AWS_PROFILE`, access key pairs
 
 ## Tools
 
-Tools enable LLMs to interact with external systems. This library uses TypeBox schemas for type-safe tool definitions with automatic validation using TypeBox's built-in validator and value conversion utilities. TypeBox schemas can be serialized and deserialized as plain JSON, making them ideal for distributed systems.
+Tools enable LLMs to interact with external systems. A tool's `parameters` is a JSON Schema object: built with `@earendil-works/pi-schema`'s builders (which also type the arguments through its `Static`), with TypeBox, or written by hand. `validateToolCall()` and `validateToolArguments()` validate with pi-schema, coercing loosely typed arguments first (numeric strings, `"true"`, a single value for an array, `null` for an omitted optional property). Schemas are plain JSON, so they serialize with the rest of a context.
 
 ### Defining Tools
 
 ```typescript
-import { Type, type Tool, StringEnum } from '@earendil-works/pi-ai';
+import { type Tool, StringEnum } from '@earendil-works/pi-ai';
+import { arr, obj, str } from '@earendil-works/pi-schema';
 
-// Define tool parameters with TypeBox
+// Define tool parameters with pi-schema's builders
 const weatherTool: Tool = {
   name: 'get_weather',
   description: 'Get current weather for a location',
-  parameters: Type.Object({
-    location: Type.String({ description: 'City name or coordinates' }),
+  parameters: obj({
+    location: str({ description: 'City name or coordinates' }),
     units: StringEnum(['celsius', 'fahrenheit'], { default: 'celsius' })
   })
 };
 
-// Note: For Google API compatibility, use StringEnum helper instead of Type.Enum
-// Type.Enum generates anyOf/const patterns that Google doesn't support
+// Note: StringEnum gives { type: "string", enum: [...] }, which every provider accepts,
+// including Google's, which rejects anyOf/const patterns
 
 const bookMeetingTool: Tool = {
   name: 'book_meeting',
   description: 'Schedule a meeting',
-  parameters: Type.Object({
-    title: Type.String({ minLength: 1 }),
-    startTime: Type.String({ format: 'date-time' }),
-    endTime: Type.String({ format: 'date-time' }),
-    attendees: Type.Array(Type.String({ format: 'email' }), { minItems: 1 })
+  parameters: obj({
+    title: str({ minLength: 1 }),
+    startTime: str({ format: 'date-time' }),
+    endTime: str({ format: 'date-time' }),
+    attendees: arr(str({ format: 'email' }), { minItems: 1 })
   })
 };
 ```
@@ -574,10 +580,7 @@ Tools can opt in to provider-side constrained sampling. For JSON-schema tools, `
 const strictTool: Tool = {
   name: 'edit_file',
   description: 'Edit a file',
-  parameters: Type.Object({
-    path: Type.String(),
-    content: Type.String()
-  }, { additionalProperties: false }),
+  parameters: obj({ path: str(), content: str() }), // obj() is closed: additionalProperties: false
   constrainedSampling: { type: 'json_schema', strict: 'prefer' }
 };
 ```
@@ -588,9 +591,7 @@ Strict JSON-schema constrained sampling is supported for OpenAI, Anthropic, supp
 const patchTool: Tool = {
   name: 'apply_patch',
   description: 'Apply a patch',
-  parameters: Type.Object({
-    input: Type.String()
-  }, { additionalProperties: false }),
+  parameters: obj({ input: str() }),
   constrainedSampling: {
     type: 'grammar',
     variants: {
@@ -694,7 +695,7 @@ for await (const event of s) {
 
 ### Validating Tool Arguments
 
-When implementing your own tool execution loop, use `validateToolCall` to validate arguments before passing them to your tools:
+When implementing your own tool execution loop, use `validateToolCall` to validate arguments before passing them to your tools. It coerces loosely typed arguments with pi-schema (`compile(schema, { coerce: true }).parse()`), returns the coerced arguments, and throws an error listing pi-schema's errors (`path: message`) followed by the arguments it received:
 
 ```typescript
 import { validateToolCall, type Tool } from '@earendil-works/pi-ai';
@@ -1722,8 +1723,8 @@ models.setProvider(openaiProvider());
 
 Rules:
 
-- `@earendil-works/pi-ai/models` exports the model runtime (`createModels`, `createProvider`, model helpers, and their types) without TypeBox, built-in catalogs, or SDK implementations. Other types can still use `import type` from the root.
-- `@earendil-works/pi-ai` is the core entrypoint and does not import built-in catalogs, real provider factories, or SDK implementations, but it eagerly imports TypeBox and schema validation. Unbundled Node scripts do not tree-shake its unused exports; prefer `./models`, `./providers/faux`, and specific `./utils/*` subpaths when those are all you need.
+- `@earendil-works/pi-ai/models` exports the model runtime (`createModels`, `createProvider`, model helpers, and their types) without schema validation, built-in catalogs, or SDK implementations. Other types can still use `import type` from the root.
+- `@earendil-works/pi-ai` is the core entrypoint and does not import built-in catalogs, real provider factories, or SDK implementations, but it eagerly imports schema validation (pi-schema). Unbundled Node scripts do not tree-shake its unused exports; prefer `./models`, `./providers/faux`, and specific `./utils/*` subpaths when those are all you need.
 - `@earendil-works/pi-ai/providers/<provider>` imports that provider's catalog and lazy API wrapper only.
 - `@earendil-works/pi-ai/providers/all` imports every built-in provider factory and all catalogs. Use it only when you want the full built-in set.
 - With code splitting, provider SDKs stay in lazy chunks and load on first request.
@@ -1924,19 +1925,19 @@ Create a new API implementation file (for example `bedrock-converse-stream.ts`)
 
 Add a lazy wrapper `src/api/<api-id>.lazy.ts` (`<name>Api()` via `lazyApi()`) so providers can reference the implementation without importing its SDK. Add any root-level `export type` re-exports in `src/index.ts` that should remain available from `@earendil-works/pi-ai`.
 
-#### 3. Model Generation (`scripts/generate-models.ts`)
+#### 3. Model Catalog (`catalog/providers/<id>.ts`)
 
-- Add logic to fetch and parse models from the provider's source (e.g., models.dev API)
+- Add a provider definition and register it in `catalog/providers/index.ts`; see `catalog/README.md` for the build order and helpers
 - Map chat/tool-capable provider data to `Model`, image-generation data to `ImageModel`, and models.dev `type: "decision"` entries to `ClassifierModel`; hydration groups the ignored `src/providers/data/<id>.json` values by API while stable `src/providers/<id>.models.ts` wrappers derive exact model/API types directly from those JSON keys
 - Keep model ids unique within each provider and model type; emit separate entries when an upstream model supports multiple operations
-- Handle provider-specific quirks (pricing format, capability flags, model ID transformations)
+- Handle provider-specific quirks (pricing format, capability flags, model ID transformations) in the provider definition
 
 #### 4. Provider Factory (`src/providers/<id>.ts`)
 
 - `createProvider()` wiring catalog + auth + the lazy API wrapper
 - Auth: `envApiKeyAuth` for standard key providers, a custom `ApiKeyAuth` for ambient auth (AWS profiles, ADC), `lazyOAuth` where an OAuth flow exists
 - Register the factory in `src/providers/all.ts`
-- If it is a new API: register it in the builtin list in `src/compat.ts` and add the package subpath export in `package.json`
+- If it is a new API: register it in the builtin list in `src/compat.ts`; the `./api/*` subpath export in `package.json` already covers the new module
 
 #### 5. Tests (`test/`)
 
@@ -1947,7 +1948,6 @@ Create or update test files to cover the new provider:
 - `abort.test.ts` - Request cancellation
 - `empty.test.ts` - Empty message handling
 - `context-overflow.test.ts` - Context limit errors
-- `image-limits.test.ts` - Image support (if applicable)
 - `unicode-surrogate.test.ts` - Unicode handling
 - `tool-call-without-result.test.ts` - Orphaned tool calls
 - `image-tool-result.test.ts` - Images in tool results
@@ -1963,15 +1963,15 @@ For providers with non-standard auth (AWS, Google Vertex), create a utility like
 
 Update `src/core/model-resolver.ts`:
 
-- Add a default model ID for the provider in `DEFAULT_MODELS`
+- Add a default model ID for the provider in `defaultModelPerProvider`
 
 Update `src/cli/args.ts`:
 
 - Add environment variable documentation in the help text
 
-Update `README.md`:
+Update `docs/providers.md`:
 
-- Add the provider to the providers section with setup instructions
+- Add setup instructions, the environment variable, and the `auth.json` key
 
 #### 7. Documentation
 

package/dist/api/anthropic-messages.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"anthropic-messages.d.ts","sourceRoot":"","sources":["../../src/api/anthropic-messages.ts"],"names":[],"mappings":"AAAA,OAAO,SAAiC,MAAM,mBAAmB,CAAC;AAoBlE,OAAO,KAAK,EASX,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAMb,MAAM,aAAa,CAAC;AAuJrB,MAAM,MAAM,eAAe,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,KAAK,CAAC;AAE1E,MAAM,MAAM,wBAAwB,GAAG,YAAY,GAAG,SAAS,CAAC;AA2ChE,MAAM,WAAW,gBAAiB,SAAQ,aAAa;IACtD;;;;;;OAMG;IACH,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B;;;;OAIG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B;;;;;;;;;;;OAWG;IACH,MAAM,CAAC,EAAE,eAAe,CAAC;IACzB;;;;;;;;;;;OAWG;IACH,eAAe,CAAC,EAAE,wBAAwB,CAAC;IAC3C;;;;;OAKG;IACH,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B;;;;OAIG;IACH,UAAU,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IACtE;;;;OAIG;IACH,MAAM,CAAC,EAAE,SAAS,CAAC;CACnB;AAoRD,eAAO,MAAM,MAAM,EAAE,cAAc,CAAC,oBAAoB,EAAE,gBAAgB,CA2WzE,CAAC;AA2BF,eAAO,MAAM,YAAY,EAAE,cAAc,CAAC,oBAAoB,EAAE,mBAAmB,CAgDlF,CAAC"}
+{"version":3,"file":"anthropic-messages.d.ts","sourceRoot":"","sources":["../../src/api/anthropic-messages.ts"],"names":[],"mappings":"AAAA,OAAO,SAAiC,MAAM,mBAAmB,CAAC;AAoBlE,OAAO,KAAK,EASX,mBAAmB,EAEnB,cAAc,EACd,aAAa,EAIb,MAAM,aAAa,CAAC;AA+JrB,MAAM,MAAM,eAAe,GAAG,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,KAAK,CAAC;AAE1E,MAAM,MAAM,wBAAwB,GAAG,YAAY,GAAG,SAAS,CAAC;AA2ChE,MAAM,WAAW,gBAAiB,SAAQ,aAAa;IACtD;;;;;;OAMG;IACH,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B;;;;OAIG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAC;IAC9B;;;;;;;;;;;OAWG;IACH,MAAM,CAAC,EAAE,eAAe,CAAC;IACzB;;;;;;;;;;;OAWG;IACH,eAAe,CAAC,EAAE,wBAAwB,CAAC;IAC3C;;;;;OAKG;IACH,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B;;;;OAIG;IACH,UAAU,CAAC,EAAE,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAA;KAAE,CAAC;IACtE;;;;OAIG;IACH,MAAM,CAAC,EAAE,SAAS,CAAC;CACnB;AAoRD,eAAO,MAAM,MAAM,EAAE,cAAc,CAAC,oBAAoB,EAAE,gBAAgB,CA8VzE,CAAC;AA2BF,eAAO,MAAM,YAAY,EAAE,cAAc,CAAC,oBAAoB,EAAE,mBAAmB,CAgDlF,CAAC"}

package/dist/api/anthropic-messages.js

@@ -5,13 +5,15 @@ import { appendAssistantMessageDiagnostic } from "../utils/diagnostics.js";
 import { AssistantMessageEventStream } from "../utils/event-stream.js";
 import { headersToRecord } from "../utils/headers.js";
 import { getIdentityHeaders } from "../utils/identity.js";
-import { parseJsonWithRepair, parseStreamingJson, parseToolCallArguments } from "../utils/json-parse.js";
+import { parseJsonWithRepair } from "../utils/json-parse.js";
+import { applyOpenRouterReportedCost, isOpenRouterModel, } from "../utils/openrouter-cost.js";
 import { getProviderEnvValue } from "../utils/provider-env.js";
 import { retryProviderRequest } from "../utils/provider-retry.js";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
+import { ToolCallArgumentsStream } from "../utils/streaming-json.js";
 import { getSystemMessageText, renderSystemMessageUpdate } from "../utils/text.js";
 import { getCurrentTools, getInitialSystemMessage, resolveTranscript, } from "../utils/transcript.js";
-import { getJsonSchemaToolParameters, resolveJsonSchemaStrictSampling, } from "./constrained-sampling.js";
+import { getJsonSchemaToolParameters, isStrictToolLimitErrorMessage, prefersJsonSchemaStrictSampling, resolveJsonSchemaStrictSampling, } from "./constrained-sampling.js";
 import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.js";
 import { adjustMaxTokensForThinking, buildBaseOptions, clampMaxTokensToContext } from "./simple-options.js";
 import { splitErrorToolResultImages, transformMessages } from "./transform-messages.js";
@@ -456,11 +458,7 @@ export const stream = (model, context, options) => {
                         !(error instanceof Anthropic.APIError) ||
                         error.status !== 400 ||
                         error.type !== "invalid_request_error" ||
-                        ![
-                            "The compiled grammar is too large",
-                            "Too many strict tools (",
-                            "Schemas contains too many parameters with union types (",
-                        ].some((message) => error.message.includes(message))) {
+                        !isStrictToolLimitErrorMessage(error.message)) {
                         throw error;
                     }
                     disablePreferredStrictTools = true;
@@ -474,7 +472,12 @@ export const stream = (model, context, options) => {
             }
             await options?.onResponse?.({ status: response.status, headers: headersToRecord(response.headers) }, model);
             stream.push({ type: "start", partial: output });
-            const blocks = output.content;
+            // Stream state per Anthropic block index, kept off the content blocks so it never reaches partials.
+            const openBlocks = new Map();
+            const openBlock = (anthropicIndex) => {
+                const open = openBlocks.get(anthropicIndex);
+                return open === undefined ? undefined : { ...open, block: output.content[open.contentIndex] };
+            };
             for await (const event of iterateAnthropicEvents(response, options?.signal)) {
                 await options?.onProviderStreamEvent?.(event, model);
                 if (event.type === "message_start") {
@@ -509,105 +512,92 @@ export const stream = (model, context, options) => {
                         continue;
                     }
                     if (event.content_block.type === "text") {
-                        const block = {
-                            type: "text",
-                            text: event.content_block.text ?? "",
-                            index: event.index,
-                        };
-                        output.content.push(block);
+                        output.content.push({ type: "text", text: event.content_block.text ?? "" });
+                        openBlocks.set(event.index, { contentIndex: output.content.length - 1 });
                         stream.push({ type: "text_start", contentIndex: output.content.length - 1, partial: output });
                     }
                     else if (event.content_block.type === "thinking") {
-                        const block = {
+                        output.content.push({
                             type: "thinking",
                             thinking: event.content_block.thinking ?? "",
                             thinkingSignature: event.content_block.signature ?? "",
-                            index: event.index,
-                        };
-                        output.content.push(block);
+                        });
+                        openBlocks.set(event.index, { contentIndex: output.content.length - 1 });
                         stream.push({ type: "thinking_start", contentIndex: output.content.length - 1, partial: output });
                     }
                     else if (event.content_block.type === "redacted_thinking") {
-                        const block = {
+                        output.content.push({
                             type: "thinking",
                             thinking: "[Reasoning redacted]",
                             thinkingSignature: event.content_block.data,
                             redacted: true,
-                            index: event.index,
-                        };
-                        output.content.push(block);
+                        });
+                        openBlocks.set(event.index, { contentIndex: output.content.length - 1 });
                         stream.push({ type: "thinking_start", contentIndex: output.content.length - 1, partial: output });
                     }
                     else if (event.content_block.type === "tool_use") {
-                        const block = {
+                        output.content.push({
                             type: "toolCall",
                             id: event.content_block.id,
                             name: isOAuth
                                 ? fromClaudeCodeName(event.content_block.name, currentTools)
                                 : event.content_block.name,
                             arguments: event.content_block.input ?? {},
-                            partialJson: "",
-                            index: event.index,
-                        };
-                        output.content.push(block);
+                        });
+                        openBlocks.set(event.index, {
+                            contentIndex: output.content.length - 1,
+                            arguments: new ToolCallArgumentsStream(),
+                        });
                         stream.push({ type: "toolcall_start", contentIndex: output.content.length - 1, partial: output });
                     }
                 }
                 else if (event.type === "content_block_delta") {
+                    const open = openBlock(event.index);
                     if (event.delta.type === "text_delta") {
-                        const index = blocks.findIndex((b) => b.index === event.index);
-                        const block = blocks[index];
-                        if (block && block.type === "text") {
-                            block.text += event.delta.text;
+                        if (open?.block.type === "text") {
+                            open.block.text += event.delta.text;
                             stream.push({
                                 type: "text_delta",
-                                contentIndex: index,
+                                contentIndex: open.contentIndex,
                                 delta: event.delta.text,
                                 partial: output,
                             });
                         }
                     }
                     else if (event.delta.type === "thinking_delta") {
-                        const index = blocks.findIndex((b) => b.index === event.index);
-                        const block = blocks[index];
-                        if (block && block.type === "thinking") {
-                            block.thinking += event.delta.thinking;
+                        if (open?.block.type === "thinking") {
+                            open.block.thinking += event.delta.thinking;
                             stream.push({
                                 type: "thinking_delta",
-                                contentIndex: index,
+                                contentIndex: open.contentIndex,
                                 delta: event.delta.thinking,
                                 partial: output,
                             });
                         }
                     }
                     else if (event.delta.type === "input_json_delta") {
-                        const index = blocks.findIndex((b) => b.index === event.index);
-                        const block = blocks[index];
-                        if (block && block.type === "toolCall") {
-                            block.partialJson += event.delta.partial_json;
-                            block.arguments = parseStreamingJson(block.partialJson);
+                        if (open?.block.type === "toolCall" && open.arguments !== undefined) {
+                            open.block.arguments = open.arguments.append(event.delta.partial_json);
                             stream.push({
                                 type: "toolcall_delta",
-                                contentIndex: index,
+                                contentIndex: open.contentIndex,
                                 delta: event.delta.partial_json,
                                 partial: output,
                             });
                         }
                     }
                     else if (event.delta.type === "signature_delta") {
-                        const index = blocks.findIndex((b) => b.index === event.index);
-                        const block = blocks[index];
-                        if (block && block.type === "thinking") {
-                            block.thinkingSignature = block.thinkingSignature || "";
-                            block.thinkingSignature += event.delta.signature;
+                        if (open?.block.type === "thinking") {
+                            open.block.thinkingSignature = open.block.thinkingSignature || "";
+                            open.block.thinkingSignature += event.delta.signature;
                         }
                     }
                 }
                 else if (event.type === "content_block_stop") {
-                    const index = blocks.findIndex((b) => b.index === event.index);
-                    const block = blocks[index];
-                    if (block) {
-                        delete block.index;
+                    const open = openBlock(event.index);
+                    openBlocks.delete(event.index);
+                    if (open) {
+                        const { block, contentIndex: index } = open;
                         if (block.type === "text") {
                             stream.push({
                                 type: "text_end",
@@ -624,11 +614,8 @@ export const stream = (model, context, options) => {
                                 partial: output,
                             });
                         }
-                        else if (block.type === "toolCall") {
-                            Object.assign(block, parseToolCallArguments(block.partialJson));
-                            // Finalize in-place and strip the scratch buffer so replay only
-                            // carries parsed arguments.
-                            delete block.partialJson;
+                        else if (block.type === "toolCall" && open.arguments !== undefined) {
+                            Object.assign(block, open.arguments.finish());
                             stream.push({
                                 type: "toolcall_end",
                                 contentIndex: index,
@@ -680,6 +667,11 @@ export const stream = (model, context, options) => {
                     output.usage.totalTokens =
                         output.usage.input + output.usage.output + output.usage.cacheRead + output.usage.cacheWrite;
                     calculateCost(usageModel, output.usage);
+                    if (isOpenRouterModel(model) && event.usage) {
+                        // OpenRouter adds billing fields to streamed message_delta usage at runtime,
+                        // although its OpenAPI schema and the Anthropic SDK do not type those fields.
+                        applyOpenRouterReportedCost(output.usage, event.usage);
+                    }
                 }
             }
             if (options?.signal?.aborted) {
@@ -708,11 +700,6 @@ export const stream = (model, context, options) => {
             stream.end();
         }
         catch (error) {
-            for (const block of output.content) {
-                delete block.index;
-                // partialJson is only a streaming scratch buffer; never persist it.
-                delete block.partialJson;
-            }
             output.stopReason = options?.signal?.aborted ? "aborted" : "error";
             output.errorMessage = error instanceof Error ? error.message : JSON.stringify(error);
             stream.push({ type: "error", reason: output.stopReason, error: output });
@@ -901,10 +888,7 @@ function buildParams(model, context, isOAuthToken, options, disablePreferredStri
     let hasPreferredStrictTools = false;
     const convertToolDefinitions = (tools, toolCacheControl) => {
         const definitions = convertTools(tools, isOAuthToken, compat.supportsEagerToolInputStreaming, compat.supportsStrictTools, toolCacheControl, disablePreferredStrictTools);
-        if (definitions.some((definition, index) => {
-            const config = tools[index].constrainedSampling;
-            return definition.strict === true && config && config.type === "json_schema" && config.strict === "prefer";
-        })) {
+        if (definitions.some((definition, index) => definition.strict === true && prefersJsonSchemaStrictSampling(tools[index]))) {
             hasPreferredStrictTools = true;
         }
         return definitions;
@@ -985,7 +969,10 @@ function buildParams(model, context, isOAuthToken, options, disablePreferredStri
             display: options?.thinkingDisplay ?? "summarized",
             block_binding: { prefix_mismatch_behavior: "drop_block" },
         };
-        params.output_config = { effort: "high" };
+        // Effort markers set the level for every reply. The top-level effort is part of the cached
+        // prompt, so it stays fixed. It is "low" because organizations can cap effort and reject
+        // requests containing any level above the cap.
+        params.output_config = { effort: "low" };
     }
     else if (model.reasoning && options?.thinkingEnabled) {
         // Default to "summarized" so Opus 4.7 and Mythos Preview behave like
@@ -1045,7 +1032,9 @@ convertToolDefinitions) {
     const params = [];
     const assistantLevels = new Map();
     // Later system messages are held back and emitted directly before the next assistant
-    // message (or at the end of the transcript). Anthropic requires `tool_result` blocks to
+    // message that is sent (or at the end of the transcript). Assistant messages that convert to
+    // no blocks are dropped, so holding past them keeps an update from landing before a user
+    // message (#11091). Anthropic requires `tool_result` blocks to
     // immediately follow their `tool_use`, so a system message between them is rejected; this
     // also mirrors where the managed-effort system messages are inserted. As a result an
     // update placed before a user message in the transcript lands after it on the wire.
@@ -1118,7 +1107,6 @@ convertToolDefinitions) {
             }
         }
         else if (msg.role === "assistant") {
-            flushPendingSystemMessages();
             const blocks = [];
             for (const block of msg.content) {
                 if (block.type === "text") {
@@ -1176,6 +1164,7 @@ convertToolDefinitions) {
             }
             if (blocks.length === 0)
                 continue;
+            flushPendingSystemMessages();
             const messageIndex = params.length;
             params.push({
                 role: "assistant",
@@ -1299,8 +1288,7 @@ function convertTools(tools, isOAuthToken, supportsEagerToolInputStreaming, supp
     if (!tools)
         return [];
     return tools.map((tool, index) => {
-        const config = tool.constrainedSampling;
-        const strict = disablePreferredStrictTools && config && config.type === "json_schema" && config.strict === "prefer"
+        const strict = disablePreferredStrictTools && prefersJsonSchemaStrictSampling(tool)
             ? undefined
             : resolveJsonSchemaStrictSampling(tool, supportsStrictTools, isAnthropicStrictUnsupportedKeyword);
         const parameters = getJsonSchemaToolParameters(tool, strict);