@earendil-works/pi-ai 1.2.0 → 1.4.0

package/README.md

@@ -18,6 +18,7 @@ Unified LLM API with provider collections, automatic auth resolution, token and
 - [Auth](#auth)
   - [How Auth Resolves](#how-auth-resolves)
   - [Transforming Request Headers](#transforming-request-headers)
+  - [Client Identity](#client-identity)
   - [Credential Store](#credential-store)
   - [Environment Variables](#environment-variables)
 - [Tools](#tools)
@@ -26,7 +27,6 @@ Unified LLM API with provider collections, automatic auth resolution, token and
   - [Streaming Tool Calls with Partial JSON](#streaming-tool-calls-with-partial-json)
   - [Validating Tool Arguments](#validating-tool-arguments)
   - [Complete Event Reference](#complete-event-reference)
-  - [Compact Assistant Message Frames](#compact-assistant-message-frames)
 - [Image Input](#image-input)
 - [Image Generation](#image-generation)
 - [Classification](#classification)
@@ -35,6 +35,7 @@ Unified LLM API with provider collections, automatic auth resolution, token and
   - [Provider-Specific Options](#provider-specific-options-streamcomplete)
   - [Streaming Thinking Content](#streaming-thinking-content)
 - [Stop Reasons](#stop-reasons)
+- [Telemetry](#telemetry)
 - [Error Handling](#error-handling)
   - [Aborting Requests](#aborting-requests)
   - [Continuing After Abort](#continuing-after-abort)
@@ -425,6 +426,33 @@ Header names are merged case-insensitively. Explicit headers override auth/model
 
 `transformHeaders` belongs to `Models`, not `Provider`. A `Models` implementation must consume it and remove it before calling `Provider.stream*()`. Provider implementations continue receiving ordinary `ApiStreamOptions` or `SimpleStreamOptions` and never handle the transform themselves. Use this option instead of calling `getAuth(model)` before `stream*()`, which would resolve request auth twice.
 
+### Client Identity
+
+`identity` controls how requests identify the calling app to providers. Set it on `createModels()` as a default, or per request; a request's fields override the `Models` default field by field:
+
+```typescript
+const models = createModels({
+  identity: {
+    userAgent: 'detailed',
+    version: '2.1.0',
+    attribution: { name: 'my-app', url: 'https://my-app.example', categories: ['cloud-agent'] },
+  },
+});
+
+await models.completeSimple(model, context, { identity: { attribution: null } });
+```
+
+- `userAgent`: `"detailed"` (default) sends `pi/<version> (<platform>; <runtime>; <arch>)`, `"short"` sends `pi`, and `{ custom: "..." }` sends the given value. On Bedrock the value is appended to the AWS SDK user agent.
+- `version`: the version in the detailed `User-Agent`. Default: the pi-ai version.
+- `attribution`: provider attribution headers. Absent (default) or `null` sends none; `null` also turns off attribution from a merged identity. `{ name, url, categories? }` maps to each provider's headers:
+  - OpenRouter: `HTTP-Referer: <url>` (the app's identity in OpenRouter's rankings, grouped by origin), `X-OpenRouter-Title: <name>`, and `X-OpenRouter-Categories` (comma-separated).
+  - NVIDIA NIM: `X-BILLING-INVOKE-ORIGIN: <name>` with its first letter upper-cased.
+  - OpenCode: `x-opencode-client: <name>`.
+
+  `PI_ATTRIBUTION` is pi's: `{ name: "pi", url: "https://pi.dev", categories: ["cli-agent"] }`.
+
+Identity headers are defaults below provider, model, and request headers, so configured headers win and a `null` request header removes one. Providers that send a fixed client identity, such as GitHub Copilot and Anthropic OAuth, keep it. `withClientIdentity(models, identity)` returns a wrapper of a `Models` (or `MutableModels`) whose requests use `identity` over the one the `Models` was created with, for libraries that receive a `Models` from their host. `identity` is plain JSON, so it can be persisted with other request options.
+
 ### Credential Store
 
 Stored credentials (API keys entered interactively, OAuth tokens) live in a `CredentialStore` — one type-tagged credential per provider. pi-ai ships an in-memory default; apps inject persistent storage:
@@ -723,34 +751,6 @@ All streaming events emitted during assistant message generation:
 
 Streaming events for different content blocks are not guaranteed to be contiguous. Providers may emit deltas for text, thinking, and tool calls in the same upstream chunk, and pi may surface corresponding events interleaved, for example `text_start`, `text_delta`, `toolcall_start`, `text_delta`, `toolcall_delta`. Consumers must use `contentIndex` to associate each delta/end event with its block and must not assume that a block's `*_start`/`*_delta`/`*_end` sequence is uninterrupted by events for other blocks.
 
-### Compact Assistant Message Frames
-
-`AssistantMessageFrameEncoder` converts one stream into compact, persistable `AssistantMessageFrame` values. Create one encoder per stream and feed it every event in order. The encoder understands that `partial` is live: a block-start event consumed after the provider has already queued later deltas snapshots the current block once, and covered queued text/thinking deltas produce no duplicate frame. It retains only per-open-block counters plus, temporarily, the raw prefix needed to synchronize an already-advanced tool call. It never clones the growing full partial per token.
-
-The start frame contains message metadata with empty content. Text and thinking frames store each generated character at most once before the authoritative end frame. Tool calls that were already advanced when their start event was consumed use one compact JSON checkpoint before ordinary deltas resume. Terminal `done` and `error` events produce no frame because final message settlement is separate. A pre-generation `error` therefore produces no frames.
-
-`reduceAssistantMessageFrames()` is the canonical pure reducer. It reconstructs text, thinking, and tool-call arguments, including interleaved blocks identified by `contentIndex`, and rejects malformed sequences. It performs a single pass over the iterable and returns `undefined` when there is no start frame. End frames replace blocks with the provider's authoritative completed content and metadata. The reducer does not validate tool arguments against a TypeBox schema; call `validateToolCall` before execution.
-
-```typescript
-import {
-  AssistantMessageFrameEncoder,
-  reduceAssistantMessageFrames,
-  type AssistantMessageFrame,
-} from '@earendil-works/pi-ai';
-
-const encoder = new AssistantMessageFrameEncoder();
-const frames: AssistantMessageFrame[] = [];
-for await (const event of s) {
-  const frame = encoder.encode(event);
-  if (frame) frames.push(frame);
-}
-
-const reconstructedPartial = reduceAssistantMessageFrames(frames);
-const finalMessage = await s.result(); // Persist terminal settlement separately.
-```
-
-An encoder rejects duplicate starts, updates before start, `done` before start, events after a terminal event, duplicate block starts, and block-kind mismatches. An `error` before start is valid and returns no frame.
-
 ## Image Input
 
 Models with vision capabilities can process images. You can check if a model supports images via the `input` property. If you pass images to a non-vision model, they are silently ignored.
@@ -895,8 +895,8 @@ Classifier models consume structured JSON state, and optionally images, and answ
 | Provider | Model IDs | Auth |
 | --- | --- | --- |
 | `typesafe` | `jev-latest` | `TYPESAFE_API_KEY` |
-| `openrouter` | `typesafe/jev-1.13`, `~typesafe/jev-latest` | `OPENROUTER_API_KEY` or OpenRouter OAuth |
-| `cloudflare-workers-ai` | `typesafe/jev`, `@cf/cloudflare/clef`, `@cf/cloudflare/clef-flash` | `CLOUDFLARE_API_KEY` and `CLOUDFLARE_ACCOUNT_ID` |
+| `openrouter` | `typesafe/jev-1.13`, `typesafe/jev-1.14`, `~typesafe/jev-latest`, `cloudflare/clef`, `cloudflare/clef-flash`, `cloudflare/clef-omni` | `OPENROUTER_API_KEY` or OpenRouter OAuth |
+| `cloudflare-workers-ai` | `typesafe/jev`, `typesafe/jev-1.14.0`, `@cf/cloudflare/clef`, `@cf/cloudflare/clef-flash`, `@cf/cloudflare/clef-omni` | `CLOUDFLARE_API_KEY` and `CLOUDFLARE_ACCOUNT_ID` |
 | `vercel-ai-gateway` | `typesafe-ai/jev` | `AI_GATEWAY_API_KEY` |
 | `opencode` | `jev-1.13`, `jev-1.13-free` | `OPENCODE_API_KEY` |
 | `openai` | `gpt-6-luna` | `OPENAI_API_KEY` |
@@ -941,6 +941,18 @@ When the service reports token counts, `result.usage` carries them with their co
 
 `ClassifierContext.images` holds image blocks (`{ type: "image", data, mimeType }`) judged together with the state. Only models whose `input` includes `"image"` accept them; `classify()` returns an error result for other models.
 
+### System One images
+
+System One services take images in three forms. The model's `compat.imageInput` chooses the form (`"assets"`, `"images"`, or `"state-image-urls"`), and `compat.sendImageOptInHeader` controls TypeSafe's opt-in header:
+
+- **TypeSafe** (`typesafe`, and Jev on Workers AI) takes PNG and JPEG images as `assets`, `{ "<image-1>": { type: "image", format: "png", data_base64 } }`. Every asset must be referenced from the request, so the state is sent as `["<image-1>", ..., state]`. Images are experimental on TypeSafe's own API and need the `TypeSafe-Danger-Zone-Images: true` header, which the API sends with requests that carry images.
+- **Clef on Workers AI** takes up to four PNG, JPEG, or WebP images as an `images` list of data URLs, next to the unchanged state.
+- **OpenRouter** takes images as `{ type: "image_url", image_url: { url } }` parts before the state: `[part, ..., state]`. It answers requests with `assets` but drops the images, without an error.
+
+Built-in catalog entries set `compat` explicitly, so they keep their form when `baseUrl` points at a proxy. For models without it, the APIs detect the form: OpenRouter by provider `openrouter` or a `baseUrl` containing `openrouter.ai`, Clef by a model ID starting with `@cf/cloudflare/clef`, and TypeSafe's form with the opt-in header otherwise. A custom provider that reaches OpenRouter through a proxy matches neither check and must set `compat.imageInput: "state-image-urls"` and `compat.sendImageOptInHeader: false`; otherwise OpenRouter drops its images.
+
+Vercel AI Gateway and OpenCode list their Jev models with text input only.
+
 ### OpenAI Decisions
 
 The `openai-decisions` API calls OpenAI's [Decisions API](https://developers.openai.com/api/docs/guides/decisions) (`POST /v1/decisions`). GPT-6 Luna is its only model. The state is sent as JSON text; with images, the input is one user message with the state followed by the images as data URLs (at most 128). `choice` and `score` questions map to the Decisions types of the same name. `bool` questions become `predicate` questions; predicates have no criteria field, so the meanings of true and false are appended to the instructions. If OpenAI refuses to answer a question, the whole result is an error. Usage reports input tokens only, priced at the catalog input rate with OpenAI's long-context multiplier.
@@ -1099,6 +1111,40 @@ Every `AssistantMessage` includes a `stopReason` field that indicates how the ge
 
 `AssistantMessage` may also include `responseId`, a provider-specific upstream response or message identifier when the underlying API exposes one. Do not assume it is always present across providers.
 
+## Telemetry
+
+Pass an explicit parent span (`TelemetrySpan`, not an adapter) through `telemetry`; omitting it uses the no-op span:
+
+```typescript
+import { InMemoryTelemetryContext } from '@earendil-works/pi-telemetry';
+
+const telemetry = new InMemoryTelemetryContext();
+await telemetry.startSpan({ name: 'application.turn' }, async (span) => {
+  await models.completeSimple(model, context, { telemetry: span });
+});
+```
+
+`Models.stream()` and `streamSimple()` create `pi.ai.request`; `streamDeferred()` creates one `pi.ai.poll` per poll. Their completion helpers use the same spans. These spans include auth/setup, end at the terminal result, and do not require the caller to consume events. Direct provider/API calls are not instrumented.
+
+`AI_TELEMETRY_SCHEMA`, exported from the root and `@earendil-works/pi-ai/models`, defines the typed Pi-owned vocabulary:
+
+- Start attributes: `provider`, `api`, `model`, and `reasoning` only when `options.reasoning` is a string. Provider-native reasoning objects are ignored.
+- End attributes: `stop`, `usage.input`, `usage.output`, `usage.cache_read`, `usage.cache_write`, `cost.total`, and optional `response.id`. `stop: "error"` sets error status with the fixed category `provider_error` and `AssistantMessage.errorMessage` truncated to 500 characters (empty when absent). Cancellation keeps status `ok` with `stop: "aborted"`.
+- `pi.ai.first_chunk` marks the first normalized event, including setup errors. Each completed text, thinking, or tool-call block emits `pi.ai.block` with `type`, `index`, and `chars`; only tool calls include `tool`. Character counts use JavaScript string length; tool counts sum argument delta lengths without serializing arguments.
+
+When `options.telemetry.metrics` is available, the same requests record metrics:
+
+| Metric | Unit | Attributes |
+| --- | --- | --- |
+| `pi.ai.request.duration` | `ms` | `provider`, `model`, `stop` |
+| `pi.ai.first_chunk.duration` | `ms` | `provider`, `model` |
+| `pi.ai.tokens` | `token` | `provider`, `model`, `kind` (`input`, `output`, `cache_read`, `cache_write`) |
+| `pi.ai.cost` | `USD` | `provider`, `model` |
+
+Durations use `performance.now()`. Pending/deferred polls do not add tokens or cost; final completed/error polls do. A parent span without metric capabilities remains valid and simply records no metrics. Adapter-created spans carry the same optional metric and propagation capabilities as their root.
+
+Telemetry does not record prompts, generated text, reasoning content, or arguments in attributes/events. Error status includes up to 500 characters of provider error text, which may contain sensitive data; apply your export policy accordingly. Telemetry observes the existing forwarding path without another stream consumer or per-delta events. Response identifiers and error messages never become metric attributes.
+
 ## Error Handling
 
 Request failures after a stream is returned never throw: when a request ends with an error (including aborts and tool call validation errors), the streaming API emits an error event and the final message carries the details. Setup failures may emit `error` without `start`; failures after generation begins emit `start`, any observed updates, then `error`. Direct API `streamSimple()` calls throw synchronously when request auth is missing:
@@ -1527,13 +1573,13 @@ Notes:
 
 ## Cross-Provider Handoffs
 
-The library supports seamless handoffs between different LLM providers within the same conversation. This allows you to switch models mid-conversation while preserving context, including thinking blocks, tool calls, and tool results.
+The library supports seamless handoffs between different LLM providers within the same conversation. This allows you to switch models mid-conversation while preserving context, including text, tool calls, and tool results.
 
 When messages from one provider are sent to a different provider, the library automatically transforms them for compatibility:
 
 - **User and tool result messages** are passed through unchanged
-- **Assistant messages from the same provider/API** are preserved as-is
-- **Assistant messages from different providers** have their thinking blocks converted to text with `<thinking>` tags
+- **Assistant messages from the same provider, API, and model** are preserved as-is, including thinking blocks
+- **Assistant messages from a different model** have their thinking blocks dropped. Signatures and encrypted reasoning only work with the model that produced them, and readable thinking sent as plain text makes the new model treat it as part of the reply
 - **Tool calls and regular text** are preserved unchanged
 
 ```typescript
@@ -1554,7 +1600,7 @@ const claude = models.getModel('anthropic', 'claude-sonnet-4-5')!;
 context.messages.push({ role: 'user', content: 'What is 25 * 18?', timestamp: Date.now() });
 context.messages.push(await models.completeSimple(claude, context, { reasoning: 'medium' }));
 
-// Switch to GPT-5 - it will see Claude's thinking as <thinking> tagged text
+// Switch to GPT-5 - it sees Claude's answer, but not Claude's thinking
 const gpt5 = models.getModel('openai', 'gpt-5-mini')!;
 context.messages.push({ role: 'user', content: 'Is that calculation correct?', timestamp: Date.now() });
 context.messages.push(await models.complete(gpt5, context));
@@ -1565,7 +1611,7 @@ context.messages.push({ role: 'user', content: 'What was the original question?'
 const geminiResponse = await models.complete(gemini, context);
 ```
 
-All providers can handle messages from other providers — text, tool calls and results (including images), thinking blocks (transformed to tagged text), and aborted messages with partial content. This enables flexible workflows: start with a fast model, switch to a more capable one for complex reasoning, or maintain continuity across provider outages.
+All providers can handle messages from other providers — text, tool calls and results (including images), thinking blocks (dropped when the model changes), and aborted messages with partial content. This enables flexible workflows: start with a fast model, switch to a more capable one for complex reasoning, or maintain continuity across provider outages.
 
 ## System Messages
 
@@ -1598,7 +1644,7 @@ getCurrentTools(messages);        // []
 
 A custom `Provider` or `ProviderStreams` implementation reads the prompt and tools the same way from `context.messages`; `context.systemPrompt` and `context.tools` do not exist at that layer.
 
-Models that accept system messages mid-conversation (`supportsMidConvoSystemMessages` in the model's compat settings, set by the generated catalog for verified models) receive each later system message in place, so the cached prefix stays intact; section changes are framed by name for the model. Every other model receives `collapseSystemMessages(transcript)`: the replayed prompt and current tools as the leading system message, with later system messages dropped. Anthropic models that also set `supportsMidConvoToolChanges` send tool changes as native blocks (`inline-tools-2026-09-15` beta): the top-level tool list holds the initial tools plus a stable deferred placeholder and never changes, later tools are defined by value in `tool_addition` blocks (a new definition under an existing name replaces the old one), and removals are `tool_removal` references, so tool changes do not invalidate the prompt cache. That needs at least one initial tool; otherwise the current tool list is sent at the top level with the system text only. OpenAI Responses models with `supportsAdditionalTools` or `supportsToolSearch` anchor additive tool changes at their message; everything else sends the current tool list at the top level.
+Models that accept system messages mid-conversation (`supportsMidConvoSystemMessages` in the model's compat settings, set by the generated catalog for verified models) receive each later system message in place, so the cached prefix stays intact; section changes are framed by name for the model. Every other model receives `collapseSystemMessages(transcript)`: the replayed prompt and current tools as the leading system message, with later system messages dropped. Anthropic models that also set `supportsMidConvoToolChanges` send tool changes as native blocks (`inline-tools-2026-09-15` beta): the top-level tool list holds the initial tools plus a stable deferred placeholder and never changes, later tools are defined by value in `tool_addition` blocks (a new definition under an existing name replaces the old one), and removals are `tool_removal` references, so tool changes do not invalidate the prompt cache. That needs at least one initial tool; otherwise the current tool list is sent at the top level with the system text only. OpenAI Responses models with `supportsAdditionalTools` declare every tool in `additional_tools` items, like Codex: the initial tools in one at the head of the input and later additions and redefinitions at their message, with no top-level tool list. A later item replaces same-name definitions from earlier items but cannot replace a top-level tool or remove any tool, so removals are announced as text, later updates restate the tools that are still unavailable, and calls to removed tools fail at execution. OpenAI Responses models with only `supportsToolSearch`, and Kimi models with `supportsMidConvoToolAdditions`, anchor new tools at their message; a removal or a redeclared name moves the tool state up to that point into the top-level list once. Everything else sends the current tool list at the top level.
 
 ## Context Serialization
 

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;AAmJrB,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;AAwRD,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,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"}

package/dist/api/anthropic-messages.js

@@ -4,8 +4,8 @@ import { calculateCost } from "../models.js";
 import { appendAssistantMessageDiagnostic } from "../utils/diagnostics.js";
 import { AssistantMessageEventStream } from "../utils/event-stream.js";
 import { headersToRecord } from "../utils/headers.js";
-import { parseJsonWithRepair, parseStreamingJson } from "../utils/json-parse.js";
-import { getPiUserAgent } from "../utils/pi-user-agent.js";
+import { getIdentityHeaders } from "../utils/identity.js";
+import { parseJsonWithRepair, parseStreamingJson, parseToolCallArguments } from "../utils/json-parse.js";
 import { getProviderEnvValue } from "../utils/provider-env.js";
 import { retryProviderRequest } from "../utils/provider-retry.js";
 import { sanitizeSurrogates } from "../utils/sanitize-unicode.js";
@@ -14,7 +14,7 @@ import { getCurrentTools, getInitialSystemMessage, resolveTranscript, } from "..
 import { getJsonSchemaToolParameters, resolveJsonSchemaStrictSampling, } from "./constrained-sampling.js";
 import { buildCopilotDynamicHeaders, hasCopilotVisionInput } from "./github-copilot-headers.js";
 import { adjustMaxTokensForThinking, buildBaseOptions, clampMaxTokensToContext } from "./simple-options.js";
-import { transformMessages } from "./transform-messages.js";
+import { splitErrorToolResultImages, transformMessages } from "./transform-messages.js";
 /**
  * Resolve cache retention preference.
  * Defaults to "short" and uses PI_CACHE_RETENTION for backward compatibility.
@@ -75,6 +75,16 @@ const fromClaudeCodeName = (name, tools) => {
     }
     return name;
 };
+function convertImageBlock(image) {
+    return {
+        type: "image",
+        source: {
+            type: "base64",
+            media_type: image.mimeType,
+            data: image.data,
+        },
+    };
+}
 /**
  * Convert content blocks to Anthropic API format
  */
@@ -92,14 +102,7 @@ function convertContentBlocks(content) {
                 text: sanitizeSurrogates(block.text),
             };
         }
-        return {
-            type: "image",
-            source: {
-                type: "base64",
-                media_type: block.mimeType,
-                data: block.data,
-            },
-        };
+        return convertImageBlock(block);
     });
     // If only images (no text), add placeholder text block
     const hasText = blocks.some((b) => b.type === "text");
@@ -157,9 +160,6 @@ function mergeHeaders(...headerSources) {
     }
     return merged;
 }
-function mergeClientHeaders(...headerSources) {
-    return mergeHeaders({ "User-Agent": getPiUserAgent() }, ...headerSources);
-}
 function hasHeader(headers, name) {
     if (!headers)
         return false;
@@ -378,14 +378,12 @@ export const stream = (model, context, options) => {
     const normalizedContext = resolveTranscript(context, getAnthropicCompat(model).supportsMidConvoSystemMessages);
     const currentTools = getCurrentTools(normalizedContext.messages);
     (async () => {
-        const providerThinkingLevel = model.compat?.supportsMidConvoEffort ? (options?.effort ?? "high") : undefined;
         const output = {
             role: "assistant",
             content: [],
             api: model.api,
             provider: model.provider,
             model: model.id,
-            ...(providerThinkingLevel === undefined ? {} : { providerThinkingLevel }),
             usage: {
                 input: 0,
                 output: 0,
@@ -421,7 +419,7 @@ export const stream = (model, context, options) => {
                 }
                 const cacheRetention = resolveCacheRetention(options?.cacheRetention, options?.env);
                 const cacheSessionId = cacheRetention === "none" ? undefined : options?.sessionId;
-                const created = createClient(model, apiKey, options?.headers, options?.fetch, copilotDynamicHeaders, cacheSessionId, federation);
+                const created = createClient(model, apiKey, options?.headers, options?.fetch, copilotDynamicHeaders, cacheSessionId, federation, options?.identity);
                 client = created.client;
                 isOAuth = created.isOAuthToken;
             }
@@ -435,6 +433,8 @@ export const stream = (model, context, options) => {
             for (;;) {
                 options?.signal?.throwIfAborted();
                 const built = buildParams(model, normalizedContext, isOAuth, options, disablePreferredStrictTools);
+                if (built.providerThinkingLevel)
+                    output.providerThinkingLevel = built.providerThinkingLevel;
                 let params = built.params;
                 const nextParams = await options?.onPayload?.(params, model);
                 if (nextParams !== undefined) {
@@ -625,7 +625,7 @@ export const stream = (model, context, options) => {
                             });
                         }
                         else if (block.type === "toolCall") {
-                            block.arguments = parseStreamingJson(block.partialJson);
+                            Object.assign(block, parseToolCallArguments(block.partialJson));
                             // Finalize in-place and strip the scratch buffer so replay only
                             // carries parsed arguments.
                             delete block.partialJson;
@@ -780,7 +780,7 @@ export const streamSimple = (model, context, options) => {
 function isOAuthToken(apiKey) {
     return apiKey.includes("sk-ant-oat");
 }
-function createClient(model, apiKey, optionsHeaders, fetch, dynamicHeaders, sessionId, federation) {
+function createClient(model, apiKey, optionsHeaders, fetch, dynamicHeaders, sessionId, federation, identity) {
     // Copilot: Bearer auth.
     if (model.provider === "github-copilot") {
         const client = new PiAnthropic({
@@ -789,14 +789,14 @@ function createClient(model, apiKey, optionsHeaders, fetch, dynamicHeaders, sess
             baseURL: model.baseUrl,
             dangerouslyAllowBrowser: true,
             fetch,
-            defaultHeaders: mergeClientHeaders({
+            defaultHeaders: mergeHeaders(getIdentityHeaders(model, identity), {
                 accept: "application/json",
                 "anthropic-dangerous-direct-browser-access": "true",
             }, model.headers, dynamicHeaders, optionsHeaders),
         });
         return { client, isOAuthToken: false };
     }
-    // OAuth: Bearer auth, Claude Code identity headers
+    // OAuth: Bearer auth, Claude Code identity headers instead of the client identity.
     if (apiKey && isOAuthToken(apiKey)) {
         const client = new PiAnthropic({
             apiKey: null,
@@ -804,7 +804,7 @@ function createClient(model, apiKey, optionsHeaders, fetch, dynamicHeaders, sess
             baseURL: model.baseUrl,
             dangerouslyAllowBrowser: true,
             fetch,
-            defaultHeaders: mergeClientHeaders({
+            defaultHeaders: mergeHeaders({
                 accept: "application/json",
                 "anthropic-dangerous-direct-browser-access": "true",
                 "user-agent": `claude-cli/${claudeCodeVersion}`,
@@ -820,7 +820,7 @@ function createClient(model, apiKey, optionsHeaders, fetch, dynamicHeaders, sess
         const header = compat.sessionAffinityFormat === "openrouter" ? "x-session-id" : "x-session-affinity";
         sessionAffinityHeaders[header] = sessionId;
     }
-    const defaultHeaders = mergeClientHeaders({
+    const defaultHeaders = mergeHeaders(getIdentityHeaders(model, identity), {
         accept: "application/json",
         "anthropic-dangerous-direct-browser-access": "true",
     }, sessionAffinityHeaders, model.headers, optionsHeaders);
@@ -911,14 +911,15 @@ function buildParams(model, context, isOAuthToken, options, disablePreferredStri
     };
     const initialTools = initialSystemMessage?.toolsAdded ?? [];
     const nativeToolChanges = compat.supportsMidConvoSystemMessages && compat.supportsMidConvoToolChanges && initialTools.length > 0;
-    const converted = convertMessages(conversationMessages, isOAuthToken, cacheControl, compat.allowEmptySignature, model.compat?.supportsMidConvoEffort === true ? model.provider : undefined, nativeToolChanges ? convertToolDefinitions : undefined);
+    const thinkingDisabled = model.reasoning && options?.thinkingEnabled === false && model.thinkingLevelMap?.off !== null;
+    // Effort cannot change between messages while thinking is disabled, so "off" sends no effort markers.
+    const effortMessages = model.compat?.supportsMidConvoEffort === true && !thinkingDisabled;
+    const converted = convertMessages(conversationMessages, isOAuthToken, cacheControl, compat.allowEmptySignature, effortMessages ? model.provider : undefined, nativeToolChanges ? convertToolDefinitions : undefined);
     const activeEffort = options?.effort ?? "high";
     const betaFeatures = getBetaFeatures(model, context, isOAuthToken, nativeToolChanges, options);
     const params = {
         model: model.id,
-        messages: model.compat?.supportsMidConvoEffort === true
-            ? insertThinkingLevelMessages(converted, activeEffort)
-            : converted.messages,
+        messages: effortMessages ? insertThinkingLevelMessages(converted, activeEffort) : converted.messages,
         max_tokens: options?.maxTokens ?? model.maxTokens,
         stream: true,
         ...(betaFeatures.length > 0 ? { betas: betaFeatures } : {}),
@@ -971,9 +972,14 @@ function buildParams(model, context, isOAuthToken, options, disablePreferredStri
             params.tools = convertToolDefinitions(tools, toolCacheControl);
         }
     }
-    // Managed effort models always use adaptive thinking so prefix mismatches can
-    // be dropped instead of surfacing as persistent 400 responses.
-    if (model.compat?.supportsMidConvoEffort === true) {
+    if (thinkingDisabled) {
+        // No effort is sent, so the model default applies. Anthropic rejects `block_binding`
+        // with disabled thinking.
+        params.thinking = { type: "disabled" };
+    }
+    else if (effortMessages) {
+        // Managed effort models use adaptive thinking so prefix mismatches can
+        // be dropped instead of surfacing as persistent 400 responses.
         params.thinking = {
             type: "adaptive",
             display: options?.thinkingDisplay ?? "summarized",
@@ -981,29 +987,24 @@ function buildParams(model, context, isOAuthToken, options, disablePreferredStri
         };
         params.output_config = { effort: "high" };
     }
-    else if (model.reasoning) {
-        if (options?.thinkingEnabled) {
-            // Default to "summarized" so Opus 4.7 and Mythos Preview behave like
-            // older Claude 4 models (whose API default is also "summarized").
-            const display = options.thinkingDisplay ?? "summarized";
-            if (model.compat?.forceAdaptiveThinking === true) {
-                // Adaptive thinking: Claude decides when and how much to think.
-                params.thinking = { type: "adaptive", display };
-                if (options.effort) {
-                    params.output_config = { effort: options.effort };
-                }
-            }
-            else {
-                // Budget-based thinking for older models
-                params.thinking = {
-                    type: "enabled",
-                    budget_tokens: options.thinkingBudgetTokens || 1024,
-                    display,
-                };
+    else if (model.reasoning && options?.thinkingEnabled) {
+        // Default to "summarized" so Opus 4.7 and Mythos Preview behave like
+        // older Claude 4 models (whose API default is also "summarized").
+        const display = options.thinkingDisplay ?? "summarized";
+        if (model.compat?.forceAdaptiveThinking === true) {
+            // Adaptive thinking: Claude decides when and how much to think.
+            params.thinking = { type: "adaptive", display };
+            if (options.effort) {
+                params.output_config = { effort: options.effort };
             }
         }
-        else if (options?.thinkingEnabled === false && model.thinkingLevelMap?.off !== null) {
-            params.thinking = { type: "disabled" };
+        else {
+            // Budget-based thinking for older models
+            params.thinking = {
+                type: "enabled",
+                budget_tokens: options.thinkingBudgetTokens || 1024,
+                display,
+            };
         }
     }
     if (options?.metadata) {
@@ -1024,17 +1025,17 @@ function buildParams(model, context, isOAuthToken, options, disablePreferredStri
     if (allowedFallbackModels && allowedFallbackModels.length > 0) {
         params.fallbacks = allowedFallbackModels.map((fallback) => ({ model: fallback.model }));
     }
-    return { params, hasPreferredStrictTools };
+    return { params, hasPreferredStrictTools, providerThinkingLevel: effortMessages ? activeEffort : undefined };
 }
 // Normalize tool call IDs to match Anthropic's required pattern and length
 function normalizeToolCallId(id) {
     return id.replace(/[^a-zA-Z0-9_-]/g, "_").slice(0, 64);
 }
-function convertToolResult(msg) {
+function convertToolResult(msg, content = msg.content) {
     return {
         type: "tool_result",
         tool_use_id: msg.toolCallId,
-        content: convertContentBlocks(msg.content),
+        content: convertContentBlocks(content),
         is_error: msg.isError,
     };
 }
@@ -1099,14 +1100,7 @@ convertToolDefinitions) {
                         };
                     }
                     else {
-                        return {
-                            type: "image",
-                            source: {
-                                type: "base64",
-                                media_type: item.mimeType,
-                                data: item.data,
-                            },
-                        };
+                        return convertImageBlock(item);
                     }
                 });
                 const filteredBlocks = blocks.filter((b) => {
@@ -1197,16 +1191,24 @@ convertToolDefinitions) {
         else if (msg.role === "toolResult") {
             // Collect all consecutive toolResult messages, needed for z.ai Anthropic endpoint.
             const toolResults = [];
+            // Anthropic rejects error tool results that contain anything but text, so their images follow
+            // all tool results of the turn, which must come first in the user message (#10794).
+            const errorImages = [];
             let j = i;
             while (j < transformedMessages.length && transformedMessages[j].role === "toolResult") {
-                toolResults.push(convertToolResult(transformedMessages[j]));
+                const toolMsg = transformedMessages[j];
+                const split = splitErrorToolResultImages(toolMsg);
+                toolResults.push(convertToolResult(toolMsg, split?.content));
+                if (split) {
+                    errorImages.push({ type: "text", text: split.caption }, ...split.images.map(convertImageBlock));
+                }
                 j++;
             }
             // Skip the messages we've already processed.
             i = j - 1;
             params.push({
                 role: "user",
-                content: toolResults,
+                content: [...toolResults, ...errorImages],
             });
         }
     }