@yourgpt/llm-sdk 2.6.0-beta.2 → 2.6.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2024 YourGPT
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -2,6 +2,13 @@
 
 Multi-provider LLM SDK with streaming. One API, any provider.
 
+## Release availability
+
+The examples below describe the SDK source on `main`. See
+[versions and compatibility](https://copilot-sdk.yourgpt.ai/docs/llm-sdk/beta-upgrade)
+for published npm versions and feature availability. The
+[Feature Lab](https://copilot-sdk.yourgpt.ai/playground/features) runs the workspace source.
+
 ## Installation
 
 ```bash
@@ -15,8 +22,10 @@ npm install @yourgpt/llm-sdk @anthropic-ai/sdk
 The SDK normalizes token limits and sampling options for recognized OpenAI model
 families, including GPT-6. Reasoning stays enabled by default. An unsupported
 `temperature` is omitted with a warning; the SDK does not silently disable
-reasoning. Use `onWarning` in the OpenAI provider configuration to handle these
-warnings yourself.
+reasoning. Responses also omits `message.output_text.logprobs` from `include`
+while reasoning is enabled, preserving other requested fields and encrypted
+replay state. Use `onWarning` in the OpenAI provider configuration to handle
+these warnings yourself.
 
 For a model that supports non-reasoning mode, select it explicitly when sampling
 control is required:
@@ -83,12 +92,28 @@ existing option; `responses.reasoningSummary` requests a reasoning summary.
 
 **Stateless replay:** use `store: false` and preserve the returned message objects,
 including `providerMetadata`. Each assistant's `providerMetadata.openai.response`
-contains the response ID, ordered native output items, raw usage, status and
-refusals. Completed encrypted reasoning, native search items, citations and
-function-call IDs survive the next turn. Automatic tool loops retain them too.
-Streams use completed items, never partially streamed reasoning. Models may
-return no reasoning items on simple requests; the SDK preserves what is returned.
-Do not edit opaque output items or carry them across unrelated conversations.
+contains minimal ordered continuation items, including function-call IDs, assistant
+phase and completed encrypted reasoning. It excludes the raw response envelope,
+native usage, logprobs, reasoning summaries and output-text annotations. Native
+tool state needed for replay is retained with diagnostic fields removed. Streams
+use completed items, never partially streamed reasoning. Keep this state intact
+and within its original conversation; do not treat it as trusted authorization.
+
+The separate **server-side model result** `providerMetadata` still contains the
+full response ID, status, native usage and output for diagnostics and stored
+continuation. Do not forward that complete model result to a public client.
+`runtime.chat()` retains top-level `usage` for billing and `events` for debugging;
+after billing, omit both before serializing it:
+
+```ts
+const result = await runtime.chat(request);
+await recordBilling(result.usage);
+const { usage, events, ...clientResult } = result;
+return Response.json(clientResult);
+```
+
+SSE response helpers omit usage by default. `runtime.stream(...).toResponse({
+includeUsage: true })` explicitly enables it for a server-controlled raw API.
 
 `generateText().response.messages` now includes the final assistant turn as well
 as tool turns. `streamText().response` returns the same complete history after
@@ -101,7 +126,8 @@ storage adapter stores display/tool records and does **not** yet guarantee
 lossless Responses replay after a remote session reload.
 
 **Stored continuation:** explicitly use `store: true`, take the returned
-`providerMetadata.openai.response.id`, and pass it as
+`generateText(...).providerMetadata.openai.response.id` (the server result, not
+the saved message metadata), and pass it as
 `openai: { responses: { previousResponseId: id } }` on the next call. Send only new
 input after that response, and resend system instructions; do not also resend
 its entire history. The SDK does not automatically advance or persist response
@@ -116,9 +142,12 @@ support remain model-specific. Raw usage preserves cache/reasoning breakdowns;
 this does not add pricing/accounting logic.
 
 Modern Responses accepts image URLs, data URLs/bytes, and file URL/base64 inputs.
-This does not add generalized image-bearing function results, native MCP approval
-flows, background jobs, or Decisions. Failed/incomplete/cancelled streams do not
-produce executable partial function calls or a successful final result.
+Function results can return images through `toModelOutput` or `_aiContent` on
+OpenAI Responses and Anthropic. Native Decisions uses the separate
+`experimental_decide` API described in [the Decisions guide](./docs/decisions.md).
+Native MCP approval flows and background jobs are not covered by this integration.
+Failed/incomplete/cancelled streams do not produce executable partial function
+calls or a successful final result.
 
 The contracts were checked against [OpenAI conversation state](https://developers.openai.com/api/docs/guides/conversation-state),
 [Responses request fields](https://developers.openai.com/api/reference/resources/responses/methods/create),
@@ -151,7 +180,7 @@ const adapter = createFallbackChain({
     const e = error as { status?: number; code?: string; param?: string };
     return (
       e?.status === 400 &&
-      e.code === "unsupported_parameter" &&
+      ["unsupported_parameter", "unsupported_value"].includes(e.code ?? "") &&
       e.param === "temperature"
     );
   },
@@ -165,6 +194,12 @@ errors after stream output or tool events never retry or switch models. Provider
 status/code/parameter fields survive OpenAI streaming errors; raw error causes
 are retained only in-process and omitted from JSON/SSE output.
 
+The live parameter regression checks both OpenAI error codes above. In particular,
+GPT-6 Luna returns `unsupported_value` when `temperature: 0` is deliberately sent
+with reasoning enabled. Normal SDK requests omit that unsupported parameter.
+`TRAINING_CONFLICT` is application policy: a provider error must not be interpreted
+as a successful guardrail decision or automatically become `respond`.
+
 Stream readers and final-result promises share one execution, including tool
 execution. Later readers replay retained stream parts; pass an abort signal to
 cancel generation. Cancellation during a tool stops subsequent tools and rejects
@@ -222,7 +257,10 @@ or `{ role: "system", content: "..." }` messages. It combines them into
 Anthropic's top-level `system` field: the request's nonempty `systemPrompt` (or
 the runtime default) first, followed by system messages in order. System messages
 are removed from the conversation sent to Anthropic. This applies to `chat`,
-`generate`, `stream`, and `processChat`.
+`generate`, `stream`, and `processChat`. If the same text is supplied through both
+forms, the explicit prompt is sent once. Null, non-text values and unsupported
+system blocks are ignored for proxy compatibility; valid text blocks are retained.
+Repeated message-only instructions are preserved in order.
 
 ```ts
 const decision = await runtime.chat({
@@ -232,8 +270,10 @@ const decision = await runtime.chat({
 ```
 
 The published `2.6.0-beta.1` Runtime adapter silently dropped system-role
-messages; use the explicit `systemPrompt` form above until you upgrade to a release
-containing this fix. The separate `generateText`/`streamText` Anthropic provider
+messages. Version `2.6.0-beta.2` first fixed that; use `2.6.0-beta.3` for its
+compatibility corrections, including tolerant system content and deduplication
+of identical explicit/system-role instructions.
+The separate `generateText`/`streamText` Anthropic provider
 already handles system-role messages.
 
 ## Selective Tool Loading
@@ -314,4 +354,34 @@ Tools can return native image content using `toModelOutput` or the existing `_ai
 
 `generateSpeech` / `transcribe` are separate server audio helpers, with `openaiSpeech` / `openaiTranscription` factories from `@yourgpt/llm-sdk/openai`. They support cancellation, validated encoded audio, and response metadata. Copilot's `VoiceRecorder` and `transcribeRecording` provide recorded microphone input and a reviewable transcript.
 
-See [image and audio contracts, examples, and limitations](./docs/multimodal-and-audio.md). Realtime voice calls and image generation are separate follow-ups.
+See [image and audio contracts, examples, and limitations](./docs/multimodal-and-audio.md).
+For realtime voice calls and image generation, see [the media guide](./docs/media.md).
+
+### Native Decisions
+
+`experimental_decide` provides native boolean, choice, and score questions with
+OpenAI, TypeSafe/Jev, OpenRouter, and Cloudflare adapters. It preserves named
+answers, distributions, refusals, provider confidence and usage. See the
+[Decisions guide](./docs/decisions.md) for supported inputs, provider-specific
+limits and the skills-demo example.
+
+### Images and realtime voice
+
+`generateImage` / `openaiImage` add image generation and editing.
+`openaiRealtime` creates server-side WebRTC sessions or short-lived credentials.
+Copilot supplies `RealtimeVoiceSession` and `useRealtimeVoice` for browser calls,
+transcripts, tool approval and cleanup. See [the media guide](docs/media.md).
+
+### Compatibility fixes
+
+OpenAI-compatible Chat transports preserve raw structured tool-message content
+instead of coercing arrays or objects to `[object Object]`. The SDK's explicit
+`modelOutput` representation still uses provider-specific conversion. This does
+not make every provider accept arbitrary object content; the provider validates
+its own schema. Use its supported content-part format for live calls.
+
+Malformed Responses function calls are rejected before tool execution with
+`status: 502`, `code: "INVALID_PROVIDER_RESPONSE"`. The fallback chain can retry or
+fall back according to its configured policy before any content has been emitted.
+It does not switch models after streaming content or tool execution has started.
+`generateText` also accepts model results that omit the optional `toolCalls` field.

package/dist/adapters/index.d.mts

@@ -1,7 +1,7 @@
-import { L as LLMAdapter, C as ChatCompletionRequest, a as CompletionResult } from '../base-BICaQaRS.mjs';
-export { A as AdapterFactory, l as AnthropicContentBlock, O as OpenAIContentBlock, j as attachmentToAnthropicDocument, i as attachmentToAnthropicImage, k as attachmentToOpenAIImage, f as formatMessages, c as formatMessagesForAnthropic, d as formatMessagesForOpenAI, b as formatTools, h as hasImageAttachments, g as hasMediaAttachments, m as messageToAnthropicContent, e as messageToOpenAIContent } from '../base-BICaQaRS.mjs';
-import { ae as OpenAIParameterOptions, O as OpenAITransportOptions, W as WebSearchConfig, i as StreamEvent } from '../types-Dg1Bcg1k.mjs';
-import { d as OllamaModelOptions } from '../types-DHQEMFHA.mjs';
+import { L as LLMAdapter, C as ChatCompletionRequest, a as CompletionResult } from '../base-DySbTP8C.mjs';
+export { A as AdapterFactory, l as AnthropicContentBlock, O as OpenAIContentBlock, j as attachmentToAnthropicDocument, i as attachmentToAnthropicImage, k as attachmentToOpenAIImage, f as formatMessages, c as formatMessagesForAnthropic, d as formatMessagesForOpenAI, b as formatTools, h as hasImageAttachments, g as hasMediaAttachments, m as messageToAnthropicContent, e as messageToOpenAIContent } from '../base-DySbTP8C.mjs';
+import { ae as OpenAIParameterOptions, O as OpenAITransportOptions, W as WebSearchConfig, i as StreamEvent } from '../types-D7lzMMLu.mjs';
+import { o as AnthropicThinking, d as OllamaModelOptions } from '../types-C7_qVSSd.mjs';
 import 'zod';
 
 /**
@@ -9,6 +9,8 @@ import 'zod';
  */
 interface OpenAIAdapterConfig extends OpenAIParameterOptions, OpenAITransportOptions {
     organization?: string;
+    project?: string;
+    timeout?: number;
     headers?: Record<string, string>;
     apiKey: string;
     model?: string;
@@ -94,14 +96,6 @@ declare class OpenAIAdapter implements LLMAdapter {
  */
 declare function createOpenAIAdapter(config: OpenAIAdapterConfig): OpenAIAdapter;
 
-/**
- * Extended thinking configuration
- */
-interface ThinkingConfig {
-    type: "enabled";
-    /** Budget for thinking tokens (minimum 1024) */
-    budgetTokens?: number;
-}
 /**
  * Anthropic adapter configuration
  */
@@ -110,8 +104,12 @@ interface AnthropicAdapterConfig {
     model?: string;
     /** Base URL for API endpoint */
     baseUrl?: string;
+    timeout?: number;
+    headers?: Record<string, string>;
+    cacheControl?: boolean;
     /** Enable extended thinking (for Claude 3.7 Sonnet, Claude 4) */
-    thinking?: ThinkingConfig;
+    thinking?: AnthropicThinking;
+    effort?: "low" | "medium" | "high" | "xhigh" | "max";
     temperature?: number;
     maxTokens?: number;
     /**

package/dist/adapters/index.d.ts

@@ -1,7 +1,7 @@
-import { L as LLMAdapter, C as ChatCompletionRequest, a as CompletionResult } from '../base-CpRPB8u8.js';
-export { A as AdapterFactory, l as AnthropicContentBlock, O as OpenAIContentBlock, j as attachmentToAnthropicDocument, i as attachmentToAnthropicImage, k as attachmentToOpenAIImage, f as formatMessages, c as formatMessagesForAnthropic, d as formatMessagesForOpenAI, b as formatTools, h as hasImageAttachments, g as hasMediaAttachments, m as messageToAnthropicContent, e as messageToOpenAIContent } from '../base-CpRPB8u8.js';
-import { ae as OpenAIParameterOptions, O as OpenAITransportOptions, W as WebSearchConfig, i as StreamEvent } from '../types-Dg1Bcg1k.js';
-import { d as OllamaModelOptions } from '../types-Dikj3En4.js';
+import { L as LLMAdapter, C as ChatCompletionRequest, a as CompletionResult } from '../base-6LzczuCg.js';
+export { A as AdapterFactory, l as AnthropicContentBlock, O as OpenAIContentBlock, j as attachmentToAnthropicDocument, i as attachmentToAnthropicImage, k as attachmentToOpenAIImage, f as formatMessages, c as formatMessagesForAnthropic, d as formatMessagesForOpenAI, b as formatTools, h as hasImageAttachments, g as hasMediaAttachments, m as messageToAnthropicContent, e as messageToOpenAIContent } from '../base-6LzczuCg.js';
+import { ae as OpenAIParameterOptions, O as OpenAITransportOptions, W as WebSearchConfig, i as StreamEvent } from '../types-D7lzMMLu.js';
+import { o as AnthropicThinking, d as OllamaModelOptions } from '../types-BXQMtaCX.js';
 import 'zod';
 
 /**
@@ -9,6 +9,8 @@ import 'zod';
  */
 interface OpenAIAdapterConfig extends OpenAIParameterOptions, OpenAITransportOptions {
     organization?: string;
+    project?: string;
+    timeout?: number;
     headers?: Record<string, string>;
     apiKey: string;
     model?: string;
@@ -94,14 +96,6 @@ declare class OpenAIAdapter implements LLMAdapter {
  */
 declare function createOpenAIAdapter(config: OpenAIAdapterConfig): OpenAIAdapter;
 
-/**
- * Extended thinking configuration
- */
-interface ThinkingConfig {
-    type: "enabled";
-    /** Budget for thinking tokens (minimum 1024) */
-    budgetTokens?: number;
-}
 /**
  * Anthropic adapter configuration
  */
@@ -110,8 +104,12 @@ interface AnthropicAdapterConfig {
     model?: string;
     /** Base URL for API endpoint */
     baseUrl?: string;
+    timeout?: number;
+    headers?: Record<string, string>;
+    cacheControl?: boolean;
     /** Enable extended thinking (for Claude 3.7 Sonnet, Claude 4) */
-    thinking?: ThinkingConfig;
+    thinking?: AnthropicThinking;
+    effort?: "low" | "medium" | "high" | "xhigh" | "max";
     temperature?: number;
     maxTokens?: number;
     /**