CHANGELOG ========= 0.13 ---- * [BC BREAK] Add `ListenerInterface::onError()` and `Result\Stream\ErrorEvent`, dispatched when draining a `StreamResult` throws, so a listener can finalize on a failed stream where `onComplete()` never fires; `AbstractStreamListener` provides a no-op default * Add `Test\Replay\CassetteHttpClient` and `Test\Replay\HttpCassette` to record real HTTP responses (when the cassette file is missing) and replay them offline through the real bridge pipeline (Contract, `ModelClient`, `ResultConverter`) in tests, including raw Server-Sent Event streams; binary response bodies (images, audio, ...) are elided to a metadata stub (content type, byte size) and replayed as a small placeholder body * `TemplateRendererListener` now renders templates without `template_vars`; consequently, an expression referencing an absent variable now throws instead of reaching the normalizer as an empty object * Add `Test\Recording\RecordingProvider` (with `Cassette`/`ResultSerializer`) to record a real provider's result once (when the cassette file is missing) and replay it offline in tests, preserving the result metadata a provider reports alongside the answer - scalars, `null`, arrays, `FinishReason\FinishReason` and token usage - so a replayed result can be asserted on for usage; a metadata value the serializer cannot rebuild throws instead of being dropped silently * Add `TokenUsage\TokenUsageAggregation::getTokenUsages()` to read back the individual usages an aggregation sums up * Encode `Test\Recording\Cassette` with `JSON_THROW_ON_ERROR` and `JSON_PRESERVE_ZERO_FRACTION` and check the write, so a result holding a value JSON cannot represent (`NAN`, `INF`, invalid UTF-8) raises instead of truncating the cassette and destroying the interactions already recorded in it, and a recorded float with no fractional part no longer replays as an integer * Add `Test\Replay\AbstractBridgeReplayTestCase` for cassette-driven bridge replay tests, and make the `examples/` corpus a record/replay harness: `examples/runner --record` captures every HTTP interaction of an example into a committed cassette and refreshes the replay goldens, `ExamplesReplayTest` re-runs each recorded example offline through the full bridge pipeline in CI, without credentials, against those goldens * Add `Result\Stream\AssistantMessageStreamListener`, which rebuilds the assistant turn - text, thinking blocks with their signatures, tool calls, in provider order - from a streamed response * Accept a `StreamResult` in `Message::ofAssistant()` and add `StreamResult::getAssistantMessage()`, so a streamed turn is replayed like a buffered one, draining the stream if the caller has not read it * Add `Result\CustomToolCallResult` for `custom_tool_call` output items reported by provider-specific server-side tools (e.g. xAI's `x_search`), converted the same way as the other built-in tool call results instead of being surfaced as a `ToolCall` the application is expected to execute 0.12 ---- * [BC BREAK] Widen `ModelRouterInterface::resolve()` to accept `non-empty-string|Model` and return the new `ModelRouter\RoutingDecision` instead of a `ProviderInterface`, so a router can select the model - by input, rule set, or cost - and rewrite the invocation options for the selected provider, not only pick the provider serving the request. `Platform` now routes `Model` instances through the router and `ModelRoutingEvent` as well, instead of resolving them on a separate path * [BC BREAK] Throw `ModelNotFoundException` instead of `RuntimeException` when a provider has no `ModelClient` registered for a model, so a routing decision naming a provider that cannot serve the model fails consistently with an unknown model name 0.11 ---- * Add optional `$context` argument to `Contract\JsonSchema\Factory` build methods to limit schema to subset of properties * Add `Tool::getMetadata()` and `Tool::getMetadataValue()` to carry arbitrary custom data attached to a tool * Add `MessageBag::isLastMessageFrom()` to check whether the last message has a given role * Add `MaxOutputTokensException` for completed responses truncated by output token limits * Convert OpenAI Responses built-in tool output items (`web_search_call`, `file_search_call`, `code_interpreter_call`, `image_generation_call`, `mcp_call`, `mcp_list_tools`, `mcp_approval_request`, `computer_call`, `local_shell_call`) into typed results instead of skipping them. Adds `Result\WebSearchResult`, `Result\FileSearchResult`, `Result\McpCallResult`, `Result\McpListToolsResult`, `Result\McpApprovalRequestResult`, `Result\ComputerCallResult`, and `Result\LocalShellCallResult` (code interpreter reuses `ExecutableCodeResult`/`CodeExecutionResult`, image generation reuses `BinaryResult`), plus the matching `Message\Content` classes so the results round-trip through `Message::ofAssistant()`. Benefits the OpenAI, Azure OpenAI, and Scaleway bridges. * [BC BREAK] Remove `Capability::INPUT_MULTIPLE` from all embedding models since every embedding bridge already accepts multiple inputs in a single API call * Add `ResultConvertedEvent` and `ResultErrorEvent`, dispatched when the deferred result is actually converted (or conversion fails) instead of when the not-yet-resolved `ResultEvent` fires, so listeners can act on the resolved result * Add `provider` and `context` arguments to `#[Schema]` plus `SchemaProviderInterface` to contribute JSON Schema fragments computed at runtime (env, database, services) for tool parameters and structured output properties; `provider` accepts any container service ID and `context` is passed to `getSchemaFragment()` * Add `PartialJsonParser` for recovering partial JSON from streaming output * Add `DeferredResult::asPartialJsonStream()` yielding the largest recoverable JSON value from each streamed delta * Add streaming support for structured output via `DeferredResult::asStreamedObject()`; combining `stream: true` with `response_format: SomeClass::class` no longer throws and yields progressively populated typed objects through the new `PartialObjectStreamListener` * Add routing-aware mock test provider (`Test\MockPlatformFactory`, `Test\MockModelClient`, `Test\MockResultConverter`, `Test\MockModelCatalog`) for any-result-type mocks in tests, complementing `Test\InMemoryPlatform` * [BC BREAK] Rework `ToolCallMessage` to hold `Content\ContentInterface` parts (variadic constructor) instead of a single string content, mirroring `UserMessage`: `getContent()` now returns `Content\ContentInterface[]` and `asText()` returns the flattened text. `Message::ofToolCall()` accepts strings, `\Stringable`, and `Content\ContentInterface` values, upcasting strings to `Content\Text`. This enables multimodal tool results (e.g. text and images): the Anthropic and Amazon Bedrock (Nova/Converse) bridges render them as `tool_result` content blocks, the Gemini and Vertex AI bridges attach them as `inline_data` parts next to the `functionResponse`, and bridges without multimodal tool-result support degrade to text. * Add `ServerException`, thrown on transient server errors (HTTP 5xx) by `HttpStatusErrorHandlingTrait` and the bridge converters so consumers can retry without parsing error messages * Add `Capability::TEXT_TO_SPEECH_ASYNC`, `Capability::VIDEO_FRAME_TO_FRAME`, `Capability::VIDEO_WITH_SUBJECT`, and `Capability::MUSIC` cases * Add OpenAI image generation for the `gpt-image-*` models (`gpt-image-1`, `gpt-image-1-mini`, `gpt-image-1.5`, `gpt-image-2`), routed to the images endpoint by the model catalog * Add OpenAI image editing via the `images/edits` endpoint: pass the source image as a `Content\Image` through the `image` option of `Platform::invoke()` * [BC BREAK] Rename the OpenAI `DallE` bridge to `Image` (`Bridge\OpenAi\DallE` → `Bridge\OpenAi\Image`, and the `Bridge\OpenAi\DallE\*` namespace → `Bridge\OpenAi\Image\*`) and remove the `dall-e-2`/`dall-e-3` catalog entries retired by OpenAI * [BC BREAK] Replace the bridge-specific `ImageResult`, `Base64Image`, and `UrlImage` classes of the OpenAI image bridge with the generic `Result\BinaryResult` (single image) and `Result\MultiPartResult` (multiple images), matching the other multi-modal bridges * Throw an exception if the `response_format` option looks like a class, but is not * Add the `finish_reason` result metadata, exposing why a model stopped generating as a `FinishReason\FinishReason` object that normalizes the provider value (`length`, `max_tokens`, `MAX_TOKENS`, ...) into a `FinishReason\FinishReasonCase` while keeping the raw value. Set for buffered and streamed results by the OpenAI, Anthropic, Gemini, VertexAI, Amazon Bedrock, Azure, Ollama, Cohere, Mistral, DeepSeek, Cerebras, Scaleway, Docker Model Runner, Perplexity, MiniMax, amazee.ai and Generic bridges 0.10 ---- * Add `RawSseStream` to parse Server-Sent Events from backends that omit the `text/event-stream` content type (used by the OpenResponses bridge for the ChatGPT Codex backend); `SseStream` now decodes only content-type-advertised SSE * Add `IncompleteStreamException`, thrown by bridge converters when a stream ends before its terminal event * Throw `ModelNotFoundException` on HTTP 404 responses in the shared `HttpStatusErrorHandlingTrait` * Add `JsonBodyEncodingTrait` so model clients can encode JSON request bodies without aborting on malformed UTF-8 * Add support for passing a fully defined `Model` instance to `Platform::invoke()` (and `Provider::invoke()`) instead of a model name string, bypassing the model catalog; widen `ProviderInterface::supports()` to `string|Model` to route a model instance to the first provider whose model clients accept it * Add in-place `MessageBag::prepend()` and `MessageBag::removeSystemMessage()` * Add `MessageBag::withoutToolMessages()` to strip tool-call messages and tool-call-only assistant messages 0.9 --- * Add `ValidatorSubscriber` to validate structured output using Symfony Validator * Add support for multiple system messages in `MessageBag` * [BC BREAK] Rework `AssistantMessage` to hold `ContentInterface` parts (variadic constructor) instead of a single string content plus separate tool-call/thinking fields. Adds `Message\Content\Thinking`, `Message\Content\ExecutableCode`, and `Message\Content\CodeExecution` content classes, and makes `Result\ToolCall` implement `ContentInterface`. `Message::ofAssistant()` accepts strings, `ContentInterface`, and `ResultInterface` values, mapping `TextResult`/`ThinkingResult`/`ToolCallResult`/`ExecutableCodeResult`/`CodeExecutionResult`/`MultiPartResult` to their content equivalents; result types without a known mapping throw `InvalidArgumentException` so unhandled cases surface instead of being silently dropped. * Add optional `signature` field to `Message\Content\Text`, `Result\ToolCall`, and `Result\TextResult` for provider-scoped signatures (currently used by Gemini/Vertex AI for `thoughtSignature` round-trip). * Memoize conversion failures in `DeferredResult::getResult()` so subsequent calls re-throw the cached exception instead of re-running the converter * Surface tool calls executed by the `ClaudeCode` and `Codex` CLI bridges as `ToolCallResult` parts of a `MultiPartResult`, mirroring the inference-API behavior of the Anthropic and Gemini bridges 0.8 --- * [BC BREAK] Reduce visibility of `ImageResult::$revisedPrompt` to `private readonly`; use `getRevisedPrompt()` instead * Add `MultiPartResult` for exposing the parts inside a message * Add `ExecutableCodeResult`, `CodeExecutionResult` for exposing the executed code blocks and results * [BC BREAK] Replace variadic constructor parameters with array parameters in `VectorResult`, `ToolCallResult`, `RerankingResult`, `ToolCallComplete`, and `ImageResult` (OpenAI DallE bridge) * [BC BREAK] Rename `#[With]` attribute to `#[Schema]` and `WithAttributeDescriber` to `SchemaAttributeDescriber` * Add `ref` property to `#[Schema]` attribute to allow providing schema as file * Add support for `DiscriminatorMap` and `DateTimeImmutable` for structured output and tool calls * [BC BREAK] `Contract::create()` no longer accepts variadic `NormalizerInterface` arguments; pass an array instead * [BC BREAK] Change `public array $calls` and `public \WeakMap $resultCache` to private in `TraceablePlatform` - use `getCalls()` and `getResultCache()` instead * Add `ProviderInterface` and `Provider` abstraction for inference backends, with `Platform` becoming a router over one or more providers * Add `ModelRouterInterface` with `CatalogBasedModelRouter` as default strategy, and `CompositeModelCatalog` merging provider catalogs * Add `ModelRoutingEvent` dispatched by `Platform` before resolution, allowing listeners to observe/modify routing or short-circuit it by setting a provider * [BC BREAK] `Platform` constructor signature changed from `(modelClients, resultConverters, modelCatalog, contract, eventDispatcher)` to `(providers, modelRouter, eventDispatcher)` * Add `HttpStatusErrorHandlingTrait` and expose API errors (`AuthenticationException`, `BadRequestException`, `RateLimitExceededException`) across the Mistral, DeepSeek, Cerebras, Perplexity, Cohere, Gemini and OpenAI (Embeddings, DallE, TextToSpeech) bridges 0.7 --- * Add `TraceablePlatform` profiler decorator moved from AI Bundle * Add `asFile()` method to `BinaryResult` and `DeferredResult` for saving binary content to a file * Add typed streaming deltas (`TextDelta`, `ThinkingDelta`, `ThinkingStart`, `ThinkingSignature`, `ToolCallStart`, `ToolInputDelta`, `BinaryDelta`, `ChoiceDelta`, `ToolCallComplete`, `ThinkingComplete`) implementing `DeltaInterface` * [BC BREAK] Remove `Symfony\AI\Platform\Bridge\Ollama\OllamaMessageChunk`; Ollama streams now yield semantic deltas (`TextDelta`, `ThinkingDelta`, `ToolCallComplete`, `TokenUsage`) like the other bridges * Add generic `MetadataDelta` streaming support and use it for Perplexity citations/search results instead of provider-specific stream delta classes and listeners * Remove `DeltaInterface` from `BinaryResult`, `ChoiceResult`, and `ToolCallResult` * Remove `Usage` and `ThinkingContent` classes in favor of `TokenUsage` and `ThinkingComplete` * Add `DeltaEvent` replacing `ChunkEvent` in `ListenerInterface` * Add reranking support via `RerankingResult`, `RerankingEntry`, and `Capability::RERANKING` * Add `description` and `example` properties to `#[With]` attribute * Generate JSON schema from Symfony Validator constraints when available * Add `asTextStream()` method to `DeferredResult` to get a stream of `TextDelta` objects only * Add `reasoning_content` serialization in shared `AssistantMessageNormalizer` for OpenAI-compatible endpoints 0.6 --- * [BC BREAK] Change `Symfony\AI\Platform\Contract\JsonSchema\Factory` constructor signature in order to make schema generation extensible 0.4 --- * Add thinking support to `AssistantMessage` * Add support for object serialization in template variables via `template_vars` option * Add support for populating existing object instances in structured output via `response_format` option 0.3 --- * Add `StreamListenerInterface` to hook into response streams * [BC BREAK] Change `TokenUsageAggregation::__construct()` from variadic to array * Add `TokenUsageAggregation::add()` method to add more token usages * [BC BREAK] `CachedPlatform` has been renamed `CachePlatform` and moved as a bridge, please require `symfony/ai-cache-platform` and use `Symfony\AI\Platform\Bridge\Cache\CachePlatform` * [BC BREAK] `Metadata::merge()` method signature has changed to accept `Metadata` instead of array * [BC BREAK] Behavior of `Metadata::add()` has changed to merge existing keys instead of overwriting them * [BC BREAK] Move `Symfony\AI\Platform\Serializer\StructuredOutputSerializer` to `Symfony\AI\Platform\StructuredOutput\Serializer` 0.2 --- * [BC BREAK] Change `ChoiceResult::__construct()` from variadic to accept array of `ResultInterface` 0.1 --- * Add the component