Is it possible to train an Adaptor for the Foundation Models to produce Generable output? If so what would the response part of the training data need to look like? Presumably, under the hood, the model is outputting JSON (or some other similar structure) that can be decoded to a Generable type. Would the response part of the training data for an Adaptor need to be in that structured format?

Hey, When generating responses with structured output and non-streaming API, it sometimes takes 3s, sometimes 10-20s. I am firing that request subsequently while testing the app. Is this by design, or any place I can learn more about what contributes to such variation?

Apologies if this is obvious to everyone but me... I'm using the Tahoe AI foundation models. When I get an error, I'm trying to handle it properly. I see the errors described here: https://developer.apple.com/documentation/foundationmodels/languagemodelsession/generationerror/context, as well as in the headers. But all I can figure out how to see is error.localizedDescription which doesn't give me much to go on. For example, an error's description is: The operation couldn’t be completed. (FoundationModels.LanguageModelSession.GenerationError error 2. That doesn't give me much to go on. How do I get the actual error number/enum value out of this, short of parsing that text to look for the int at the end? This one is: case guardrailViolation(LanguageModelSession.GenerationError.Context) So I'd like to know how to get from the catch for session.respond to something I can act on. I feel like it's there, but I'm missing it. Thanks!

After reviewing the WWDC sessions on Foundation Models and Core AI, I had a few questions around the practical limits and architectural direction of the platform. From my understanding, on-device Foundation Models remain optimized for privacy, latency, and efficiency, which naturally introduces constraints around context length and agent complexity. Has anything changed regarding the effective context window available to developers, or should we still design around similar context-management constraints as before? Core AI appears to introduce a more structured approach to building AI-powered applications. For developers building sophisticated assistants, how should we think about the boundary between application-level orchestration and framework-level orchestration? For example, are advanced patterns such as sub-agents, hierarchical planning, dynamic tool availability, and workflow decomposition expected to remain developer-managed, or are these areas Core AI aims to support more directly over time? I am also curious about Apple's vision for model interoperability. While Foundation Models provide an excellent on-device experience, many production-grade agent systems combine multiple specialized models for planning, reasoning, retrieval, and execution. Does Apple envision future pathways for integrating external models into Core AI driven workflows while maintaining the privacy and performance principles of the platform? Finally, for teams pushing the limits of on-device AI assistants, what architectural patterns do you recommend for handling long-horizon tasks, large context requirements, evolving toolsets, and multi-step reasoning within the current Foundation Models ecosystem?

Hello Team, I'm currently working on a proof of concept using Apple's Foundation Model for a RAG-based chat system on my MacBook Pro with the M1 Max chip. Environment details: macOS: 26.0 Beta Xcode: 26.0 beta 2 (17A5241o) Target platform: iPad (as the iPhone simulator does not support Foundation models) While testing, even with very small input prompts to the LLM, I intermittently encounter the following error: InferenceError::inference-Failed::Failed to run inference: Context length of 4096 was exceeded during singleExtend. Has anyone else experienced this issue? Are there known limitations or workarounds for context length handling in this setup? Any insights would be appreciated. Thank you!

What kind out-of-box on-device RAG support exists in the foundation models framework? (vector DBs, embedding methods etc., agentic RAG hooks?)

Hello, I recently learned about Private Cloud Compute (PCC): https://developer.apple.com/private-cloud-compute/ I am currently using a standard Developer Program account, and it seems that I cannot apply for the program directly. Is there an alternative? Also, is there any additional fee for using this service? If I want to call PCC in the app, for example, using the following code: let session = LanguageModelSession( model: PrivateCloudComputeLanguageModel() ) Do I need to apply for a specific plan to ensure that my App is successfully published on the App Store and available to users? Thank you!

I've created the following Foundation Models Tool, which uses the .anyOf guide to constrain the LLM's generation of suitable input arguments. When calling the tool, the model is only allowed to request one of a fixed set of sections, as defined in the sections array. struct SectionReader: Tool { let article: Article let sections: [String] let name: String = "readSection" let description: String = "Read a specific section from the article." var parameters: GenerationSchema { GenerationSchema( type: GeneratedContent.self, properties: [ GenerationSchema.Property( name: "section", description: "The article section to access.", type: String.self, guides: [.anyOf(sections)] ) ] ) } func call(arguments: GeneratedContent) async throws -> String { let requestedSectionName = try arguments.value(String.self, forProperty: "section") ... } } However, I have found that the model will sometimes call the tool with invalid (but plausible) section names, meaning that .anyOf is not actually doing its job (i.e. requestedSectionName is sometimes not a member of sections). The documentation for the .anyOf guide says, "Enforces that the string be one of the provided values." Is this a bug or have I made a mistake somewhere? Many thanks for any help you provide!

Download the Foundation Models Adaptor Training Toolkit Hi, after I clicked on the download button, I was redirected to this page https://developer.apple.com and did not download the toolkit.

Reports indicate PCC now extends to NVIDIA hardware in Google Cloud datacenters, and the flagship cloud model is refined using Gemini outputs. Now that PCC spans infrastructure outside Apple's own datacenters, what attestation or verifiable transparency is available to developers and users about where a given request was processed, and do the original "data unreachable even by Apple" guarantees hold unchanged across all hardware vendors? For apps with enterprise or regulated users, is there documented data residency behavior for PCC and for third-party model routing, and any contractual/compliance posture (e.g., regional pinning) developers can rely on? Given the EU and China availability gaps at launch, what is the recommended graceful-degradation path for apps that must function in those regions — fall back to on-device only, to a developer-supplied provider, or disable AI features? Does routing to a third-party cloud provider through the framework carry the same PCC privacy guarantees, or are those guarantees specific to Apple's own cloud models?

Hello! I'm following the Foundation Models adapter training guide (https://developer.apple.com/apple-intelligence/foundation-models-adapter/) on my NVIDIA DGX Spark box. I'm able to train on my own data but the example notebook fails when I try to export the artifact as an fmadapter. I get the following error for the code block I'm trying to run. I haven't touched any of the code in the export folder. I tried exporting it on my Mac too and got the same error as well (given below). Would appreciate some more clarity around this. Thank you. Code Block: from export.export_fmadapter import Metadata, export_fmadapter metadata = Metadata( author="3P developer", description="An adapter that writes play scripts.", ) export_fmadapter( output_dir="./", adapter_name="myPlaywritingAdapter", metadata=metadata, checkpoint="adapter-final.pt", draft_checkpoint="draft-model-final.pt", ) Error: --------------------------------------------------------------------------- ModuleNotFoundError Traceback (most recent call last) Cell In[10], line 1 ----> 1 from export.export_fmadapter import Metadata, export_fmadapter 3 metadata = Metadata( 4 author="3P developer", 5 description="An adapter that writes play scripts.", 6 ) 8 export_fmadapter( 9 output_dir="./", 10 adapter_name="myPlaywritingAdapter", (...) 13 draft_checkpoint="draft-model-final.pt", 14 ) File /workspace/export/export_fmadapter.py:11 8 from typing import Any 10 from .constants import BASE_SIGNATURE, MIL_PATH ---> 11 from .export_utils import AdapterConverter, AdapterSpec, DraftModelConverter, camelize 13 logger = logging.getLogger(__name__) 16 class MetadataKeys(enum.StrEnum): File /workspace/export/export_utils.py:15 13 import torch 14 import yaml ---> 15 from coremltools.libmilstoragepython import _BlobStorageWriter as BlobWriter 16 from coremltools.models.neural_network.quantization_utils import _get_kmeans_lookup_table_and_weight 17 from coremltools.optimize._utils import LutParams ModuleNotFoundError: No module named 'coremltools.libmilstoragepython'

Hi everyone, I've been building an on-device AI safety layer called Newton Engine, designed to validate prompts before they reach FoundationModels (or any LLM). Wanted to share v1.3 and get feedback from the community. The Problem Current AI safety is post-training — baked into the model, probabilistic, not auditable. When Apple Intelligence ships with FoundationModels, developers will need a way to catch unsafe prompts before inference, with deterministic results they can log and explain. What Newton Does Newton validates every prompt pre-inference and returns: Phase (0/1/7/8/9) Shape classification Confidence score Full audit trace If validation fails, generation is blocked. If it passes (Phase 9), the prompt proceeds to the model. v1.3 Detection Categories (14 total) Jailbreak / prompt injection Corrosive self-negation ("I hate myself") Hedged corrosive ("Not saying I'm worthless, but...") Emotional dependency ("You're the only one who understands") Third-person manipulation ("If you refuse, you're proving nobody cares") Logical contradictions ("Prove truth doesn't exist") Self-referential paradox ("Prove that proof is impossible") Semantic inversion ("Explain how truth can be false") Definitional impossibility ("Square circle") Delegated agency ("Decide for me") Hallucination-risk prompts ("Cite the 2025 CDC report") Unbounded recursion ("Repeat forever") Conditional unbounded ("Until you can't") Nonsense / low semantic density Test Results 94.3% catch rate on 35 adversarial test cases (33/35 passed). Architecture User Input ↓ [ Newton ] → Validates prompt, assigns Phase ↓ Phase 9? → [ FoundationModels ] → Response Phase 1/7/8? → Blocked with explanation Key Properties Deterministic (same input → same output) Fully auditable (ValidationTrace on every prompt) On-device (no network required) Native Swift / SwiftUI String Catalog localization (EN/ES/FR) FoundationModels-ready (#if canImport) Code Sample — Validation let governor = NewtonGovernor() let result = governor.validate(prompt: userInput) if result.permitted { // Proceed to FoundationModels let session = LanguageModelSession() let response = try await session.respond(to: userInput) } else { // Handle block print("Blocked: Phase \(result.phase.rawValue) — \(result.reasoning)") print(result.trace.summary) // Full audit trace } Questions for the Community Anyone else building pre-inference validation for FoundationModels? Thoughts on the Phase system (0/1/7/8/9) vs. simple pass/fail? Interest in Shape Theory classification for prompt complexity? Best practices for integrating with LanguageModelSession? Links GitHub: https://github.com/jaredlewiswechs/ada-newton Technical overview: parcri.net Happy to share more implementation details. Looking for feedback, collaborators, and anyone else thinking about deterministic AI safety on-device.

I'm experimenting with Foundation Models and I'm trying to understand how to define a Tool whose input argument is defined at runtime. Specifically, I want a Tool that takes a single String parameter that can only take certain values defined at runtime. I think my question is basically the same as this one: https://developer.apple.com/forums/thread/793471 However, the answer provided by the engineer doesn't actually demonstrate how to create the GenerationSchema. Trying to piece things together from the documentation that the engineer linked to, I came up with this: let citiesDefinedAtRuntime = ["London", "New York", "Paris"] let citySchema = DynamicGenerationSchema( name: "CityList", properties: [ DynamicGenerationSchema.Property( name: "city", schema: DynamicGenerationSchema( name: "city", anyOf: citiesDefinedAtRuntime ) ) ] ) let generationSchema = try GenerationSchema(root: citySchema, dependencies: []) let tools = [CityInfo(parameters: generationSchema)] let session = LanguageModelSession(tools: tools, instructions: "...") With the CityInfo Tool defined like this: struct CityInfo: Tool { let name: String = "getCityInfo" let description: String = "Get information about a city." let parameters: GenerationSchema func call(arguments: GeneratedContent) throws -> String { let cityName = try arguments.value(String.self, forProperty: "city") print("Requested info about \(cityName)") let cityInfo = getCityInfo(for: cityName) return cityInfo } func getCityInfo(for city: String) -> String { // some backend that provides the info } } This compiles and usually seems to work. However, sometimes the model will try to request info about a city that is not in citiesDefinedAtRuntime. For example, if I prompt the model with "I want to travel to Tokyo in Japan, can you tell me about this city?", the model will try to request info about Tokyo, even though this is not in the citiesDefinedAtRuntime array. My understanding is that this should not be possible – constrained generation should only allow the LLM to generate an input argument from the list of cities defined in the schema. Am I missing something here or overcomplicating things? What's the correct way to make sure the LLM can only call a Tool with an input parameter from a set of possible values defined at runtime? Many thanks!

I’d like to submit a feature request regarding the availability of Foundation Models in MessageFilter extensions. Background MessageFilter extensions play a critical role in protecting users from spam, phishing, and unwanted messages. With the introduction of Foundation Models and Apple Intelligence, Apple has provided powerful on-device natural language understanding capabilities that are highly aligned with the goals of MessageFilter. However, Foundation Models are currently unavailable in MessageFilter extensions. Why Foundation Models Are a Great Fit for MessageFilter Message filtering is fundamentally a natural language classification problem. Foundation Models would significantly improve: Detection of phishing and scam messages Classification of promotional vs transactional content Understanding intent, tone, and semantic context beyond keyword matching Adaptation to evolving scam patterns without server-side processing All of this can be done fully on-device, preserving user privacy and aligning with Apple’s privacy-first design principles. Current Limitations Today, MessageFilter extensions are limited to relatively simple heuristics or lightweight models. This often results in: Higher false positives Lower recall for sophisticated scam messages Increased development complexity to compensate for limited NLP capabilities Request Could Apple consider one of the following: Allowing Foundation Models to be used directly within MessageFilter extensions Providing a constrained or optimized Foundation Model API specifically designed for MessageFilter Enabling a supported mechanism for MessageFilter extensions to delegate inference to the containing app using Foundation Models Even limited access (e.g. short text only, strict execution limits) would be extremely valuable. Closing Foundation Models have the potential to significantly raise the quality and effectiveness of message filtering on Apple platforms while maintaining strong privacy guarantees. Supporting them in MessageFilter extensions would be a major improvement for both developers and users. Thank you for your consideration and for continuing to invest in on-device intelligence.

No matter what, the LanguageModelSession always returns very lengthy / verbose responses. I set the maximumResponseTokens option to various small numbers but it doesn't appear to have any effect. I've even used this instructions format to keep responses between 3-8 words but it returns multiple paragraphs. Is there a way to manage LLM response length? Thanks.

Since agents in Xcode 27 can now interact with the accessibility tree and screenshots, can we provide 'developer hints' in our code to help the agent distinguish between decorative UI and critical interactive elements during automated testing? Can the Evaluations framework be used to 'score' the efficiency of an agent’s navigation path through the app, helping us identify where our App Intents might be creating confusing or redundant loops for Apple Intelligence?

Hello, I am studying macOS26 Apple Intelligence features. I have created a basic swift program with Xcode. This program is sending prompts to FoundationModels.LanguageModelSession. It works fine but this model is not trained for programming or code completion. Xcode has an AI code completion feature. It is called "Predictive Code completion model". So, there are multiple on-device models on macOS26 ? Are there others ? Is there a way for me to send prompts to this "Predictive Code completion model" from my program ? Thanks

I can’t seem to find a way to include an image when prompting the new on-device model in Xcode, even though Apple explicitly states that the model was trained and tested with image data (https://machinelearning.apple.com/research/apple-foundation-models-2025-updates). Has anyone managed to get this working, or are VLM-style capabilities simply not exposed yet?

When doing some exploratory research into using Apple Intelligence in our aviation-focused application, I noticed that there were several times that key phases would be marked as inappropriate. I tried to stifle these using prompts and rules but couldn't get it to take hold. I was encouraged by an Apple employee to go ahead and post this so that the AI team can use the feedback. There were several terms that triggered this warning, but the two that were most prominent were: 'Tailwind' 'JFK' or 'KJFK' (NY airport ICAO/IATA codes)

@Generable enum Breakfast { case waffles case pancakes case bagels case eggs } do { let session = LanguageModelSession() let userInput = "I want something sweet." let prompt = "Pick the ideal breakfast for request: (userInput)" let response = try await session.respond(to: prompt,generating: Breakfast.self) print(response.content) } catch let error { print(error) } i want to test the @Generable demo but get error with below:decodingFailure(FoundationModels.LanguageModelSession.GenerationError.Context(debugDescription: "Failed to convert text into into GeneratedContent\nText: waffles", underlyingErrors: [Swift.DecodingError.dataCorrupted(Swift.DecodingError.Context(codingPath: [], debugDescription: "The given data was not valid JSON.", underlyingError: Optional(Error Domain=NSCocoaErrorDomain Code=3840 "Unexpected character 'w' around line 1, column 1." UserInfo={NSJSONSerializationErrorIndex=0, NSDebugDescription=Unexpected character 'w' around line 1, column 1.})))]))