tell me more about language session models

Asked on 2026-06-18

1 search

Tailscan

This post is brought to you by Tailscan, the best devtool for Tailwind CSS.

Want to sponsor? We'd love to chat →

If by “language session models” you mean the LanguageModelSession / session-based model APIs in Foundation Models, here’s the short version:

What a language model session is

A language model session is the object you use to talk to a model in a stateful way. In these sessions:

  • you send prompts
  • the model responds
  • the interaction is stored in a transcript
  • the session can carry instructions and other context across turns

This is covered most directly in:

  • Deep dive into the Foundation Models framework
  • Meet the Foundation Models framework
  • What’s new in the Foundation Models framework

Why sessions matter

Sessions let the model remember the conversation, so follow-up prompts can refer back to earlier turns. Apple emphasizes that sessions are stateful, and that the transcript records prompts and responses.

Customizing a session

You can shape a session with:

  • instructions — developer-provided guidance on role, tone, style, and behavior
  • structured output — via APIs like Generable
  • tool calling — to let the model use app capabilities
  • dynamic schemas and, in newer sessions, Dynamic Profiles

Newer session model capabilities

The newer WWDC 2026 sessions add a lot more flexibility:

  • Dynamic Profiles let a single LanguageModelSession switch between different modes/models
  • you can configure different models per phase
  • transcript history can be transformed to trim, redact, or simplify context before prompting
  • sessions can expose additional metadata and lifecycle behavior

The most relevant newer sessions are:

  • What’s new in the Foundation Models framework
  • Build agentic app experiences with the Foundation Models framework
  • Bring an LLM provider to the Foundation Models framework

If you meant models behind the sessions

Apple also talks about multiple kinds of models that can power the same session API:

  • Apple’s on-device Foundation Model
  • Private Cloud Compute models
  • partner/open-source models via the model abstraction layer
  • your own model via Core AI / provider integration

If you want, I can also give you:

  1. a plain-English explanation of LanguageModelSession, or
  2. a comparison of sessions across WWDC 2025 vs WWDC 2026.
Bring an LLM provider to the Foundation Models framework

Bring an LLM provider to the Foundation Models framework

Extend the Foundation Models framework by implementing a LanguageModelExecutor for new models. Explore how to interface with the LanguageModelSession’s transcript, manage session state effectively, and optimize KV cache utilization. Find out how to support custom segment types and unlock advanced capabilities for your generative AI features.

What’s new in the Foundation Models framework

What’s new in the Foundation Models framework

Explore what’s new in the Foundation Models framework. Learn how to access Private Cloud Compute, integrate third-party and open source models, and work with vision capabilities. Discover context management APIs, built-in semantic search, and powerful primitives for creating agentic experiences in your apps.

Build agentic app experiences with the Foundation Models framework

Build agentic app experiences with the Foundation Models framework

Learn how to take your intelligence features further with Foundation Models framework primitives for dynamic context and agentic workflows. We’ll walk through engineering shared context, setting up privacy boundaries, and managing key value caching. Discover how to orchestrate smooth handoffs between local and server models.