AI Core

AI Core is a runtime for executing AI Core packages from SDKs, applications, services, CLIs, and language bindings. It owns runtime orchestration: package install, model calls, Wasm actions, pipeline checkpoints, direct model chat, transcription handoff, and events.

Host applications own product state, user accounts, billing, durable storage, secrets, queues, UI, deployment policy, retries, and long-term logs.

Start Here

• Quick Start
• Language Bindings
• Process Host CLI
• AI Packages
• Runtime Contracts
• Speech To Text
• Local LLM
• API Reference

Integration Choices

Use the shared library when your host can load native code and wants the C ABI. Use the process binary when the host wants an async subprocess worker boundary. Use the Rust SDK artifact when the host is Rust. Use aip-pack only for package authoring and distribution tasks.

Every path uses the same concepts: apply host-owned config, install a package, start or resume work, and store durable state in the host.

Version Compatibility

The public docs track AI Core v0.1. Packages declare requirements.core in the package manifest. AI Core checks that requirement before activation, so ship packages and runtime artifacts from compatible release lines.

Read Next

• Quick Start: normal runtime flow.
• Language Bindings: FFI functions, result envelopes, memory ownership, and wrapper examples.
• Process Host CLI: async worker examples for Rust services, Flutter desktop, and Go.
• AI Packages: .aip package authoring, aip.template.json, pipeline.json, package requirements, and tool variants.