KaozKit 0.2.0

Agents that checkpoint themselves, email, and a first contact that works.

New

• Agent-driven checkpoint and restore: host.snapshot(), a restore event on resume, kaoz --state-auto
• Email tools (send_email / read_email, HTML, attachments), news_search, a much more capable fetch_url
• kaoz --version; an unknown tool is a warning naming what to set, not a failed turn
• A package importable by its bare name; timers that carry their own callback
• OpenAI /v1/responses client for reasoning models; MLX reasoning effort and a loadingModel event
• Quick start, install script, and a debugging section (xsbug, xsdb) in the README

Changed — check your agents

• list_directory returns JSON instead of a text table
• Every runtime message is now English (tool errors, provider availability, engine errors)
• A heap snapshotted by another build is refused; the orchestrator marker is now generation 3
• The engine fixtures moved from agents/ to Sources/KaozJSTests/fixtures/; agents live in demo/

Builds without warnings on Swift 6.3 (two remain from mlx-swift-lm's own manifest). macOS 26+, Apple Silicon.

KaozKit 0.1.1

First public release.

JavaScript LLM agents on the XS engine (Moddable), embedded in Swift:

• Providers: Anthropic, OpenAI, Google, Mistral, Ollama, MLX (on-device), Apple Intelligence, ComfyUI, and more
• Tools with confinement by construction — secrets never enter JS
• Resident agents whose full state survives restarts via XS heap snapshots
• Sub-agents as isolated XS machines
• kaoz CLI for headless and daemon use

macOS 26+, Apple Silicon. MIT licensed (XS engine linked separately under its own license — see README).

KaozKit v0.1.0

First public release of KaozKit — a single Swift package for autonomous LLM agents written in JavaScript, running on an embedded XS (Moddable) engine.

Layered products: KaozJS / KaozJSCore for the bare JS↔Swift bridge, KaozKit for the agent runtime (providers, tools, memory, channels), KaozMLX for on-device MLX inference, and kaoz, the headless CLI / resident daemon. Requires macOS 26+ on Apple Silicon and a local Moddable checkout — see the README for setup.