v0.1.0

The first release. Katari is a language for orchestration logic for AI agents: you write a program, the compiler lowers it to an IR, and a persistent runtime executes it. Every step of a run is durable, so a program can wait months on a human answer and pick up where it left off; parallelism and concurrency are in the language rather than in a library; and an effect row makes what a program may perform — which secrets it reads, whose calendar it writes, which questions it escalates to a person — something the compiler checks and the toolchain reports.

This release freezes that surface for the first time: the language and its diagnostics, the standard library, the IR and its schema version, the runtime API, the CLI and its lockfile discipline, the LSP and the editor extension, the package registry with its immutable snapshots, and the documentation site — served both to people and, as an MCP server, to the coding agents that write Katari.

Getting it

npm install -g @katari-lang/cli
katari init my-agent && cd my-agent
docker compose up -d && katari apply && katari run

The runtime image is ghcr.io/katari-lang/katari:0.1.0. The editor extension is katari-lang.katari-vscode, on the VS Code Marketplace and on Open VSX.

Where to start

• Quickstart — a project, a runtime, a run, and the escalation you answer.
• The tutorial — five chapters from an empty project to a deployed Discord bot.
• Examples — four complete, deployable projects: a release monitor with no model in it, a Slack standup bot with a human-approved digest, a two-desk Discord concierge, and a Gmail-to-calendar butler behind an approval gate.

Stability

Pre-1.0. A minor version may ship breaking changes, each recorded in these notes. Pin what you deploy: katari.lock pins the package closure, and the runtime image tag pins the runtime.