iris-runtime

The iris command — author an agent, compile it to a portable image, run it anywhere. The install-free CLI for Iris: scaffold a config-plus-folder agent, build it into an open, content-addressed image, and run / serve / chat / deploy it — across hosts, models, and vendors — with the durable journal that makes a session resume byte-identical wherever it lands.

What it is

The unscoped npm package iris-runtime installs the iris binary. It wraps the Iris runtime behind a zero-dependency argv dispatcher: each subcommand is a unit-tested command function over @irisrun/agent, wired here to real fs / host defaults (the SQLite store + the provider named by the image's model-id prefix). Node ≥ 24, no build step.

Use it

npm i -g iris-runtime        # or run ad hoc: npx iris-runtime <command>

iris init my-agent           # scaffold agent.yaml + instructions.md + a tools/now tool
cd my-agent
iris build --out ./image     # compile the content-addressed agent image
iris chat ./image --session s1 --fake   # talk to it — no API key needed

Drop --fake and set the provider key (ANTHROPIC_API_KEY / OPENAI_API_KEY) for a real model.

Command groups:

• Author & inspect — init, build, inspect, schema, providers
• Distribute — verify, push, pull (local OCI layout; a real registry is manual)
• Run — run, serve, chat, deploy (--target <name> scaffolds for 9 platforms across edge / container / serverless; --list-targets)
• Operate & assure — audit, eval, schedule, journal

See the CLI reference for every command's flags, and the docs for the guided path from init to a deployed agent.

Part of Iris — own, portable, verifiable state.