AI agents

Everything an AI coding agent needs to work with RunWisp is in the binary or at a stable URL.

Point your agent at the docs

Add one line to your project’s AGENTS.md, CLAUDE.md, or README:

This project uses RunWisp. Machine-readable reference: https://docs.runwisp.com/llms.txt

• llms.txt is a compact index of the documentation for agents. runwisp --help prints the same pointer, and so does every subcommand’s --help when its output isn’t a terminal, which is how agents run commands.
• Agent reference is a dense summary of the whole runwisp.toml schema, the CLI, and the REST API. Start there.
• Every page is also Markdown. Append .md to a page URL, or send Accept: text/markdown.

The config schema

runwisp schema

This prints the JSON Schema for runwisp.toml, also published at https://docs.runwisp.com/config.schema.json. To get validation and completion in editors, see Editor support.

Edit, validate, reload

To change tasks, an agent edits runwisp.toml, checks it, and tells the daemon to reload:

runwisp validate --json   # non-zero exit on error; each error has key, line, column
runwisp reload            # apply it to the running daemon
runwisp run backup --json # run a task; the outcome is JSON on stdout, logs on stderr

runwisp status --json and runwisp list --json show the daemon’s health, each task’s last run, and the configured tasks. The output shapes are in JSON output and on each command in the CLI reference.

To see why a run failed, runwisp logs backup -n 50 prints the last 50 lines of the task’s latest run and how it ended. It never waits for more output unless you pass -f. See runwisp logs.

An agent on the same machine as the daemon needs no password: the CLI talks to it over the control socket. Neither the socket nor the REST API can change a task definition; only runwisp.toml can.

Learn more

• CLI reference
• API reference, or runwisp openapi offline
• Configuration overview