Skip to Content
Runners

Runners

A runner is a small daemon you start on a machine — your laptop, a cloud VM, a Kubernetes pod, a Docker container, an edge device — anywhere Node runs. It connects that machine to almyty and exposes it to your agents: they run commands and use the local tools and files there, while every LLM call still goes through the backend. Your code and credentials never leave the machine.

A runner's detail page — detected binaries and the Coding agents card

Start a runner

Authenticate once, then start the daemon. It registers with almyty, reports the binaries it finds on the machine, and stays online listening for work.

npx @almyty/auth login
npx @almyty/runner start --name my-laptop
npx @almyty/runner status

It appears in the Runners page as online. Stop it with npx @almyty/runner stop or Ctrl-C in its terminal.

What an agent can do on it

The runner exposes a generic process surface — spawn, shell, read, write, signal — scoped to a workspace. So an agent can do whatever the machine can: run a build, query data that only lives there, operate a server, or drive a CLI it detects. The orchestration lives in the agent’s prompts; the runner just runs the processes.

Its capabilities (such as shell.exec and runner.info) are published as tools, tagged with the runner’s name. Attach them to a gateway and any agent can call them — one you built, or a Claude Code session over MCP.

Coding agents

The runner natively drives the major coding-agent CLIs — Claude Code, Codex, Gemini, Cursor, aider, and more. It detects which are installed at startup (shown in the Coding agents card above) and can launch them as unattended members: it injects the provider key for headless auth, isolates each one’s config and session, enables auto-approve, and reports live status (busy / idle / awaiting input). That is the building block for a multi-vendor coding team — one agent driving several CLIs on the same codebase. You compose the orchestration; the runner runs the members.

Workspaces

Work happens inside a workspace: a reservation of a working directory on the runner, with a time limit. Processes are isolated per workspace and cleaned up when it is released, so one job cannot reach another’s files.

Notes

  • One runner per account in v1.0; multi-machine scheduling is on the way.
  • Where LLM calls happen. almyty’s own agents run their LLM calls through the backend, not the runner — the runner just executes local processes. The exception is the coding-agent CLIs you spawn on the runner: they authenticate to their provider directly with a key you supply, so those calls leave from the machine itself.
ComplianceReferrals