Own your fleet

agent

Turn any machine into a fleet node with one install — connect a device, dispatch real work.

github → MIT the instruments fleet node

01What it is

An AI workforce is only as useful as the machines it can actually touch. agent is the device-side half of an askalf fleet — the connector you run on a server, a laptop, or a remote box so the rest of the fleet can do real things on real computers, not just produce text.

You install it once on a machine you want the fleet to be able to act on, and it opens a persistent connection back to your own forge. From then on the device is a fleet node: it can be dispatched shell commands or Claude Code tasks, report what it has installed, and stream results back as they run. It is not a chat client or an autonomous coder — it is the substrate that lets a workforce reach beyond the prompt. Windows, Linux, and macOS all enroll the same way.

agent · enroll

# install on any machine
npm install -g @askalf/agent

# enroll it as a fleet node
askalf-agent connect <api-key> \
  --url wss://your-forge \
  --name laptop --install

Fig. 1 — one install, one connect, the device joins the fleet.

02What it does

Enrolls a device in one command

A single connect writes the config, registers the machine with your forge, and installs a system service that auto-starts. Close the terminal and the agent keeps running — the device stays a reachable fleet node across reboots and logins.

Runs the same on every OS

The same install path works on Windows, Linux, and macOS. Service install picks the native backend — a systemd unit on Linux, a launchd plist on macOS, a Scheduled Task on Windows — so auto-start is handled for you per platform.

Dispatches real work

Once connected, agents in the fleet can run shell commands or Claude Code tasks on the machine and watch the output stream back line by line while it is still running. A capability scan reports CPU, RAM, OS, and which tools are present — git, docker, node, python, claude — so work lands where it can run.

A built-in security policy

Every task is screened before it runs, with no configuration. A dangerous-command blocklist stops destructive patterns, output is scrubbed for secrets before it streams back, and every task — success, failure, or blocked — is appended to a local audit log. An optional policy file adds approval prompts, trusted agents, and path restrictions.

Stays connected on your terms

The agent heartbeats with memory and uptime, auto-reconnects with backoff after a drop, and connects over an encrypted wss:// link by default — plain ws:// is rejected for non-loopback hosts. It is self-hosted and MIT-licensed: your box, your forge, your terms.

03Where it sits

Part of The Instruments.

agent turns hardware into capacity. The Instruments are the agents that do real work with real tools — research, computer use, fleet nodes, and a browser — not just chat completions.

deepdive

A research agent — plan, search, fetch, synthesize, cite.

hands

A cross-platform computer-use agent with an audit log.

browser-bridge

Hardened headless Chromium in Docker, on your own CDP endpoint.

Reach beyond the prompt.

agent is open source and MIT-licensed. Read the code, install it, enroll your own machine.

View agent on GitHub →