Fleet

Write a Python automation once. Run it across a fleet of machines —
self-healing, observable, and wired together end to end.

Fleet is a distributed automation framework. You write the work as a small Python class; Fleet distributes it to your workers, keeps them alive, restarts the ones that fall over, and streams their output wherever it needs to go — including into other automations, across hosts, with zero glue code.

Proven in production as the engine behind a Cloudflare Turnstile-solving farm — 8 hosts, 42 workers, 200+ operations per minute.

Why Fleet

• One control plane, many workers. The master is the single source of truth for configuration. Change it in one place and every worker reconciles to match — no redeploys, no manual restarts.
• Self-healing by design. A crashed slot is respawned on the next tick. A worker that drops off reconnects and resumes. A task that fails is retried with exponential backoff and dead-lettered after its attempt budget.
• Automations compose across machines. One workload's output stream is another's input — cross-host, type-checked, and durable. You wire pipelines, not plumbing.
• A browser pool that beats anti-bot. A built-in headed-Chromium pool ships in the box: statistically-weighted fingerprinting, per-flow proxy routing with a real Chrome TLS signature, and Cloudflare Turnstile solving.
• Backends that fit the deployment. Redis for horizontal scale, SQLite for a single VPS, in-memory for tests — one Backend interface, identical semantics, swap by URL.
• Operable from day one. Scoped auth tokens, a live dashboard, a Prometheus /metrics endpoint, and OpenTelemetry tracing are all wired up. REST and WebSocket APIs, plus a full CLI.
• Extend by installing a package. Automations, proxy providers, and anti-bot solvers are discovered via entry-points. Ship one as a pip package and Fleet finds it automatically.

How it works

One master runs on one host — where you set what to do and read what got done. Many workers, one per machine, check in, pull their config, and do the work. The work itself is a small Python class you write, ship as a pip package, and install on the machines; Fleet discovers it on its own and hands it out.

Quickstart

pip install fleet-framework

A complete automation is about 30 lines:

import asyncio
import httpx
from fleet.core import BaseConfig, ContinuousAutomation, register

class PingerConfig(BaseConfig):
    url: str
    interval_seconds: float = 5.0

@register("pinger")
class Pinger(ContinuousAutomation[PingerConfig]):
    Config = PingerConfig

    async def run_slot(self, ctx):
        async with httpx.AsyncClient(proxy=ctx.proxy) as client:
            while not ctx.shutdown.is_set():
                r = await client.get(ctx.config.url)
                await ctx.emit({"status": r.status_code})
                await asyncio.sleep(ctx.config.interval_seconds)

Full walkthrough — master, worker, config, output — at http://fleet.hackmap.win/getting-started/quickstart.

Documentation

Full docs live at http://fleet.hackmap.win/.

• Introduction
• Installation · Quickstart · Your first automation
• Architecture · Automations · Primitives
• Continuous automations · Batch automations · Inter-automation comms
• Browser-based automations · Deployment
• REST API · WebSocket protocol · CLI
• Auth · Observability · Troubleshooting

License

MIT. See LICENSE.