apra-fleet

About apra-fleet

apra-fleet is an open-source MCP server that turns machines you already own into a fleet of AI agents. It routes work across mixed providers like Claude, Antigravity, Codex, Copilot, and local models via OpenCode, and runs durable, multi-hour workflows that can operate unattended. The project is Apache-2.0 licensed and free, launched this week on Product Hunt.

Review

apra-fleet acts as a control plane for AI agents running on real hardware - laptops, build servers, and VMs. It started as an internal tool at Apra Labs to combine several machines and provider subscriptions into one coherent system, and the team has been its own primary user for eight months. The core workflow: agents plan work from an issue tracker, dispatch tasks across machines, have different AI providers review changes, run integration tests, and file bugs for the next cycle.

One unusual design choice: the workflow engine doesn't know what a sprint is. It runs durable programs, and the team uses it for software delivery, but they also cite production deployments tracking cloud spend across a SaaS deployment and reconciling construction project financials. The emphasis is on "explore with agents, operate with programs" - an LLM figures out a workflow once, and then deterministic code handles the mechanical steps.

Key Features

• Runs agents across your existing machines, including laptops, build servers, and VMs
• Mixes provider tiers - premium models for planning and review, cheap or local models for mechanical work
• Durable, resumable workflows with stall detection and crash recovery
• Quick start via npm: `npm install -g @apralabs/apra-fleet`, then `apra-fleet install`, then `apra-fleet start`
• Self-building capability: the tool plans work from an issue tracker and fixes bugs against its own codebase

Pricing and Value

apra-fleet is free and open source under the Apache-2.0 license. You supply your own LLM API keys or subscriptions, so the cost is whatever your providers charge. The team reports their own e2e setup step went from a few dollars per run to roughly zero once the workflow became a deterministic program.

Pros

• Open source (Apache-2.0), so you can inspect the workflow engine and contribute back
• No vendor lock-in on models - you mix providers by cost tier and judgment needs
• Designed for unattended execution with stall detection and resumable state
• Works with machines already in your possession, no cloud infrastructure to spin up
• Self-building loop: it plans, executes, tests, and files bugs against its own codebase

Cons

• Single autonomous sprints on a registered fleet are solid, but the multi-sprint supervisor dashboard is still a preview
• Some permission setup remains manual, which could slow initial configuration on certain machines
• Not well suited for teams that want a fully managed SaaS solution with no interest in running their own infrastructure or managing API keys

apra-fleet suits engineers who already have a few machines and want to put them to work without paying for a managed agent platform. Teams with no tolerance for manual infrastructure setup or who prefer a fully hosted product should wait - or look elsewhere. The tool is at its best for those willing to trade a bit of configuration effort for full control over which models handle which tasks and where the work runs.