apra-fleet
### Architecture & Positioning apra-fleet is an open-source multi-agent orchestration fleet developed by Apra Labs, engineered to coordinate AI coding agents across distributed physical and virtual machines from a single centralized conversational interface. While single-agent coding assistants are confined to the local filesystem and prone to unchecked hallucinations, apra-fleet implements a collaborative multi-node topology where autonomous agents write code, conduct peer reviews, and dispatch remote infrastructure tasks across an organization's device fleet. ### Core Capabilities & Differentiation The architecture is structured around two-agent consensus and remote execution: Doer-Reviewer Loops (pairing an authoring agent with an adversarial reviewing agent that enforces linting rules, security policies, and edge-case testing before changes merge), Multi-Machine Orchestration via SSH (distributing heavy compile jobs, container testing, and distributed trace analysis across remote nodes), Project Manager Skill Layer (breaking complex epics into structured multi-step sprints), Multi-Provider Interoperability (orchestrating Claude, Gemini, Codex, and Copilot in mixed configurations), and an Out-of-Band Credential Store. ### Deployment & Best Practices Installation is accomplished using the standalone binary installer: `curl -fsSL ... | ./apra-fleet-installer install`. In enterprise environments, configure SSH key-based authentication with restricted sudo profiles across target fleet nodes. When delegating complex multi-file refactors, pair a high-reasoning model as the Reviewer with a fast authoring model as the Doer; mandate that the Reviewer agent verifies local build and test execution logs before approving staged commits to guarantee codebase stability.