Mayros: Local-first framework for private multi-agent AI and messaging
Mayros, developed by ApiliumCode, is a local-first framework for deploying and managing AI agents on personal hardware. It acts as a local gateway that routes model interactions to a terminal TUI, messaging channels, and orchestrates multi-agent workflows while connecting to external models and tools and supporting MCP interoperability. Built-in features include an AIngle-powered semantic graph memory, plugin extensibility, and markdown-defined agent behaviors, plus a Gateway control plane for messaging channels. The tool targets developers, researchers, and privacy-conscious power users who need on-device control and cross-tool compatibility.
What tasks can you actually use it for?
The framework runs agents locally as a control plane that links models to interactive terminals and messaging channels, enabling remote agent interaction through apps like Telegram and terminal-based workflows. It supports multi-agent orchestration and plugin hooks, so teams can assign separate agents to research, tool use, or automation tasks. The combination of a TUI, markdown-defined behaviors, and a Gateway control plane maps directly to developer-driven automation and messaging integration scenarios.
How reliable and verifiable are agent memories?
The tool uses a semantic graph engine to store agent memory as structured, verifiable relationships rather than flat files. That AIngle-based Cortex stores claims with graph relationships and cryptographic proofs, which helps agents retrieve context with provenance and makes post-hoc verification possible. For knowledge-heavy tasks, this model produces more traceable retrievals, though using graph memory requires users to adopt graph concepts and storage practices.
Does it require technical knowledge to get useful results?
The framework requires a developer-oriented environment, it runs on Node.js (recommended 22+) and installs via npm or from source, and it supports Linux, Raspberry Pi, macOS, WSL2, and Docker. The terminal TUI helps with interactive coding and agent management, but administrators should expect command-line setup and familiarity with TypeScript and dependency management. Dual MCP operation increases integration options, but adds configuration steps for tool exposure and consumption.
A practical choice for technically skilled users who need sovereign agents
The tool is a practical option for developers and researchers who require on-device agent control and verifiable memory, provided they accept a developer-focused setup and ongoing maintenance. Expect an initial configuration and a learning curve around graph-based memory and integration points. For teams that can manage local dependencies and command-line workflows, the framework delivers functional sovereignty and extensibility for agent-driven projects.
