mkurman/zorai

What it solves

Zorai is designed for operators who need AI agents that can handle long-running, durable work without losing state when a user interface is closed. It solves the problem of ephemeral agent sessions by providing a persistent daemon runtime that manages memory, tools, and goals across multiple control surfaces.

How it works

The system operates as a daemon that owns the execution state, including threads, workspace tasks, and goal runs. This allows users to interact with the agentic runtime via various interfaces—such as an Electron GUI, a TUI (Terminal User Interface), a CLI, or MCP clients—all of which reconnect to the same shared state. The runtime can plan work, spawn bounded sub-agents, and use tools, while maintaining a durable history and execution queue.

Who it’s for

It is built for operators and engineers who prefer terminal-first workflows but require the stability of a persistent agentic platform for complex, long-term tasks like investigating failing builds or tracing bugs across a workspace.

Highlights

• Durable Execution: Work continues in the daemon even after the UI is closed.
• Multi-Agent Orchestration: Features a set of specialized personas (e.g., Swarog for planning, Weles for governance) that can be added to threads or contacted via internal DMs.
• Multi-Surface Control: Unified state across a desktop app, TUI, and CLI.
• Integrated Voice: Built-in support for speech-to-text and text-to-speech workflows.
• Observability: Ability to export agent turns and tool calls to MLflow for conversation tracing.