hristo2612/jinn

What it solves

Jinn provides a management layer for AI agent command-line interfaces (CLIs), transforming standalone agents into a structured, persistent "AI company." It solves the problem of agent fragmentation by adding shared organizational structure, durable task tracking, and reusable automation that persists across individual sessions.

How it works

Jinn operates as a local gateway daemon and web dashboard that orchestrates existing agent CLIs installed on your system (such as Claude Code, Codex, or Grok). It does not provide its own AI logic; instead, it acts as a "bus" that dispatches work to these engines. It manages company state through YAML-defined roles, a durable Todo ledger, and a graph-based workflow system, providing a unified interface at localhost:7777 to operate the organization.

Who it’s for

It is designed for users who already use agent CLIs and want to scale their usage into complex, multi-agent pipelines, such as content creation pipelines, automated research organizations, or support desks requiring human approval.

Highlights

• Organizational Model: Define named employees with ranks, departments, and reporting lines in YAML.
• Durable Work Ledger: A Todo system that tracks assignees, priorities, and approvals across sessions.
• Visual Workflows: Reusable graph procedures with sequential, conditional, and parallel paths.
• Engine Agnostic: Supports multiple agent CLIs (Claude, Codex, Grok, Hermes, etc.) and allows for fallback chains if one engine fails.
• Knowledge Management: Durable Markdown notes and an "Experiments" system for tracking hypotheses and metrics.
• Extensibility: A plugin system for adding dashboard pages and background tasks, and a built-in MCP for company operations.