# Gini Agent

A personal AI agent runtime that treats chat as just one interface onto a system of record for conversations, runs, tasks, approvals, memory, skills, jobs, tools, traces and audit events, reachable identically from a CLI, a web app, mobile and messaging bridges.

Gini Agent is built around a runtime rather than a chat interface: a single process per instance owns state and does the actual work of running tasks, calling tools, remembering things and tracking approvals, while a Next.js web app, a CLI, a mobile app, MCP surfaces and messaging bridges are all clients of that same runtime.

When the agent needs a decision from the user, it surfaces a purpose-built control, such as a secure input field, a choice, a sign-in handoff or a confirm-before-send step, wherever the work is happening, and because every client speaks to the same runtime, that control reaches a phone as readily as a desktop. A skill-learning mechanism improves the agent's own skills from task outcomes using a two-tier reward and a daily human-reviewed gate.

It supports OpenAI, Anthropic, Bedrock, Azure, OpenRouter, DeepSeek and local model providers, can run headless in a Docker container with a real browser under Xvfb, exposes several tunnel modes for remote access, and includes a documented path for migrating an existing OpenClaw installation.

Website: https://github.com/Open-Curiosity/gini-agent