Open-Curiosity/gini-agent

The agent that remembers and learns.

What it solves

Gini Agent is a personal AI agent designed to move beyond simple chat interfaces. It solves the problem of "stranded" agents—where an AI cannot complete a task because it lacks a specific credential, needs a human to solve a CAPTCHA or sign in to a website, or needs a human to make a choice between several valid paths. Instead of asking the user in prose, Gini provides purpose-built interactive controls (like secure fields for passwords or live browser handoffs) directly in the chat interface to unblock the agent in real-time.

How it works

Gini uses a central gateway (a Bun runtime process) that manages the agent's state, memory, tools, and jobs. This gateway acts as the system of record for all conversations and tasks. Multiple clients—including a Next.js web app, a CLI, and an iOS app—connect to this same gateway via an authenticated API.

To handle complex tasks, the agent can run scheduled jobs and use approval-gated tools for files, terminals, and code. It also features a memory system with local embeddings and reranking to retain and recall information. The agent can improve its own skills based on task outcomes through a two-tier reward system and human gating.

Who it’s for

Individuals who want a highly capable personal agent that can perform multi-step work across various tools and platforms, and who prefer a professional product experience over a raw CLI or basic chat box.

Highlights

  • Interactive Controls: Replaces prose instructions with secure credential fields, choice prompts, and confirmation buttons for sensitive actions.
  • Browser Handoff: Allows users to sign in to websites or complete steps within the agent's own browser session from any device.
  • Secure Credential Management: Secrets are entered into secure cards that flow directly to the gateway, bypassing the LLM and the chat transcript.
  • Broad Provider Support: Compatible with OpenAI, Anthropic, Amazon Bedrock, Azure OpenAI, OpenRouter, DeepSeek, and local LLM servers.
  • Cross-Platform Access: A single runtime gateway can be accessed via web, mobile, and CLI surfaces.