starnet

Product Introduction

1. Definition: Starnet is a local-first, desktop-based autonomous AI agent orchestration platform. It is a technical application that allows users to deploy, manage, and visually monitor AI agents (a "crew") that execute multi-step tasks directly on the user's local machine.
2. Core Value Proposition: It exists to provide developers and technical users with full control, transparency, and offline capability for AI workflows, solving the problem of opaque, cloud-dependent AI tools by enabling local AI agent execution with user-owned API keys.

Main Features

1. Project-Centric COMMS Orchestrator: Each project operates within a persistent, centralized conversation interface called COMMS. This orchestrator manages the project's context, selected AI crew, and all delegated work, bringing coordination, conversation history, and task status into a single, unified glass-themed panel. It maintains durable worker conversations and returns results to the main orchestrator for review.
2. Local-First Agent Execution & Control: AI agents run locally on the user's desktop. The platform provides granular control, allowing users to send directions to working agents, request work stoppage, and view expandable activity details including the agent, task, tools used, and outputs. It features agent pathfinding (e.g., yielding in hallways) within a customizable 3D station environment.
3. Secure External Integrations (Google Files & Windows CUA): Offers secure, scoped integrations. For Google, it uses the limited drive.file OAuth scope via a picker to grant tools for specific Docs and Sheets only, not broad Drive access. For Windows, it provides an optional CUA (Accessibility) driver for computer control, enabling agents to perform UI automation, screenshots with semantic elements, and input simulation, with credentials encrypted via the OS credential store.

Problems Solved

1. Pain Point: Eliminates reliance on opaque cloud AI services where users have no visibility into data processing, incur unpredictable costs, and lack customization. It solves the "black box" problem in autonomous AI agent platforms.
2. Target Audience: Primary users are software developers, DevOps engineers, data scientists, and tech-savvy tinkerers who build and customize automation workflows. Secondary users are technical project managers seeking transparent oversight of AI-assisted tasks.
3. Use Cases: Essential for scenarios requiring privacy-sensitive data processing, reproducible local automation, complex multi-agent research orchestration, and custom AI toolchain development where cloud API latency or cost is prohibitive.

Unique Advantages

1. Differentiation: Unlike cloud-based competitors (e.g., some aspects of LangChain/LangGraph cloud, SmythOS), Starnet is fundamentally local-first and BYOK (Bring Your Own Keys), ensuring no data leaves the machine unless via user-authorized APIs. Unlike pure CLI frameworks, it adds a persistent visual interface (the 3D station) for agent oversight.
2. Key Innovation: Its durable, recoverable conversation architecture ensures delegated agent work and context survive app restarts and failures. The COMMS orchestrator seamlessly blends project context, crew management, and activity monitoring into one recoverable session, a significant UX innovation over separate chat and agent logs.

Frequently Asked Questions (FAQ)

1. Is Starnet completely offline? No, Starnet is local-first but not offline-only. The core application and agent orchestration run on your desktop, but agents typically require external API calls (e.g., to OpenAI, Anthropic) using your keys. Some abilities, like Google file access, also require internet connectivity.
2. How does Starnet handle data privacy and security? Starnet prioritizes privacy by executing all agent logic locally. Your API keys and Google connector credentials are encrypted using your operating system's credential store. It uses least-privilege OAuth scopes (like drive.file) for Google integrations, limiting access to only files you explicitly select.
3. What AI models can I use with Starnet? Starnet is model-agnostic. You can use any AI model's API that you have access to, including OpenAI GPT, Anthropic Claude, or open-source models via compatible endpoints, by providing your own API keys in the application configuration.
4. Can I use Starnet for automated UI testing or RPA? Yes, on Windows x64, the optional CUA (Accessibility) computer control driver allows authorized AI agents to perform UI automation tasks, including window discovery, semantic interaction, and input simulation, making it suitable for Robotic Process Automation (RPA) and testing scenarios.
5. How does project recovery work in Starnet? Starnet employs a robust durable-store recovery system. It preserves the last known state of your station, projects, and conversation contexts. If a crash occurs, it recovers conversations and delegated work states, using ownership generation to prevent data races during restoration.