This project provides an AI agent framework designed as a "Pinix Clip," offering an agentic loop with integrated memory, tool use, and vision capabilities. It targets developers and power users seeking a robust, extensible agent architecture that can be easily deployed and managed within the Pinix ecosystem, benefiting from asynchronous execution and multimodal understanding.

How It Works

The agent operates via an LLM-driven loop, managed by the Pinix framework. Core components include Topics for namespaced conversation history stored in SQLite, and Runs representing individual agentic cycles. Memory is layered, comprising persistent Facts, LLM-generated Summaries with embeddings, and semantic search capabilities. Vision is integrated by automatically capturing browser screenshots, attaching them as multimodal input to LLM calls, allowing the agent to perceive its environment. The system supports Clips, which are external services invoked via a dedicated command, enabling extensibility and cross-service file transfer, including support for Edge Clips. Command chaining (|, &&, ;) and a comprehensive set of built-in tools for file I/O, memory management, and browser control are central to its operation.

Quick Start & Requirements

Local development is initiated with make dev, which builds a macOS binary and initializes data. Frontend dependencies are managed via pnpm install within the ui directory. An API key must be added to data/config.yaml. For deployment, make deploy enables workdir mode for live changes, while make package creates a distributable .clip file for Pinix. Installation uses the Pinix CLI: pinix clip install dist/agent.clip. Prerequisites include Go, pnpm, the Pinix CLI, and a configured LLM provider API key.

Highlighted Details

• Extensive Tooling: Features built-in commands for file system operations (ls, cat, write, etc.), memory management (search, store, forget), topic management, browser control with automatic screenshotting, and command chaining.
• Vision Integration: Automatically captures browser screenshots, saving them and attaching them as vision content to LLM API calls, enabling visual context for agent decisions.
• Pinix Ecosystem Integration: Designed as a distributable "Pinix Clip" for straightforward installation, upgrades, and management via the Pinix CLI, supporting remote deployment.
• Dual Interface: Supports both a raw CLI output and a web-based UI rendering JSONL data.

Maintenance & Community

No specific details regarding maintainers, community channels (e.g., Discord, Slack), or roadmap links are provided in the project description.

Licensing & Compatibility

The project's license is not explicitly stated in the provided information, which is a critical omission for assessing commercial use or derivative works. Compatibility is primarily tied to the Pinix ecosystem.

Limitations & Caveats

Adoption is contingent on familiarity with and setup of the Pinix ecosystem. The project requires configuration of LLM API keys and specific provider settings. Crucially, the absence of explicit licensing information presents a significant adoption blocker, preventing clear understanding of usage rights and restrictions.