GantryGraph

10 or later on macOS, Linux, and Windows. It enables developers to create agents capable of perception, tool use, browser automation, and memory, supporting integration with large language models from Anthropic, OpenAI, or any LangChain-compatible provider. The framework is positioned as LangGraph-native and comes with enterprise guardrails built in.

The tool allows users to quickly set up agents that can perform tasks such as querying the current working directory, interacting with web pages, and extracting information from browser sessions. Features demonstrated include the ability to add browser-based perception, use browser tools to interact with websites, and stream agent events in real time for integration with user interfaces. Agents can be extended with additional tools, security guardrails, and event streaming capabilities. The agent loop follows an observe, think, act, and review cycle, and the framework encourages adding guardrails before granting agents wider access.

GantryGraph is delivered as a Python package installable via pip, with optional extras for browser and search capabilities. It integrates with Playwright for browser automation and supports models accessed through API keys from Anthropic, OpenAI, or LangChain-compatible providers. The framework is distributed under the MIT License, allowing users to extend and modify it as needed. Documentation, API references, and community resources are available to support developers working with the tool.