Announcing Griptape

In 2022, the ReAct paper introduced a novel approach that synergizes reasoning and acting in Large Language Models (LLMs), such as GPT and BERT. This advancement bridged the gap between the ability of LLMs to reason, through prompting a chain of thought, and to act, by generating action plans to achieve goals. Consequently, hundreds of new projects emerged, exposing LLMs to APIs, knowledge bases, and data sources, and pushing the boundaries of possibility. New application architectures were developed, with initial efforts focusing on flexible frameworks that combine integrations into chains. However, these architectures are primarily suitable for hobbyists and small projects, lacking the core elements enterprises expect from middleware frameworks.

Businesses are adopting LLMs to enhance their operations in three ways. First, they are implementing conversational agents or chat-driven applications across various environments to streamline communications and customer service. Second, the copilot application model, popularized by GitHub, has expanded beyond just developers. Businesses are creating copilot or AI assistant capabilities that help improve efficiency in day-to-day task completion. Finally, businesses are beginning to develop autonomous agents that tackle more open-ended and ambiguous tasks within predefined constraints. Autonomous agents may be scheduled, event-triggered, or queue-driven.

Introducing Griptape, the first enterprise-grade opinionated Python framework that enables developers to fully harness the potential of LLMs while enforcing strict trust boundaries, schema validation, and activity-level permissions. Griptape can be used to create conversational, copilot, and autonomous agents. Its core design tenet is to maximize the REasoning and enforce ACTing capabilities of LLMs, allowing developers to unleash the LLMs’ reasoning potential while adhering to strict policies regarding their ability to take action.

Griptape offers developers the ability to build AI systems that operate across two dimensions: predictability and creativity. For predictability, software structures like sequential pipelines or directed acyclic graphs (DAGs) are enforced. Creativity, on the other hand, is facilitated by safely prompting large language models (LLMs) with tools that connect to external APIs and data. Developers can move between these two dimensions according to their use case. For instance, if they need control over the order of execution, they may choose to use a pipeline with one tool per task. If they need to build an agent that can handle ambiguity, they can incorporate more tools into a single task.

Griptape is venture-funded. We are dedicated to maintaining our open source framework and will be building a business around managed offerings based on Griptape. If you’d like to contribute, visit our GitHub page. We are currently seeking to hire Python developers with LLM application experience.