Griptape offers developers the ability to build AI systems that operate across two dimensions: predictability and creativity.

For predictability, software structures like sequential pipelines and directed acyclic graphs (DAGs) are enforced. Creativity, on the other hand, is facilitated by safely prompting LLMs with Griptape Tools that connect to external APIs and data sources. Developers can move between these two dimensions according to their use case.

Please refer to Griptape Docs for:

• Getting started guides.
• Core concepts and design overviews.
• Examples.
• Contribution guidelines.

First, install griptape:

pip install griptape -U

Second, configure an OpenAI client by getting an API key and adding it to your environment as OPENAI_API_KEY. By default, Griptape uses OpenAI Completions API to execute LLM prompts.

With Griptape, you can create structures, such as Agents, Pipelines, and Workflows, that are composed of different types of tasks. Let's build a simple creative agent that dynamically uses two tools with shared short-term memory.

from griptape.structures import Agent
from griptape.tools import WebScraper

agent = Agent(
    tools=[WebScraper()]
)

agent.run(
    "based on https://www.griptape.ai/, tell me what Griptape is"
)

And here is the output:

Q: based on https://www.griptape.ai/, tell me what Griptape is
A: Griptape is an opinionated Python framework that enables developers to fully harness the potential of LLMs while enforcing strict trust boundaries, schema validation, and activity-level permissions. It offers developers the ability to build AI systems that operate across two dimensions: predictability and creativity. Griptape can be used to create conversational and autonomous agents.

During the run, the Griptape agent loaded a webpage, stored its full content in the short-term memory, and finally queried it to answer the original question. The important thing to note here is that no matter how big the webpage is it can never blow up the prompt token limit because the content never goes to memory instead of the main prompt.

By default, Griptape uses OpenAI's gpt-4 to drive the core agent logic. Other framework components responsible for summarization, querying, and text extraction use gpt-3.5-turbo. All of those are customizable through prompt drivers. For example, if you don't have access to gpt-4, you can change the quick start snippet like this:

from griptape.drivers import OpenAiChatPromptDriver
from griptape.structures import Agent
from griptape.tools import WebScraper

agent = Agent(
    prompt_driver=OpenAiChatPromptDriver(
        model="gpt-3.5-turbo"
    ),
    tools=[WebScraper()]
)

agent.run(
    "based on https://www.griptape.ai/, tell me what Griptape is"
)

If you are running into gpt-4 rate limiting issues, specify a custom number of max_tokens in the driver:

OpenAiChatPromptDriver(
    model="gpt-4",
    max_tokens=200
)

Check out our docs to learn more about how to use Griptape with other LLM providers like Anthropic, Claude, Hugging Face, and Azure.

Griptape is in constant development and its APIs and documentation are subject to change. Until we stabilize the API and release version 1.0.0, we will use minor versions (i.e., x.Y.z) to introduce features and breaking features, and patch versions (i.e., x.y.Z) for bug fixes.

Contributions in the form of bug reports, feature ideas, or pull requests are super welcome! Take a look at the current issues and if you'd like to help please submit a pull request with some tests.

Griptape is available under the Apache 2.0 License.