# Griptape

> Modular Python framework for building AI agents, pipelines, and workflows with chain-of-thought reasoning, tools, and memory — open source under Apache 2.0.

Griptape is an open-source Python framework developed by Griptape, Inc. (now part of Foundry) for building generative AI applications including agents, pipelines, workflows, and RAG implementations. The core library is available on GitHub under the Apache 2.0 license and is complemented by Griptape Cloud (a managed hosting layer) and Griptape Nodes (a visual no-code desktop builder for creatives). The latest release as of June 2026 is v1.10.3.

## What It Is

Griptape is a developer-first Python framework in the AI agent and workflow category. It provides clean abstractions for composing LLM-powered applications without requiring deep prompt engineering expertise. The framework's homepage describes its goal as letting "any developer become an AI developer" by keeping business logic in predictable, programmable Python rather than relying on prompt-based control flow. Its Off-Prompt™ concept routes large or sensitive data through Task Memory rather than sending it directly to the LLM, which the project claims improves security, performance, and cost.

## Core Architecture

The framework is organized around a hierarchy of composable components:

- **Structures**: Agents (single-task), Pipelines (sequential tasks), and Workflows (parallel tasks)
- **Tasks**: Core building blocks that interact with Engines, Tools, and other components
- **Memory**: Conversation Memory (cross-interaction retention), Task Memory (keeps large/sensitive outputs off the prompt), and Meta Memory (additional metadata context)
- **Drivers**: Swappable integrations for LLMs, embeddings, vector stores, file systems, SQL databases, image generation, text-to-speech, audio transcription, web search, web scraping, and observability platforms
- **Engines**: Higher-level wrappers for RAG pipelines, JSON/CSV extraction, text summarization, and output quality evaluation
- **Tools**: Built-in and custom capabilities that LLMs can invoke to interact with data and external services

## LLM and Provider Support

Griptape's driver model is designed for provider interchangeability. The GitHub README and topics reference OpenAI (GPT models), Anthropic (Claude), and HuggingFace integrations. Prompt Drivers manage textual and image interactions, while Assistant Drivers support various "assistant" service APIs. Embedding Drivers and Vector Store Drivers handle the retrieval side of RAG pipelines, and Rerank Drivers improve search result relevance.

## Update: v1.10.3

The repository shows active development with the latest release v1.10.3 published on June 16, 2026. The project was created in January 2023 and has accumulated over 2,500 GitHub stars and 241 forks as of mid-2026. The repository's last push was June 16, 2026, indicating ongoing maintenance. The project uses Semantic Versioning and publishes releases to PyPI.

## Ecosystem: Framework, Cloud, and Nodes

Griptape ships as three distinct products under one platform:

- **Griptape Framework**: The open-source Python library installable via PyPI (`pip install griptape`)
- **Griptape Cloud**: A managed cloud layer for hosting and operating ETL pipelines, RAG structures, and serverless application runtimes — skipping infrastructure management
- **Griptape Nodes**: A visual drag-and-drop desktop application targeting creative professionals (artists, filmmakers, designers) who want to build AI workflows without writing code; it also supports Python scripting for users who want to go further

## Team Background

The founding team brings backgrounds from AWS and the creative industry. CEO Kyle Roche previously served as GM at AWS and founded 2lemetry (acquired by Amazon in 2015). Product lead Jason Schleifer co-founded Nimble Collective (acquired by AWS in 2019) and has nearly 30 years in animation and tech. The engineering leadership also draws heavily from AWS distributed systems experience, with a dedicated Head of Open Source role reflecting the project's community focus.

## Features
- Agents, Pipelines, and Workflows as composable Python structures
- Conversation Memory, Task Memory, and Meta Memory
- Off-Prompt™ architecture for keeping sensitive data out of LLM context
- Swappable Drivers for LLMs, embeddings, vector stores, file systems, SQL, and more
- Built-in RAG Engine with modular retrieval-augmented generation pipelines
- Extraction Engine for JSON and CSV from unstructured text
- Summary Engine and Eval Engine for text quality scoring
- Image generation, text-to-speech, and audio transcription drivers
- Web search and web scraping drivers
- Observability drivers for trace and event data
- Built-in and custom Tool support for LLM-callable capabilities
- Rulesets for steering LLM behavior with minimal prompt engineering
- Loaders, Artifacts, Chunkers, and Tokenizers for data handling
- Griptape Cloud for managed hosting and serverless runtime
- Griptape Nodes visual no-code desktop builder for creatives

## Integrations
OpenAI, Anthropic (Claude), HuggingFace, DuckDuckGo Web Search, AWS, PyPI, Griptape Cloud, Griptape Nodes

## Platforms
WEB, API, DEVELOPER_SDK, CLI

## Pricing
Open Source

## Version
v1.10.3

## Links
- Website: https://www.griptape.ai
- Documentation: https://docs.griptape.ai/stable/
- Repository: https://github.com/griptape-ai/griptape
- EveryDev.ai: https://www.everydev.ai/tools/griptape