Gestell Documentation

View the API Reference

Contact us For guides, support and more.

Introduction

Gestell enframes the world

Gestell offers accurate, scalable and customizable ingestion, structuring and search processes for building AI Applications

Gestell can be broken down into Frames, Categories, Canons and Collections

Frames natural language structuring

Frames are the basic unit of data structuring for Gestell. A frame is a natural language structuring of an unstructured data, created by an LLM. Here, an LLM evaluates the contents of uploaded data, extracts the meaning from it and structures it into distinct natural language outputs - 'frames'. The key pieces of information that make up data are semantically summarized and outputted to form a frame.

For natural language processing to work best, it requires natural language structuring - Gestell's frames. Lacking data structuring, an LLM will attempt to generate results for users based upon its best guess. This best guess is prone to error, resulting in faulty or hallucinated outputs. Gestell's frames exist to unlock and empower the best machine understanding

Canons principles of relation

Frames don't exist in the ether, they are rather organized into Canons. Canons are mapped and structured hierarchies of frames. Canons dictate the overall structure of the results of Gestell. Frames are canonized to create maps of meaning. A canon can be crawled by an LLM to service results in the fastest and most effective way.

Collections contents of judgement

Collections is user uploaded data, canonized and enframed, ready for output and usage. Collections are the primary way users are going to interact with Gestell. The best way to get started with Gestell is to create an account. In order to get acquainted, we also recommend to read Using Gestell

Why Gestell

Scalable

Your enterprise needs, from gigabytes to petabytes, Gestell can serve as your enframing platform. Thanks to the scalable and flexible focus of Gestell's architecture - any multimodal ingestion needs are covered. We've seen that Gestell truly shines at large-scale applications, especially those where you have complex retrieval needs

Customizable

Gestell gets your ingestion, structuring and search processes working, on whatever your specific need might be. Gestell is designed to be incredibly generalizable across domains and across use cases, however, you also have powerful tooling to customize your data structuring. Utilizing Categories we give you full control for how your data is enframed, indexed and canonized. With our library of documentation and guides, we will help you learn how exactly you can use Gestell

Pragmatic

Gestell is fundamentally a pragmatic tool focused on suiting your real-world case. As Gestell becomes more ingrained in your workflows, it should get better at the specific tasks you need it to do. Gestell improves and scales alongside your usage of it. We also have a number of workflow specific toolings to get Gestell easily embedded in your workflows

Architecture Concepts

The meaningful objects... among which we live are not a model of the world stored in our mind or brain; they are the world itself
- Hubert Dreyfus

Gestell presents a fulsome solution for Data Enframing and Disclosure. One challenge presented by most ingestion, structuring and search processes today is that they are piecemeal, putting together separate parts from various disjointed entities that eventually result in subpar retrieval. Gestell proposes that since search and reasoning itself is a fulsome process across databases, the structuring and setting up of the environment for this search and reasoning ought to also be fulsome - we call this structuring ‘Enframing’ and this search and reasoning ‘Disclosure’. There is fundamentally no separation from the quality of the search and reasoning process from the structure of the database and presentation of the data itself. LLM results are the expression of the interaction of the LLM with the database. From document enframing to canonization, Gestell allows the user to set up the optimal environment for the entire Disclosure process

Pre-Processing

Pre-Processing efforts derive from strategizing and contextualizing. Gestell collects guidance from the user as to what sort of structuring is required, and if there are any Categories that should be reflected. This guidance and context serves to inform the creation of the strategy for processing: what modalities are utilized, the sort of frames that are required to be constructed, the type of language or images that will be processed and the structure of those possible relations between frames. Nuance that traditionally wouldn‘t inform a model in data processing is made clear in the pre-processing efforts

Processing

The Processing of data is informed by the Pre-Processing strategies and contextualization. Holism of frames is achieved by first understanding the distinctive patterns of the optimal retrieval and the proper perception of significance in the data. In the example of a textual data, paragraphs might flow across pages, errors might be improperly scanned or concepts could be vague in their layout. All of these might jeopardize the proper processing of the data, however, Gestell circumvents this through the relation of Pre-Processing context and Processing execution

Post-Processing

Post-Processing then consists of the establishing of connections between frames such that they are categorized and canonized for optimal retrieval. Post-Processing serves to nullify the scaling problem of traditional search processes - reasoning becomes embedded in the structure of the relations of that data itself such that the best answer is readily available across dataset sizes. Computational efficiency results from maximizing relational awareness in context-dependent tasks. The structure of the dataset thus affords itself to the intentional arc of the Disclosure process. There is no ‘fixed’ database structure in Gestell but one rather contingent upon the parameters needed for retrieval in the first place

Agents

Thinking agents are operative throughout Gestell‘s entire database process. These agents follow user instructions and communicate to each other to make up the entire arc of processing. Through first analyzing the task environment, separate agents are deployed and consistently reach back to the initial tasking agent to update and inform the task outputs. The primitive facts of the data and the simple routine of database construction are folded onto one another again and again by these agents to make a complete and complex circle of knowledge. This emergent knowledge structure of agent activity forms the representation of the data such that what traditionally would be a ‘rule’ in non-agent-based systems is rather a ‘heuristic’ enabling database complexity while minimizing computational effort

Disclosure

Gestell enables you to build your full Disclosure process in-app, allowing for search, retrieval and prompting. Each prompt is tested by agents for generalization and conceptualization across the database to minimize representational noise and overhead while still obtaining characteristic features and relations from the data. The database makes up the world of the Disclosure process. Citation and organizational sequencing allow for a grounded coherence and sociality within the database. The layers of facts of the world which would traditionally be intractable in complex retrieval rather become the resources and nodes of search agents to allow for proper Disclosure outputs

Pricing

Plans

Gestell offers two monthly or annual plans:

Save 15% when you subscribe annually

Want to try the preview first? Get Started for free.

Billing

Your card is billed monthly based on your plan and usage of credits. Each month, you are granted a certain number of credits according to your plan. If you use more credits than granted, we will charge you according to your plan‘s by credit cost. There are two types of credits - Processing Credits and Prompt Credits

Processing

Processing Credits are a standardized unit following the below parameters:

Prompting

Prompt Credits are a standardized unit following the below parameters:

Reprocessing

For Reprocessing jobs, credits are calculated according to the sort of reprocessing you utilize:

Overage Usage

If you use more credits than granted, we will charge you according to your plan‘s by credit cost. The per credit costs are defined below:

Frequently Asked Questions

What is Gestell?

Gestell is a new ETL architecture for LLMs, from Enframing to Disclosure

What tools does Gestell replace for me?

Gestell can replace your current data ingestion tools (like Google‘s Document AI), database and vectorization tools (like Chroma), and your RAG and prompting tools (like Amazon Bedrock)

What common use cases can I accomplish with Gestell?

With Gestell you can:

Ingest and enframe your data

Canonize and structure your data

Prompt and retrieve across your data

How do you calculate credits?

Credits are standardized units across processing and search, we recommend reading the Billing portion of our docs for more info

What if I go over my granted number of credits?

If you use more credits than granted, we will charge you according to your plan‘s by credit cost. See Billing for more information

Do you keep my data?

Gestell doesn‘t keep your data. Your data is entirely yours. Read our Privacy Policy and Terms of Service for more information

Is my data secure?

Your data in encrypted at both rest and in transit