Honcho is a platform for making AI agents and LLM powered applications that are personalized to their end users. It leverages the inherent theory-of-mind capabilities of LLMs to cohere to user psychology over time.
Read about the project here.
Read the user documentation here
The Honcho project is split between several repositories with this one hosting the core service logic. This is implemented as a FastAPI server/API to store data about an application's state.
There are also client-sdks that are created using Stainless. Currently, there is a Python and TypeScript/JavaScript SDK available.
Examples on how to use the SDK are located within each SDK repository. There is also SDK example usage available in the API Reference along with various guides.
Currently, there is a demo server of Honcho running at https://demo.honcho.dev. This server is not production ready and does not have an reliability guarantees. It is purely there for evaluation purposes.
A private beta for a tenant isolated production ready version of Honcho is currently underway. If interested fill out this typeform and the Plastic Labs team will reach out to onboard users.
Additionally, Honcho can be self-hosted for testing and evaluation purposes. See Contributing for more details on how to setup a local version of Honcho.
The functionality of Honcho can be split into two different services: Storage and Insights.
Honcho contains several different primitives used for storing application and user data. This data is used for managing conversations, modeling user psychology, building RAG applications, and more.
The philosophy behind Honcho is to provide a platform that is user-centric and easily scalable from a single user to a million.
Below is a mapping of the different primitives.
Apps
└── Users
├── Sessions
│ ├── Messages
│ └── Metamessages
└── Collections
└── Documents
Users familiar with APIs such as the OpenAI Assistants API will be familiar with much of the mapping here.
This is the top level construct of Honcho. Developers can register different
Apps for different assistants, agents, AI enabled features, etc. It is a way to
isolation data between use cases.
Users
Within an App everything revolves around a User. the User object
literally represent a user of an application.
The Session object represents a set of interactions a User has with an
App. Other application may refer to this as a thread or conversation.
Messages
The Message represents an atomic interaction of a User in a Session.
Messages are labed as either a User or AI message.
A Metamessage is similar to a Message with different use case. They are
meant to be used to store intermediate inference from AI assistants or other
derived information that is separate from the main User App interaction
loop. For complicated prompting architectures like metacognitive prompting
metamessages can store thought and reflection steps along with having developer
information such as logs.
Each Metamessage is associated with a Message. The convention we recommend
is to attach a Metamessage to the Message it was derived from or based on.
At a high level a Collection is a named group of Documents. Developers
familiar with RAG based applications will be familar with these. Collections
store vector embedded data that developers and agents can retrieve against using
functions like cosine similarity.
Developers can create multiple Collections for a user for different purposes
such as modeling different personas, adding third-party data such as emails and
PDF files, and more.
As stated before a Document is vector embedded data stored in a Collection.
The Insight functionality of Honcho is built on top of the Storage service. As
Messages and Sessions are created for a User, Honcho will asynchronously
reason about the User's psychology to derive facts about them and store them
in a reserved Collection.
To read more about how this works read our Research Paper
Developers can then leverage these insights in their application to better
server User needs. The primary interface for using these insights is through
the Dialectic Endpoint.
This is a regular API endpoint that takes natural language requests to get data
about the User. This robust design let's us use this single endpoint for all
cases where extra personalization or information about the User is necessary.
A developer's application can treat Honcho as an oracle to the User and
consult it when necessary. Some examples of how to leverage the Dialectic
API include:
- Asking Honcho for a theory-of-mind insight about the
User - Asking Honcho to hydrate a prompt with data about the
Users behavior - Asking Honcho for a 2nd opinion or approach about how to respond to the User
Honcho is licensed under the AGPL-3.0 License. Learn more at the License file