SAGA AI Agents - Latest Release
SAGA’s latest release brings some important and helpful new features to tell you about. If you are new to the project, SAGA is Skill -> Action Generation for (AI) Agents and it's open-source. You can read more about how it works here. It had an important refactor since last release that allows for more features and more flexibility. We’re growing a community around Multi-Agent Simulations, so your input is welcome! Check it out on Github and let us know in the issue queue how it could be even better!
New Features
Improved Space-Colony Demo
The space-colony demo allows you to run a simple text-based simulation of a spaceship crew within SAGA. Instead of interacting with a Sim over the network, it demos the use of fable-saga directly as a python library. Here’s an example of it generating options for the ship’s captain.
Conversations
SAGA now generates Conversations as well as Actions. That means your sim can do more by leveraging SAGA instead of having to do it yourself. It also means the built-in demo sim, “Space Colony” will now generate conversations if you want to test that out. In case you missed it, the space-colony demo allows you to run a simple text-based simulation of a spaceship crew.
SAGA is still stateless when it comes to Actions and Conversations, so you need to supply it the appropriate context it needs.
New Ways to Connect to SAGA
Now in addition to Socketio, you can connect via Websockets and a REST HTTP. Just choose the option on startup.
Conversations are their own Endpoint as well as the new Embeddings endpoint.
Embeddings and Vector Store
SAGA now provides a helpful endpoint for embeddings. You can use it to generate embeddings as well as store them and use similarity search for retrieval. Very useful if you want to experiment with RAG or Retrieval Augmented Generation techniques and cut down on your context size.
Token Usage Reporting
Token usage is now report back to the client. This allows tracking how much you are spending per session based on the model you choose to use. For instance we are using it to show a dollar amount in the Thistle Gulch UI.
Choose Your Own Model
SAGA uses OpenAI by default. You can now specify which OpenAI model you want to use per-request. A good example is action generations using GPT-4, which are typically much better than the default GPT-3.5-turbo, but only doing that for specific characters or at a specific time to save on costs.
SAGA also builds upon LangChain, allowing you to swap out OpenAI completely for another model they support like Claude or Ollama to run a local model.
