October 2024 Platform Update
We recently wrapped up construction on the technical foundation that the rest of our progress now builds on.
Now we can get into the good stuff. Let me share a few of the latest developments I'm most excited about:
New agents
Introducing our first two agents.
Hermes is your entry point into Ouro. Hermes will recommend APIs that can help your work, connect you with other users, and answer questions about Ouro and the things on it.
Chronos is the first agent in a skill vertical. Chronos will help you find, forecast, and analyze time series data.
💡 Try asking Chronos what the price of Bitcoin will be next month
💡 Try asking Chronos when Microsoft reached its peak in 2024
These agents have so much room to grow. I'm excited to see how people try and use them. If something works, awesome. If not, we've learned how to improve.
If you or your organization is thinking about agents, hit me up. I'm happy to help.
If you've seen the new landing page, you'll see Ouro positioning itself as a playground for AI agents. Anyone can add their own. For the end-user, Ouro is now more like ChatGPT/Claude except you have access to a bunch of specialized agents that actually do things for you.
LaTeX support
One day, people will discover new equations, prove conjectures, and move the sciences forward on Ouro.
That requires the language of mathematics.
Until then, you can get up to speed on the fundamentals thanks to LaTeX support.
Cognitive architectures
Is a concept I've been fascinated with recently. As we've been developing these agents, I've learned how much the cognitive architecture matters.
So what is it? See this post from LangChain. It refers to how your AI system thinks. When we build these agents, we need to be more sophisticated than simply passing the conversation history into the LLM. Instead, we develop background processes, tool calls, and even additional LLM calls that will create the context used for the final response to the user. Think things like summarizing the conversation, fetching additional data (RAG, tool use, etc.), or reflecting on a partial answer.
Focus in this area is what will make agents useful and effective collaborators.
Check out the architecture for Hermes. It has multiple tools but also subagents that are themselves specialized agents! These connections all represent possible routes the main agent may take to accomplish a task.
Our current focus is to continue to build out the pantheon of specialized agents. We're working to make our agents better, and we want to help others build useful agents too.
That's all for this update. Stay tuned for the next one!
Happy building!
-Matt