Introducing Cybernaut-1: Agentic Search using MCTS

Today I am proud to announce the release of Cybernaut-1. Cybernaut-1 combines our powerful hybrid-3 search algorithm with LLM-guided Monte Carlo Tree Search to deliver world class search results on difficult queries. Cybernaut-1 is available via our V2 Search API and our Python Package.

from nosible import Nosible

with Nosible(nosible_api_key="YOUR API KEY HERE") as nos:
  print(nos.search(prompt="Find technical blogs about Monte Carlo Tree Search"))

We trust Cybernaut-1 with our signals

Cybernaut-1 is what we call a high-trust agentic search algorithm. What that means is that Cybernaut-1 has direct access to the internal logic in NOSIBLE. Our recent blog – "The Road to Cybernaut-1: Rebuilding Search for AI" – goes into a lot of detail about what that internal logic encompasses.

Cybernaut-1 uses them to self-improve

Cybernaut-1 uses LLM-guided Monte Carlo Tree Search to iteratively construct a high-quality search that aligns with your given prompt. It balances exploration, exploitation, and inference cost by slowly moving from wide and shallow searches to narrow and deep searches. This approach is illustrated below:

So that you always get the best results

Next week, we will be open-sourcing a comprehensive set of evaluations showing how Cybernaut-1 as well as Hybrid-3, the algorithm it uses, consistently match or outperforms leading search engines, even as we continue expanding our web coverage (currently growing at ~20 million webpages per day).

In the meantime, we’d love for you to try it out. We are offering $10,000 in Cybernaut-1 credits to the first 20 AI startups that sign up early.

SPONSORED

Building AI and looking for better context? Let's chat!

P.S. Looking for more of a technical deep dive? Go check out our blog: "The Road to Cybernaut-1: Rebuilding Search for AI". Or, alternatively, check out:

1. Our Official Docs - how to use the Python package.
2. Our Python Package - simply pip install nosible.
3. Our Swagger API Page - For non-Python users.