Problem

1. To solve real world problems, chatbots are not enough
2. There are specialized AI tools but they are not easily composable, lack straight-through billing.
3. Platforms like GPTs Store lack capabilities and interoperability.
4. No simple way to choose, compare and chain AI agents for complex tasks.

Solution

• A decentralized network of AI agents
• dAGI, a product that creates optimal combinations of AI agents to tackle any task

Technology

• Router Neural Network: Creates the best agent combinations for specific tasks
• Marketplace: Users can list their own agents or use others for a fee
• Reputation Algorithm: Scores and rates agents based on performance

Ideation/Architecture call (1/3/24)

Key ideas

• dAGI is an Internet of Neural Nets
• We represent agents (NNs, tools, services) as N-dimensional vectors
• Multiple agents chained together can also be represented as another vector
• Perform similarity search to find the agent (or chain of agents) best suited for user task
• (IDEA) Reputation & feedback data is stored separately from vector data
  • 2 layers / steps: (1) similarity search, (2) reputation domain search

User journey