EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·Aug 21·5 min read
Seed: Minimal, self-modifying agent harness
TL;DR
A seed agent is a minimal, self-modifying framework that connects a language model to a single tool, allowing for dynamic prompt management. This innovative approach simplifies agent development by eliminating the need for complex frameworks.
✦ Why It Matters
Consider experimenting with seed agents to streamline your AI project development today.
Key Takeaways
- Clone the Seed repository to start building your agent.
- Use the REPL to interact with your agent in real-time.
- Ensure important data is saved in the self/ directory after each session.
- Experiment with different language models using the llm library.
How It Works
Seed operates by allowing users to create agents that evolve through interaction. The core file, seed.py, connects a language model to a tool (exec) and manages the agent's system prompt.
Each agent's growth is recorded in a self/ directory, where it can store tools and session data, enabling personalized development.