Alter

Alter lets anyone create an AI digital twin that represents their expertise, thinks step-by-step, and evolves over time. Each twin is guided by structured prompts, example-based training, and a transparent knowledge base. Using an OpenClaw-style reasoning loop, the AI can use tools, remember past interactions, and generate structured outputs. Instead of static chatbots, Alter enables dynamic, identity-driven AI agents that continuously improve through memory and interaction.

Alter is built as a monorepo with a Next.js frontend and a Fastify backend. AI agents run on an OpenClaw-style execution loop, enabling step-by-step reasoning, tool usage, and structured outputs. We use 0G Compute for inference and 0G Storage to persist agent configuration, memory, and knowledge, allowing agents to evolve over time. ENS is used for identity resolution and metadata, while iNFTs (ERC-721 based) represent ownership and composability of agents on-chain. We implemented a lightweight RAG system by storing knowledge in 0G and retrieving it during inference via custom tools, avoiding heavy vector databases. A key hack was building a memory DAG where each interaction updates a new root, making agent evolution traceable and verifiable.