Agent framework

​

Problems with existing frameworks

While our team built Daige, we tried every major Web3 agent framework and ran into well-known issues most of them have:

1. Rigid & Complex: Too many layers of abstraction, making them hard to debug and extend.
2. Workflow-Focused: They emphasize workflows over autonomous, goal-solving agents.
3. Overgrown & Hard to Customize: Production often requires rewriting the whole framework from scratch.
4. Weak MVP Focus: They can spin up basic agents quickly but struggle with truly powerful ones.

We knew creating a world-class agent framework from scratch would take serious effort and R&D. That’s why our solution is built on smolagents from Hugging Face—a lightweight, powerful library that enables autonomous agents.

​

Galadriel framework

We believe that agent development requires a robust, vertically integrated stack — a strategy proven by successful ecosystems such as Swift & iOS and Rust/Anchor & Solana. Following this principle, Galadriel’s framework is highly optimized for and tightly integrated with Galadriel L1.

The framework’s design and development are guided by the following core principles:

1. Simplicity - Minimal abstractions, with ~1,000 lines of agent logic.
2. Autonomous Agents - Enables agents to plan and control their own workflows.
3. Broad Tool Support - Integrates with LangChain, Anthropic’s MCP, and more.
4. Web3 Functionalities - Provides out-of-the-box support for on-chain and off-chain tools, wallet integration, and payments.
5. Most used Clients - Integrates your agent with X, TG, Discord, and other main clients.
6. Model-Agnostic – Works seamlessly with any LLM, whether local or hosted.
7. Modality-Agnostic – Supports text, vision, video, and audio inputs.

Framework architecture: