Yellow Cursor

Yellow Cursor is a specialized web code editor and AI-driven integration engine designed to make blockchain payments invisible to the developer. By combining a Retrieval-Augmented Generation (RAG) pipeline with autonomous coding agents, it allows anyone to add Yellow Network streaming payments to a GitHub repository through a single natural language prompt.

Who is it for?

->Crypto dApp Builders: Focus on your core logic while the agent handles gas-fee reduction and streaming payment implementation.

->Web2 Developers: Access international, low-cost crypto payments without needing to learn blockchain fundamentals or complex SDKs.

Yellow Community: Rapidly prototype and deploy projects by letting the AI handle the repetitive integration heavy lifting.

Proof of Concept: Built Projects

->Yellow Flappy Bird: We took a standard game repo and used the agent to inject code that rewards players with 0.001 ETH per pipe crossed. The AI utilized the Yellow SDK and Nitrolite protocol to create instant, real-time payment channels.

->True Pay-As-You-Go AI Wrapper: Traditional AI providers charge upfront credits. Our agent refactored a standard API wrapper into a streaming model. It tracks token usage in real-time and settles payments only when the chat ends, ensuring fair, per-token pricing

The system is built as a Microservices-based Web App using LangGraph to manage complex, multi-stage reasoning and coding tasks.

1. RAG-Powered Intelligence Unlike general-purpose coding assistants, Yellow Cursor uses a dedicated RAG pipeline equipped with the latest, and even confidential, Yellow Network documentation.

Context Advantage: Because it is a web app, the coding agent stays updated with the most recent SDK changes and documentation patches in real-time.

Code Analysis: The agent reads your entire project structure to identify specific payment functions and target integration points.

1. Autonomous Coding & Error Correction The system doesn't just "guess" code; it follows a rigorous Plan-Act-Verify workflow:

Architecture Planning: The AI creates a strategic blueprint based on the user's prompt (e.g., "reduce gas fees") and the retrieved docs.

Tool Calls: To prevent hallucinations, the LLM uses specific tool calls to interact with the file system and build tools.

Self-Healing Loop: The agent automatically builds the project and runs tests. If errors are detected, it triggers an Error Analysis node to diagnose the root cause, refactors the plan, and writes corrected code until the build succeeds.

1. Developer Dashboard The web interface includes a dashboard that tracks your connected GitHub repositories, manages configuration files, and provides a history of AI-generated refactors.