Skip to content

Kelvin Github Coding Agent

AI-Accelerated Development with the Kelvin SDK

Kelvin SmartApps™ is optimized for Agentic Workflows.

By embedding specific instruction sets for AI assistants, like GitHub Copilot, Claude, and Cursor, into our kelvin app create structure we enable the agent to understand our SDK and Kelvin Platform structure, reducing hallucinations and making natural language development fast and precise.

Note

We recommend using GitHub Codespace or the Copilot extension in Visual Studio Code.

What is an Agentic Workflow?

Unlike standard autocomplete, an agentic workflow involves an AI "agent" that understands the context of your entire project. When you use the Kelvin SDK, these agents don't just guess code—they follow our pre-defined Instruction Files to:

  • Implement Features: Write boilerplate-free code using the latest SDK syntax.
  • Validate Logic: Automatically generate tests that match our internal standards.
  • Self-Correct: Debug SDK-specific errors by referencing our "Known Patterns" documentation.

Key AI Integration Files

To ensure your AI assistant is "SDK-aware," we have included the following configuration files in the repository root:

File Supported Agent Purpose
.github/copilot-instructions.md GitHub Copilot Provides specific rules on Kelvin class structures and preferred API methods.
CLAUDE.md Claude / Claude Code Defines the project architecture and coding style guidelines for high-level reasoning.
AGENT.md Generic / Custom Agents A universal instruction set providing the "source of truth" for the SDK's logic and constraints.

Getting Started

  1. Create the project using one of two agent-optimized methods:
    • Kelvin UI: Select Create SmartApp to launch a pre-configured GitHub Codespace. This environment automatically initializes the required instruction files, ensuring your AI agent is instantly "SDK-aware."
    • Kelvin CLI: Execute kelvin app create in your terminal to bootstrap the project structure locally, including the AGENT.md and configuration files necessary for natural language development.
  2. Open the project in your preferred AI-enabled IDE (VS Code with Copilot, Cursor, etc.).
  3. Verify the instructions: The IDE will automatically detect the configuration files listed above.
  4. Prompt with intent: Instead of writing code, ask the agent: "Use the Kelvin SDK to create a new authenticated session and list all active resources."

Note

Pro Tip: When using Copilot Chat, you can reference our specific implementation guide by typing # followed by the filename of our SDK examples to provide the agent with perfect context.

In this interactive demonstration below we ask GitHUb Copilot the following;

Create an event detection app that will monitor two data streams in real time and if the difference between the two is more that 5% then change the setpoint of the third data stream up by 10%.

Now click on Get Started button below to see for yourself in this interactive tour how GitHub Copilot works.