lightcode

Lightcode

Lightcode is a light weight coding agent written in Go.

Requirements

• Go 1.25+
• At least one OpenAI-compatible endpoint configured (see Models below)

Configuration

Settings live under ~/.lightcode/. The app creates this directory and a default config.json on first run.

Models

Providers must speak the OpenAI Chat Completions API. Add entries to the models array in ~/.lightcode/config.json. Each entry is an object with:

Example:

{
  "theme": "light",
  "skills_path": "choose_any_skill_path"
  "port: "8000"
  "providers": [
    {
      "models": ["another-model-id"],
      "base_url": "https://your-gateway.example/v1",
      "api_key": "..."
    },
    {
      "models": ["another-model-id-2]",
      "base_url": "https://your-gateway.example/v1",
      "api_key": "..."
    }
  ]

• In the TUI, run /models to select one of the models

Skills (greatly improves performance)

Put skills in ~/.lightcode/skills/, each in its own subdirectory containing a SKILL.md file.

Server Port

The HTTP API defaults to port 8080. To use a different port, Set port in ~/.lightcode/config.json.

go run ./cmd/lightcode/main.go

Quick start

Run the API server (by default listens on :8080) and TUI:

go run ./cmd/lightcode/main.go

Todo

• copy paste multiple lines into a [ paste #1 13 lines ]

• better tool and thinking formating

• Skills

• grep tool

• first make the ui work

• UI upgrades

• Make config files

• improve tools and make test

• Fix the database bug

• Limit accessible directory to working dir

• question tool - homepage 381, just need to create a ui and send chat completion request

• Show Code changes

• Plan mode - prompt and tool filter

• todo tool - handle in the ui and send it as context in agent.go

• json data for model selection etc

• need to also integrate anthropic go sdk cause response format for tool calling in models like glp and claude (fixed without adding anthropic api)

• Add a nicer way to use multiple agents like plan, build etc.

• File tracker

• Check for credentials before making an api call in .env

• queue

• handle errors