Transforms all files in specified directory using a large language model. You can specify conditions for file transformation along with a prompt, which LLM to use and which model should be used . The transformed files are saved in a new directory. The file structure is preserved.

docker run -v $(pwd):/data ghcr.io/blwsh/llmt analyze

Download the latest release from the releases page. Extract the archive and run the binary.

llmt \ --config <config_file> \    # optional parameter, default is config.yaml in current directory
      analyze ./myProject ../docs  # analyzes files in ./myProject and outputs them as markdown in ../docs (maintains file structure)

Example config.yaml file:

#$schema: https://raw.githubusercontent.com/blwsh/llmt/main/schema.json
version: "0.1"

analyzers:
  - prompt: Write docs for this file
    analyzer: openai
    model: gpt-4o-mini
    regex: ^. \.php$
    not_in:
      - vendor

See schema.json for the full config schema.

You can find a comprehensive list of examples here. Below is a simple example which has similar behaviour to the command line analyze command.

package main

import (
	"context"
	"io/ioutil"
	"strings"

	"github.com/blwsh/llmt/pkg/analyzer"
	"github.com/blwsh/llmt/pkg/analyzer/item_analyzer/openai"
	"github.com/blwsh/llmt/pkg/analyzer/project_analyzer/chat"
)

func main() {
	chat.New().
	  AnalyzeProject(context.Background(), "./myProject", "../docs", []analyzer.FileAnalyzerConfig{
			{
				Prompt:    "document this files behaviour",
				Analyzer:  openai.New("OPENAI_TOKEN_HERE", "gpt-4o-mini"),
				Condition: func(path string) bool { return strings.HasSuffix(path, ".php") },
				ResultHandler: func(destFilepath string, result string) error {
					return ioutil.WriteFile(destFilepath, []byte(result), 0644)
				},
			},
		})
}

All contributions welcome! For the next release I want to offer more analyzers with greater configurability. I also plan on adding more hooks for file analyzers to allow for more complex transformations.