What it does Drop a list of topics into topic.txt, your OpenAI token into token.txt, and run ./cgp. The tool queries GPT for each topic, formats the response into slides, and writes an HTML file. That’s the whole workflow. It supports multiple renderers (Remark is default) and an interactive mode (-i) that lets you approve or edit each generated slide before continuing.

The interesting bit The entire project is deliberately “all in one single file”—a Go binary you can drop anywhere without dependency hell. The architecture splits cleanly into pluggable clients (currently GPT-3.5) and renderers, so swapping in a different model or slide engine is meant to be trivial.

Key highlights

• Single static binary, no runtime dependencies
• Interactive review mode to catch GPT hallucinations before they reach your deck
• Multiple renderer backends (Remark, others shown in docs)
• Multi-language output supported
• MIT licensed, actively seeking PRs

Caveats

• Requires an OpenAI API token; no local model support mentioned
• The “all in one file” claim refers to the binary distribution, not the source architecture
• Changelog notes the project was rewritten in Go March 2023; earlier iterations may have been different

Verdict Worth a look if you frequently churn out informational slide decks and trust GPT to not embarrass you in front of an audience. Skip it if you need fine-grained visual control or work somewhere that frowns on cloud LLM dependencies.