What it does

susi_shell is a bash-driven front-end for AI services. It pipes your terminal input to the OpenAI API and returns answers, completions, translations, or transcriptions. Think of it as a thin wrapper around remote models with a Unix-philosophy interface: echo "hello world" | ./susi_shell.sh ask gpt-4.

The interesting bit

The project explicitly positions itself as “similar to ollama” but currently lacks the local-model part that makes Ollama useful. The roadmap promises self-hosted services via the companion susi_api project, which is where the actual heavy lifting is supposed to happen. For now, it’s essentially a curl-with-delusions-of-grandeur.

Key highlights

• Interactive chat mode via ./susi_shell.sh run <model-name>
• Batch mode accepting stdin for scripted pipelines
• Audio pipeline: listen → transscript (note the charming double-s spelling)
• Text-to-speech via say, macOS-only by default
• Explicitly designed to migrate from OpenAI dependency to local inference

Caveats

• Every command listed is “to be implemented” — the README is aspirational, not descriptive
• Currently requires OpenAI API access; local models are future work in a separate repo
• say command is macOS-only, limiting cross-platform utility

Verdict

Worth watching if you’re invested in the susi_api ecosystem or want a template for shell-based LLM tooling. Skip it if you need working local models today — Ollama, aichat, or shell-gpt already exist and actually ship their features.