feat: add LLM suggestions to CLI

[ryan-timothy-albert]

A sample CLI that communicates with downstream speakeasy API and LLM server.

Normal Testing (CLI)

1. Make sure you have OPENAI_API_KEY set.
2. go run main.go validate openapi --schema openapi-invalid.yaml --fix

Testing with Local API
Local Testing

cd speakeasy-registry (API)
1 Run local registry API
2. Make docker-llm (wait until Chroma is up to continue, you should see Uvicorn running on http://0.0.0.0:8000)
3. export SPEAKEASY_SERVER_URL=http://localhost:35290

[chase-crumbaugh]

Overall looking good! A couple comments:

• You mentioned moving over to use promptui--I would definitely recommend it! It's super easy to use and works great.
• I think it would good if this was also accessible through a higher-level CLI command, e.g. speakeasy suggest -s openapi.yaml. It will read better in an article IMO and be more discoverable in the CLI instead of buried as an optional flag. Can leave it accessible via --fix as well though

[ryan-timothy-albert]

Overall looking good! A couple comments:

• You mentioned moving over to use promptui--I would definitely recommend it! It's super easy to use and works great.
• I think it would good if this was also accessible through a higher-level CLI command, e.g. speakeasy suggest -s openapi.yaml. It will read better in an article IMO and be more discoverable in the CLI instead of buried as an optional flag. Can leave it accessible via --fix as well though

I think a higher-level CLI command is reasonable. In your mind would this CLI command automatically run validate, then suggest fixes based on that validation? Same behavior just set at a higher level

[ryan-timothy-albert]

Overall looking good! A couple comments:

• You mentioned moving over to use promptui--I would definitely recommend it! It's super easy to use and works great.
• I think it would good if this was also accessible through a higher-level CLI command, e.g. speakeasy suggest -s openapi.yaml. It will read better in an article IMO and be more discoverable in the CLI instead of buried as an optional flag. Can leave it accessible via --fix as well though

I think a higher-level CLI command is reasonable. In your mind would this CLI command automatically run validate, then suggest fixes based on that validation? Same behavior just set at a higher level

Done with promptui

[chase-crumbaugh]

nit/question: Are these suggestions technically being generated by ChatGPT? Or is it simply GPT3?

[ryan-timothy-albert]

GPT3.5 which I believe is a chat model

[chase-crumbaugh]

Some very minor things then gtg!