Add support for OpenAI-compatible API endpoints

[joaopalmeiro]

Closes #57

Test plan

Install Ollama (if necessary).

Start Ollama:

ollama serve

Download an LLM (Gemma 3 270M or another):

ollama pull gemma3:270m

Install the documentation dependencies:

npm run build

npm run docs:install

Launch the documentation:

npm run docs:dev

Select Option A (Automatic Descriptions) and fill out the form with the following values:

• API key: ollama
• Model: gemma3:270m
• Base URL: http://localhost:11434/v1/ or http://localhost:11434/v1

Since the page contains 8 charts, it will take a bit to get all the descriptions. Monitor them from the Ollama logs. Additionally, since models like Gemma 3 270M are small models (sometimes designed for later fine-tuning), expect some "bad descriptions."

Examples

Ollama logs

gemma3:270m

qwen2.5:0.5b

Considerations

• Following the sort-of convention of specifying a dummy API key even when using local models (a result of using SDKs like OpenAI's, which have a required API key parameter, for example), I kept the API key as a required prop.
• For now, if a reasoning model is used and the output includes its thinking output (between <think> tags, as in Qwen 3, for example), the entire output is added to the chart description. Future work is needed to properly accommodate this type of models.

References

• https://platform.openai.com/docs/api-reference/chat
• https://github.com/openai/openai-node/blob/4dc2e234f015f45ccd82212694995e6fe0e915f1/src/client.ts#L240-L245
• https://github.com/openai/openai-node/blob/4dc2e234f015f45ccd82212694995e6fe0e915f1/src/client.ts#L372
• https://github.com/ollama/ollama/blob/main/docs/openai.md

TODO

• Add model ID and base URL props to example components
• Update error messages
• Add base URL param
• Test automatic descriptions with Ollama