Skip to main content

OpenAI Provider

💡TL;DR

OpenAI uses the shared OpenAI-compatible transport. The current preset defaults to gpt-4o, https://api.openai.com/v1, temperature: 0.5, and the models-then-chat connection test. Use this page when you need exact field semantics; use the LLM Providers overview for category-level selection.

This is part of the Obsidian AI Knowledge Management Guide.

Setup​

Create an OpenAI API key, add an OpenAI provider profile in Notemd settings, and keep the default Base URL unless you are intentionally pointing at an OpenAI-compatible gateway. The model field can stay on gpt-4o for general work or be replaced by another OpenAI model ID that your account can call.

Use task-specific models for cost control: keep stronger models for research, concept extraction, and long-form restructuring; use cheaper or faster models for translation, link suggestions, and short summaries.

Endpoint And Authentication​

FieldCurrent preset
Transportopenai-compatible
API keyRequired
Base URLhttps://api.openai.com/v1
Default modelgpt-4o
Temperature0.5
Connection test/models, then /chat/completions

Notemd normalizes OpenAI-compatible base URLs before appending chat/completions or models. Do not include a trailing /chat/completions in the Base URL; that creates a duplicated path.

Model Discovery​

OpenAI uses the generic OpenAI-compatible model discovery path. The settings UI first asks /v1/models, then validates the selected model with a lightweight chat request. If discovery fails but chat still works, check whether your key has model-list permission or whether a gateway blocks model listing.

Reasoning models are treated as a protocol edge case: Notemd merges system and user prompt content for OpenAI reasoning IDs that reject a separate system role, and it exposes reasoningEffort only where the provider definition supports it.

Troubleshooting​

  • 401 or 403: verify the key, project access, billing state, and whether the selected model is enabled for the key.
  • 404 on chat: the Base URL is usually wrong or already includes /chat/completions.
  • Model appears in docs but not in the selector: model-list access and chat access are separate failure points; test the exact model ID manually in the provider profile.
  • Reasoning model rejects messages: use a model ID covered by the OpenAI reasoning handling path or switch to a non-reasoning chat model.

When To Use​

Use OpenAI when you want the most predictable default hosted path, when /v1/models discovery is important, or when a workflow depends on OpenAI reasoning model semantics. Prefer a gateway or local provider when routing, privacy, or cost controls matter more than direct OpenAI API behavior.

Next Steps​