OpenAI-Compatible API

This provider works with any API that follows OpenAI's chat completions format: which includes OpenAI itself, many local model servers like Ollama, and a lot of third-party model APIs.

Setup

1. Open Settings → AI Providers
2. Find OpenAI-compatible API and click Configure
3. Enter:
   • API URL: the base URL of the endpoint
   • API Key: your API key (or any string if the server doesn't require one)
   • Model: the model name to use
   • Service tier: optional, OpenAI-only advanced setting for api.openai.com
4. Click Save

Common setups

OpenAI

• API URL: https://api.openai.com/v1
• API Key: your OpenAI key from platform.openai.com
• Model: any model from your account (see table below)
• Service tier: leave on Auto unless you specifically want Flex, Default, or Priority

Available OpenAI models:

Model	Good for
gpt-5	Most capable, best answers
gpt-5-mini	Fast and capable, good default
gpt-5-nano	Fastest, lowest cost
gpt-4o	Previous generation, still solid
gpt-4o-mini	Previous generation, cheap

Ollama (local)

Run models locally with no API key and no data leaving your machine.

1. Install Ollama and pull a model: ollama pull llama3.2
2. In Coeus:
   • API URL: http://localhost:11434/v1
   • API Key: ollama (any string works)
   • Model: llama3.2

Other providers

Any provider with an OpenAI-compatible endpoint works the same way: Groq, Together AI, Mistral (with their compatibility layer), etc. Check their docs for the base URL.

Choosing a model

For answering questions about your notes, a mid-size model (GPT-4o-mini, Llama 3.2, Mistral 7B) works well and is fast and cheap. You only need a larger model for long, involved questions.

OpenAI service tiers

If you use the official OpenAI API at https://api.openai.com/v1, Coeus lets you choose an optional service tier.

• Auto: default behavior
• Default: standard pricing and latency
• Flex: lower cost, but slower and less available
• Priority: higher priority processing when your account supports it

This setting is ignored for other OpenAI-compatible endpoints like Ollama, LM Studio, OpenRouter, and similar providers.

Attachments

Image attachments work if the model supports vision (like GPT-4o). Text-based models will skip image content.