AI model: StockAI's or your company's

Which language model the assistant, summaries and data diagnosis use: StockAI's default or your own account at OpenAI, Claude, Gemini, Groq, OpenRouter, DeepSeek or a compatible server.

On this page
  1. What is in use
  2. Use your own provider
  3. An OpenAI-compatible server
  4. Back to the default

Open AI model from Settings → AI assistant. An administrator edits it; an analyst sees it read-only ("Only an administrator can change the AI provider."); a read-only user does not see it.

AI model screen with the In use: StockAI's default model box, usage over the last 7 and 30 days and the list of providers

What is in use

In use
"StockAI's default model" or "{provider} · {model}" if the company chose its own. If the installation has no StockAI model, it tells you to choose your own provider to turn the assistant on.
State
With your own provider: Working, Failing, Paused (it failed 5 times in a row; retried by itself after a short wait or when you press "Test connection") or Retried on the next request.
Last successful answer / Last problem
When it last answered correctly and, if it failed, which problem it had and how to fix it.
Last 7 and 30 days
Requests (and how many failed), tokens sent and received, and the estimated cost if you entered your provider's prices. Without prices it says "No cost estimate".

Use your own provider

Form with OpenAI chosen, the hidden API key, the gpt-4.1-mini model, advanced options open with prices 0.15 and 0.6, and the Test connection and Save buttons
Form filled with a made-up key, not saved.
  1. Under Provider, pick one: OpenAI, Claude (Anthropic), Google Gemini, Groq, OpenRouter, DeepSeek or Base URL (OpenAI-compatible). Those with a free tier carry a "Free option" badge (with the provider's usage limits).
  2. Read the provider's hint and, if you have no key, press "Get a key" (opens the provider's page).
  3. Paste the key into API key. The eye shows or hides it.
  4. Check the Model: a suggested one is filled in; you can type another or pick from the list.
  5. Press "Test connection": it tests what is in the form without saving it and answers "Connection works: {model} answered in N ms", or the problem. After a successful test the model list becomes your account's real one.
  6. Press "Save". You will see "AI provider saved" and the box at the top switches to your provider.
Organization / Project (optional)
Only for providers that use them (OpenAI).
Advanced options → Temperature
From 0 to 2. Empty = the provider's.
Maximum answer length in tokens
Empty = each screen's own, up to the maximum in the label.
Price per million tokens sent / received
In US dollars, per your contract. With them the screen estimates the cost; it is an estimate that applies today's prices to the whole period and may differ from your invoice.
Use this provider
Off: the settings are kept but StockAI's default is used.

An OpenAI-compatible server

Form with the Base URL option chosen and the field holding https://api.together.xyz/v1

For Ollama, LM Studio, vLLM, Together, Mistral or Azure OpenAI: enter the Base URL that ends before /chat/completions. The key is optional if your server asks for none. If the installation does not allow internal network addresses, the hint says so: use a public https URL.

Back to the default

With a provider saved, "Back to StockAI's default" appears. Confirming deletes the saved key and the assistant goes back to StockAI's model.