Skip to content

DocsConfiguration

Choose and configure a model provider

Hosted or local, and how to change your mind later.

About 4 min read

Any provider that speaks the OpenAI chat-completions format works. Change it whenever you like under Settings → AI provider; existing analyses are not affected.

Your options

KindBase URL looks likeData leaves your network?
Hosted providerhttps://api.openai.com/v1Yes — redacted context only
Ollama, on your machinehttp://localhost:11434/v1No
LM Studio, on your machinehttp://localhost:1234/v1No

Running a local model

./scripts/setup-ollama.sh      # install and pull a model
./scripts/ollama-manage.sh     # start, stop, and check it

Then set the base URL to http://localhost:11434/v1 and leave the API key empty. A local model is slower than a hosted one, and that is usually an acceptable trade for keeping incident context inside the network.

What is sent

Before anything is sent, fields whose name matches the sensitive-key list are replaced with [REDACTED] — passwords, tokens, authorization headers, cookies, API keys, card numbers, and more.

Matching is by field name. A secret pasted into a free-text message, or stored under a name nobody anticipated, is not detected. What your application puts into an error payload is the control that matters most.

If the provider is unreachable

The analysis is retried, and if it still fails a fallback record is written at MEDIUM risk so the incident is never silently dropped. You will see it in the list with a generic summary — that is the signal to check your provider settings.