LLM Providers

Kure Monitor uses an LLM to generate contextual fixes for pod failures and security findings. The provider is configured after install from the Admin Panel → AI Configuration — no API keys at install time.

2.4.0 — the LLM is now also used by the new AI Advice tab. Explanations for advice findings are generated lazily on card expand (one LLM call per finding, cached) so a scan with N findings does not cost N requests up front. The advice explainer prompt is constrained to only reference data present in the finding’s evidence dict — no invented replica counts, image names, ports, or labels.

2.3.4 — the AI Configuration panel defaults to Groq for new installations (was Ollama). Existing installs are unaffected; this only changes the initial pre-selection. Model catalogs across all six providers were refreshed to current (May 2026) latest models.

Supported providers

Provider	Alias	Default Model	Pricing
Groq (default)	groq, groq_cloud	meta-llama/llama-4-scout-17b-16e-instruct	groq.com/pricing
OpenAI	openai	gpt-5.5-mini	openai.com/pricing
Anthropic	anthropic, claude	claude-sonnet-4-6	anthropic.com/pricing
Google Gemini	gemini, google	gemini-3-flash	ai.google.dev/pricing
GitHub Copilot (GitHub Models)	copilot, github, github_models	openai/gpt-5.5-mini	GitHub Models
Ollama (local)	ollama	llama4:scout	Free / self-hosted

Model catalogs (dropdown options)

Provider	Models in the dropdown (default in bold)
OpenAI	gpt-5.5, gpt-5.5-mini, gpt-5.4-mini
Anthropic	claude-opus-4-7, claude-sonnet-4-6, claude-haiku-4-5
Google Gemini	gemini-3.1-pro, gemini-3-flash, gemini-3.1-flash-lite
Groq	openai/gpt-oss-120b, meta-llama/llama-4-scout-17b-16e-instruct, llama-3.3-70b-versatile, openai/gpt-oss-20b
Ollama	llama4:scout, llama3.3, qwen3
GitHub Copilot	openai/gpt-5.5, openai/gpt-5.5-mini, anthropic/claude-sonnet-4-6

GitHub Copilot (GitHub Models)

• Display name: GitHub Copilot (GitHub Models)
• Auth: GitHub Personal Access Token (fine-grained, with the Models permission)
• Base URL: https://models.github.ai/inference
• API: OpenAI-compatible
• Example models: openai/gpt-5.5, openai/gpt-5.5-mini, anthropic/claude-sonnet-4-6

Ollama

For air-gapped clusters. Run Ollama in your cluster, point Kure at it, and your cluster data never leaves your network. Default model: llama4:scout. Other models in the dropdown: llama3.3, qwen3.

Recommendations

Use case	Provider	Model
Best quality	Anthropic	claude-opus-4-7
Best value	OpenAI	gpt-5.5-mini
Fastest	Groq	meta-llama/llama-4-scout-17b-16e-instruct
Free tier	Groq	meta-llama/llama-4-scout-17b-16e-instruct
Google ecosystem	Google	gemini-3-flash
GitHub ecosystem	GitHub Copilot	openai/gpt-5.5-mini
Local / air-gapped	Ollama	llama4:scout

Configuring a provider

1. Open the dashboard
2. Admin Panel → AI Configuration
3. Pick a provider
4. Paste your API key (or PAT for GitHub Copilot)
5. Pick a model from the dropdown
6. Click Test Connection — verifies the key works
7. Click Save Configuration

If the LLM call fails at runtime, Kure falls back to rule-based solutions so the dashboard stays useful.

Rotating an API key

1. Generate a new key with your provider
2. Admin Panel → AI Configuration
3. Replace the existing key
4. Test Connection → Save

LLM API keys are encrypted at rest using a Fernet key (security.encryptionKey in Helm values). If left empty at install time the chart auto-generates one.

Removing a provider

DELETE /api/admin/llm/config

…or click Delete in the Admin panel. Kure reverts to rule-based solutions.