Release Notes — 2.5.0

Released: 2026-09-09

Kure Monitor 2.5.0 adds multi-provider LLM management, custom Markdown instructions for AI prompts, automatic provider failover, a complete website and dashboard redesign, and a new GitHub Actions CI pipeline built for FluxCD GitOps.

Headline Changes

• Multi-LLM Provider Profiles: Register multiple AI providers simultaneously (Ollama, OpenAI, Anthropic, Gemini, GitHub Copilot, custom endpoints). Set a default with automatic failover — if the primary provider fails, Kure tries the next active provider before falling back to rule-based analysis. Engineers can also pick a specific model per-triage from a dropdown in Pod Details.
• Custom LLM Instructions: Cluster admins can write a Markdown instruction document (up to 10KB) that gets prepended to every AI prompt. Write directly in the Admin Panel, or import/export .md files. Use it to document service mesh conventions, require Helm values.yaml diffs instead of raw kubectl, or point the model at internal runbooks.
• GitHub Copilot / Models Provider: Added copilot as a first-class provider type for GitHub Models inference.
• Encryption Key Preservation: The Helm chart now uses lookup to preserve the existing Fernet encryption key Secret across helm upgrade, preventing InvalidToken errors that previously invalidated all stored API tokens on every upgrade.
• Website Redesign: Complete landing page rebuild, contact page, three new guides (Air-Gapped Ollama, Cascading Failures Playbook, K8s AI Debugging), new comparison pages, legal pages, and updated interactive demo components.
• Dashboard UI Refresh: Consistent dark mode across all components, multi-LLM management panel, per-finding model attribution in AI Advice, and model selector dropdown in TroubleshootSection.

Multi-LLM Provider Profiles

You can now register multiple AI providers simultaneously — for example, a local Ollama instance alongside Anthropic Claude and GitHub Copilot — and manage them all from the Admin Panel.

• Provider Pool: Each registered LLM has a human-readable name, provider type, model, and active/default status. Register as many as you need.
• Designated Default: Mark any profile as the cluster default. All automated scans and initial pod diagnoses use this provider.
• Automatic Failover: If the default provider times out, hits rate limits, or returns errors, Kure automatically tries the next active provider in your pool before falling back to deterministic rule-based analysis.
• On-Demand Model Selection: Engineers inspecting a pod failure can pick a specific registered model from a dropdown to regenerate diagnoses, rather than being locked to the default.
• New API Endpoints: Full CRUD on /api/admin/llm/configs, plus /api/admin/llm/configs/{id}/default and /api/admin/llm/configs/{id}/test. A read-only /api/llm/available endpoint powers the frontend model selector dropdowns.

See LLM Providers & Multi-Model Support for configuration details.

Custom LLM Instructions

Cluster administrators can now write a Markdown instruction document that gets prepended to every AI prompt Kure generates — pod triage, AI Advice explanations, security fix proposals, and manifest generation.

• In-Dashboard Editor: Write and save instructions directly in Admin Panel → AI Configuration.
• Import / Export: Upload an existing .md file or download your active instructions to commit them to version control.
• Scoped Injection: Instructions are inserted as a clearly delimited block between the system preamble and the Kubernetes telemetry, so the model treats them as authoritative context.

Typical use cases: documenting service mesh conventions (Istio sidecar inspection order), requiring Helm values.yaml diffs instead of raw kubectl commands, or pointing the model at internal runbooks.

See Custom Instructions for examples and architecture details.

GitHub Copilot / Models Provider

Added GitHub Copilot and GitHub Models as a first-class provider option (copilot) with the models.github.ai/inference endpoint and default model openai/gpt-5.5-mini.

Encryption Key Preservation

The Helm chart’s encryption-secret.yaml template now uses lookup to check whether the encryption key Secret already exists in the cluster before generating a new one. Previously, helm upgrade would regenerate the Fernet key on every run, invalidating all stored API tokens and causing InvalidToken / HTTP 500 errors until the LLM provider was re-saved.

Website Redesign

• Complete landing page rebuild with a new hero section, interactive demo embed, trust bar, features grid (now including Multi-LLM Profiles and Custom Instructions cards), installation snippet, and “Where Kure fits” comparison table.
• Added contact page with a working form.
• Added new documentation pages: features/custom-instructions, updated configuration/llm-providers with multi-model coverage, and expanded the features/overview table.
• Added three new guides: Air-Gapped Local LLMs (Ollama), Cascading Pod Failures Playbook, and Kubernetes AI Debugging Solution.
• New comparison pages: K8sGPT vs Kure, Open Source Alternatives.
• Legal pages: Privacy Policy, Terms of Service, Cookie Policy.
• Updated the interactive demo app components to match the production dashboard redesign.

Dashboard UI Refresh

• Redesigned all dashboard components with consistent dark mode support across settings, modals, status badges, and the manifest viewer.
• Updated LLM Settings panel with multi-config management UI (add/edit/delete/test individual providers, set default, custom instructions editor with character count).
• Improved TroubleshootSection and PodDetails with model selector dropdown for on-demand LLM switching.
• Updated AdviceFindingCard with per-finding model attribution.

Full Changelog

See CHANGELOG.md in the repository for the full commit history.