Liz: The Rancher AI Assistant v1.1.0 Release Highlights

OAuth2 authentication for MCP and agent connections: Liz and its agents can now authenticate to external MCP servers using OAuth2, following the Model Context Protocol authorization specification. (#191, #10)

Custom headers authentication for MCP connections: MCP server configurations now support custom header-based authentication secrets, enabling integrations such as the SUSE Observability MCP server. (#188, #182)

Per-agent model configuration: Each agent can now be configured with its own model, allowing cheaper models for simple questions and larger models for complex operations. (#170)

Global HTTP_PROXY support: Added support for configuring HTTP_PROXY globally across the LLM provider, the Rancher MCP server, and calls to external MCP servers. (#208)

Custom AWS Bedrock endpoint URL: The AWS Bedrock endpoint is no longer hard-coded, allowing environments without direct internet access to use AWS PrivateLink. (#206)

New "Generic" provider tile: A new Generic provider tile simplifies configuring OpenAI-compatible endpoints, replacing the previous advanced-settings flow. (#187)

SUSE Virtualization related resources support: "Related Resources" links now correctly resolve SUSE Virtualization (Harvester) resources such as virtual machines. (#241)

Fleet renamed to Continuous Delivery: References to Fleet throughout the UI, including agent mode selection, have been renamed to Continuous Delivery. (#245)

Sub-agent architecture: Introduced a sub-agent architecture that improves agent handoff and enables cross-agent MCP tool calls for complex, multi-step prompts. (#169, #207)

Fleet/Continuous Delivery MCP tooling: Added a getHelmOp tool so agents can inspect Continuous Delivery Helm operations directly. (#51)

More consistent MCP tool schemas: Standardized tool field definitions across MCP tools to use InputSchema, improving reliability of tool calls. (#48)

Better resource filtering: Identified and implemented a filtering approach for the listKubernetesResources tool, improving results for large clusters. (#44)

Fixed an issue where chats could not be stored when using a custom CA with insecureSkipTls=false. (#326)

Fixed an agent deadlock caused by multiple pending interrupts requiring an explicit interrupt ID when resuming. (#318)

Fixed permanent reconciliation failures on AIAgentConfig caused by transient MCP server unavailability. (#205)

Updated the Gemini API model selection list, removing deprecated and preview models. (#265)

Fixed sliding badges breaking table layouts in the UI. (#122)