RM-LLM 0.7.1

RM-LLM 0.7.1 is now compatible with the Easy-Use UI. Model, Simple and Advanced sections fold without leaving disconnected ghost sockets; connected sockets and their wires stay attached. This layout was verified by the user on Windows.

Also in this release:

• Connected text, number, dropdown and toggle controls show the values received on the last run without saving those values over manual settings.
• One provider selector serves both Direct APIs and OpenRouter hosting providers.
• Unsupported optional controls and media are omitted from requests while their connections remain available for a compatible model.

The internal ComfyUI node ID remains RM_LLM_050 for workflow compatibility. See CHANGELOG.md for the full dated notes.

RM-ComfyUI-LLM 0.7.0

• 2026-10-17

• Show API-reported precision beside OpenRouter hosts, including Precision unknown when undisclosed.

• Stabilise RM-LLM labels and row dimensions against theme styling while preserving collapsible sections; Easy-Use visual verification remains user-led.

• Add router as the first inline dropdown: OpenRouter or Direct.

• Support model-first and hosting-provider-first discovery through OpenRouter's official endpoints, with a five-minute provider/filter cache.

• Enforce selected hosts and use endpoint-specific controls and limits; Auto retains automatic routing.

• Preserve legacy provider inputs, node IDs, widget order and workflow connections; add router and hosting_provider inputs at the end of the schema.

• Keep OpenRouter credentials separate from direct provider credentials. No local-server support or extra generation calls.

RM-LLM 0.6.0

• Give api_key_system the same inline searchable picker, reusing the Provider picker while preserving the two credential modes and key storage.

• Replace the Provider select with the same inline searchable list styling as Model: up to ten visible rows, provider subtitles and keyboard selection; stored provider names remain unchanged.

• Add a provider status line below request_timeout_seconds, with a coloured circle, short updates/errors and explanatory tooltips; use existing request results only.

• Preserve the existing Set button; separate provider status from key storage, including authentication failures, rate limits, model availability, service errors, credits and access restrictions.

• Ignore stale request results after selection or key changes; retain native ComfyUI execution diagnostics and existing node IDs/connections.

RM-LLM 0.5.1

• Use lowercase underscore input labels, mark model controls [API], and supply a default user prompt for new nodes.

• Give automatic budgets a display scale so nonzero usage remains visible; use lowercase budget labels.

• Add connectable Input, Thinking and Output token budgets in Advanced; zero preserves existing/default limits.

• Map output and supported thinking targets to provider controls. Shared generation limits combine thinking and answer allowances; unsupported numeric thinking targets remain planning allowances.

• Check estimated text input before generation without truncating prompts or counting base64 media as text.

• Show three green usage bars within Advanced only after a successful run, using API counts where available and labelled estimates/unknowns otherwise. No numeric labels or extra API calls.

• Keep 0.5.0 node IDs, credentials, cache and native socket/collapse handling compatible.

• Add display-only provider subtitles to the dropdown; plain provider values, socket inputs and the header remain unchanged.

RM-LLM 0.5.0

[0.5.0] - 2026-10-04

• Add OpenAI, Google Gemini, Anthropic, DeepSeek, Groq, Mistral AI, xAI, Together AI and Fireworks AI, retaining OpenRouter, Featherless and LithosAI.
• Adapt authenticated model discovery and pagination to each provider, with cached catalogues and manual model IDs.
• Add native Claude/Gemini requests and streaming, supported media, reasoning controls and normalized outputs retaining native response data.
• Reuse the provider-specific environment/session key process and stable native frontend hooks.
• Register separate 0.5.0 node IDs and routes for installation alongside 0.4.0.
• Restore native socket dragging by keeping the HTML widget wrapper from intercepting socket clicks; confirmed working by the user.
• Expand the README with all twelve providers, Featherless model discovery and LithosAI's speed focus.
• Add offline provider contract tests. User-tested locally on Windows; full live provider/account coverage and non-Windows checks remain pending.

RM-ComfyUI-LLM 0.4.0

[0.4.0] - 2026-10-03

• Use native ComfyUI extension hooks and widget layout for the collapsible Model, Simple and Advanced sections.
• Keep input sockets stable through section changes and browser refresh; update wire positions with their sockets.
• Add inline model search with manual model IDs, cached catalogues and download progress. Featherless catalogue downloads use public requests and retain completed pages for resuming.
• Improve persistent API-key setup and availability checks across providers on Windows and Linux. Keys remain outside saved workflows.
• Describe model discovery, supported image/video inputs, Thinking/Reasoning, linked Creativity controls and live console streaming in the package listing and README.
• Include the matching dated release notes when publishing to Comfy Registry.

This version registers separate 0.4.0 node IDs. Existing 0.3.x workflows are not automatically converted; retain the older package when using those nodes.