Censhare AI: Release information

📦 v0.1.3

Parameter Naming & Backwards Compatibility

The single-letter parameter a was renamed to enableAlphaMatting for better readability and maintainability with no breaking changes.

• Renamed a → enableAlphaMatting

• Legacy field names remain fully supported via fallback logic

• New resolveRuntimeAlphaMatting() function handles multiple field name variants

• Updated all bindings: HTML form inputs, TypeScript interfaces, and state persistence logic

ℹ️ Existing configurations continue to work seamlessly — no migration required.

📦 v0.1.2

Full-featured asynchronous job handling with intelligent polling, queueing, and live status monitoring.

Asynchronous Job Polling

• Automatically polls the server's /queue/:id endpoint when the API returns a queued job

• Waits for completion and returns the final result — no separate polling flow needed

Configurable Polling Controls

• Enable/disable auto-polling

• Set poll interval (default: 2 seconds)

• Set poll timeout (default: 5 minutes)

• Configurable in the editor or overridden per message

Queue & Concurrency Control

• Polling toggle: when disabled, immediately returns the queue ticket (msg.payload.id) for downstream manual polling

• Local queueing: the "Queue?" option strictly sequences incoming messages — preventing Node-RED memory flooding and protecting the remote API from concurrent request overload

Live Status Indicator

• Node badge shows aggregated active + queued requests plus live job progress

• Examples: 3/10 in queue or 2 ahead

Lifecycle & Reliability

• Safe abort on close/redeploy: aborts in-flight requests and clears pending polling intervals to prevent memory leaks

• Smarter result decoding: extracts image buffers from raw buffers, data URLs, base64 strings, serialized buffers, byte arrays, and common wrapper objects

• Richer job metadata: successful async jobs attach the full job status object to msg.job (id, runtime, duration, etc.)

Configuration & Docs

• Default API URL now sources from an environment variable instead of a plain string

• Editor help updated to cover polling, queueing/concurrency, the status indicator, and safe-abort behavior

📦 v0.0.8

Google GCP / Vertex AI Integration

• New configuration files add full Vertex AI support as an LLM backend

• Configuration UI provides all fields needed to connect to a GCP project and use Vertex AI-hosted models

• New dependency @langchain/google-vertexai powers the integration

LLM Node — UI & Logic Overhaul

• HTML template substantially reworked to accommodate new provider options and improve UX

• Backend updated with new routing logic, provider selection handling, and GCP/Vertex AI support

AWS Bedrock Refinements

• Minor fixes and adjustments to align with the broader configuration interface improvements

📦 v0.0.4

Google Vertex AI (GCP) Support

• Replaced & unified google package - supports both Google AI Studio and Vertex AI via ChatGoogle

• New Platform selector: choose Google AI Studio (API key auth) or Vertex AI (service account auth)

• Full GCP service account credentials form: Project ID, Private Key ID, Private Key, Client Email, Client ID, Cert URL

• Region/Location, Endpoint, and API Version fields added

• Model instantiation explicitly sets responseModalities (TEXT, IMAGE, or AUDIO)

Image Output as Binary Buffer

• New Image (Binary Buffer) output format for Google and OpenAI providers

• Google uses responseModalities: ["IMAGE"]; OpenAI uses ChatOpenAITools.imageGeneration()

• msg.payload contains the binary buffer; msg.filename is set automatically (e.g. generated-image-1.png)

Censhare Private Server Model Type

• Added as a dedicated model type and set as the new default, routing through ChatOllama

Keep Alive Control (Ollama)

Request Queue

• New Queue toggle in the Advanced tab — processes messages sequentially via an in-memory promise queue

• Node status shows live queue depth (e.g. processing request (2 in queue)...)

Multimodal Content Blocks (msg.content)

📝 For Ollama/Censhare, image URLs are automatically fetched and converted to base64 via the new fetchImageAsBase64() helper — Ollama does not support direct URLs for binary inputs.

Structured Output (JSON Schema)

• Schema-based output now uses standard JSON Schema (replaces the previous Zod string approach)

• Built-in default schema ships with the node

• Format JSON button validates and prettifies the schema; editor switched to JSON mode

UI & Configuration Changes

• New dedicated Output tab — format options (Raw / Structured / Image) moved out of the Advanced tab

• Removed: API Type radio (Public/Private) and confirmation dialog, Streaming and Memory toggles, isTracing

• ollama-gen model type deprecated/hidden (standard Ollama covers both use cases)

• Conditional fields: API URL/Key hidden for Vertex AI; HTTP Headers hidden for Google

• Node label shows the model name only when it's a static string

Backend Refactoring

• handleMessage() extracted as a standalone async function consolidating model rebuild, status updates, and output routing

• buildConversation() and toLangChainBlock() now async to support URL-to-base64 fetching

• buildModelInstance() refactored to accept the full node object

• Processing queue initialised at startup and reset on close

• Unused imports removed; all var declarations replaced with const/let

Build & Documentation

• README.md created from scratch — setup guide, prerequisites, environment variables, build/deploy instructions, and Confluence links

• Build script updated: cpx replaced with copyfiles, nodemon added, unused watch scripts removed

• Node entry paths in package.json corrected to include the dist/ prefix

📦 v0.0.3

Generative AI

• LLM node (Private/Public AI models)

• Image Background (Private/Public AI service)

Agentic AI

• Agent node (Private/Public AI models)

• Tool calls with low-code (trigger & response handling)

• Tracing

• Ollama models