> ## Documentation Index
> Fetch the complete documentation index at: https://docs.squared.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# March 2026 Releases

> Release updates for the month of March 2026

## Previous releases

## 🚀 Features

* **Agentic Coding (App Generation)**
  Introduced Agentic Coding feature for creating and managing AI-powered applications with interactive development sessions.
  * Create apps with workspace and user ownership
  * Interactive coding sessions with three states: active, paused, and ended
  * Prompt-based development flow with user and assistant roles
  * Deployment tracking with status monitoring (pending, running, succeeded, failed)
  * Session management with sandbox environments and preview URLs
  * Configurable agent models and session settings
  * Role-based permissions for app creation and management (Admin/Member: full CRUD, Viewer: read-only)

* **Prompt to Workflow Enhancements**
  Enhanced the prompt-to-workflow feature with improved UI and functionality for creating workflows from natural language descriptions.
  * Removed timestamps and chat names from AI workflow builder for cleaner interface
  * Improved workflow generation from prompts with better context handling
  * Added context to prompt functionality for more accurate workflow generation
  * Enhanced database support for prompt-to-workflow sessions
  * Added AI Workflow Builder panel with chat-driven workflow creation
  * Context component selector for adding workflow components as context
  * Workflow response visualization showing added, modified, and deleted components
  * Session history drawer with categorized workflow sessions
  * Suggested questions in empty state for guided workflow creation

* **Hosted Data Store Catalog Refresh**
  Implemented automatic catalog refresh for Hosted Data Stores to keep connector schemas synchronized.
  * Automatic catalog refresh when creating or updating data store tables
  * Refreshes both source and destination connector catalogs
  * Resilient error handling - refresh failures are logged but don't block table operations
  * Ensures connector catalogs stay up-to-date with schema changes

* **Human Approval Component**
  Introduced Human Approval Component for human-in-the-loop workflows, enabling manual review and approval gates within automated processes.

* **Tools Component UI Update**
  Enhanced the Tools selection interface for workflows with improved categorization and filtering.
  * Added tabbed interface to filter tools by category (All Tools, AI Squared, Custom)
  * Streamlined tool configuration flow with metadata handling
  * Improved error handling and validation for tool selection

* **Workflow Analytics - Token and Cost Tracking**
  Added comprehensive token usage and cost tracking to workflow analytics.
  * Track tokens used per workflow execution
  * Monitor estimated costs based on LLM usage
  * View detailed analytics in workflow dashboard

* **Firecrawl MCP Integration**
  Added Firecrawl Model Context Protocol (MCP) server integration for enhanced web scraping capabilities within workflows.
  * Secure API key-based authentication
  * Streamable HTTP transport protocol
  * Bearer token authentication support

* **PostgreSQL Batch Support**
  Enhanced PostgreSQL destination connector with high-performance batch write capabilities.
  * Bulk operations supporting up to 10,000 records per batch
  * Automatic fallback to individual writes on batch failure
  * Efficient upsert operations using PostgreSQL's ON CONFLICT clause

* **Devin Integration Enhancement**
  Switched Devin integration from SSE to Streamable HTTP for improved performance and reliability.

* **Agent-to-Agent (A2A) Workflow Component**
  Added a new A2A Agent component type for AI workflows, enabling agent-to-agent communication with a full configuration UI.
  * Connect to external AI agents using JSON-RPC protocol via the Agent Workflow editor
  * Configure agent endpoint URL and API key with a "Connect & Load Skills" button
  * Auto-discover remote agent skills from the standard `/.well-known/agent.json` endpoint
  * View discovered skills as read-only tags in the workflow sidebar
  * Support for multiple auth types including Bearer token and Basic auth
  * Secure credential handling with configuration values masked in API outputs

* **Configuration Security Enhancement**
  Improved security for workflow components with sensitive authentication data.
  * Auth configuration values are automatically masked when returned via API
  * Applies to A2A Agent components and other components with sensitive credentials
  * Prevents accidental exposure of secrets in API responses

* **Weaviate Destination Connector**
  Added Weaviate as a new destination connector for syncing vector data to Weaviate collections.
  * API key-based authentication
  * Automatic schema discovery from Weaviate classes
  * Batch write support with upsert capability
  * Property type coercion for text, integer, number, boolean, and date fields

## 🐛 Bug Fixes

* **Source Connector Delete**
  Fixed issue with deleting source connectors to ensure proper cleanup and resource management.

* **Guardrails Component Text**
  Corrected text for "Vulgar Content or Profanity" in guardrails component for improved clarity.

* **Synk Security Vulnerability**
  Fixed critical security vulnerability identified in Synk code scanning.

* **Nokogiri Security Update**
  Updated nokogiri gem to address security vulnerabilities.

* **Workflow Edge IDs**
  Fixed edge ID handling in workflows, converting from string to integer for better data consistency.

* **HITL Pause Check TypeError**
  Fixed a TypeError that occurred when checking for human-in-the-loop pause status on non-HITL workflow components.

* **Text to SQL Cyclic Dependency**
  Resolved a cyclic dependency issue in the Text to SQL workflow component.

* **SSE Streaming Buffering**
  Fixed SSE streaming buffering and zombie connection issues to improve real-time event delivery reliability.

* **API Base URL Handling**
  Fixed an issue where HTTPS was not automatically prepended to API base URLs, causing connection failures.

* **A2A Edge Fetching**
  Fixed agent-to-agent component fetching from workflow edges for correct data flow resolution.

* **Prompt-to-Workflow Optimization**
  Improved prompt-to-workflow generation performance and accuracy.

* **User Message Text Color**
  Fixed user message text color display issue in the chat interface.

* **CI Optimizations**
  Improved CI pipeline performance with caching, sharding, and parallelization.

* **OpenRouter LLM Usage Logging**
  Added Rails loggers for OpenRouter in LLM usage logs for better observability.

## ⚡ Improvements

* **Python Executor Import Helper Text**
  The Python Executor component now includes inline helper text in the default code template listing all supported libraries and a reminder that imports must be placed inside the `run` function.

## ⚙️ Miscellaneous Tasks

* **Testing & Quality Improvements**
  Expanded test coverage across multiple modules.
  * Added tests for Enterprise/Settings-and-Views/Models
  * Added tests for Enterprise P2 modules
  * Added tests for views/Settings
  * Added comprehensive test coverage for Agentic Coding models (App, Session, Prompt, Deployment)
  * Added tests for AI Workflow Builder components and hooks
  * Added tests for workflow session management and message handling

* **Error Message Improvements**
  Extracted error messages for Agent Component to improve debugging and user feedback.

* **Dependency Updates**
  Added tiktoken\_ruby to Gemfile for improved token counting capabilities.

* **Documentation Workflow**
  Added update-docs workflow to multiwoven-ee repository for automated documentation updates.

* **Feature Flags**
  Added feature flag support for prompt-to-workflow functionality to enable controlled rollout.

* **TextInput Max Length**
  Added max length validation to TextInput components across the UI for improved data integrity.

* **Text to SQL Workflow Fix**
  Fixed the text to SQL workflow component for correct query generation.

* **Security Vulnerability Fixes**
  Addressed critical vulnerability identified in code scanning to improve platform security.
