v.26.11.19.1

Estimated reading: 7 minutes

This release introduces new features, components, and enhancements in Robility Flow to improve AI-powered workflow execution, agent collaboration, memory management, language support, and workflow configuration.

Deprecated Components

The following legacy components are deprecated in this release and may be removed in a future Robility Flow release. While they will continue to work in existing flows, please review your workflows and migrate to the recommended alternatives listed below to maintain compatibility with future releases.

Deprecated Component Recommended Alternative
Text Input Chat Input
Text Output Chat Output
Text Operations Data Operation
JSON Operations Data Operation
Table Operations Data Operation

Note: The agent_description field within the Agent component is also deprecated in this release.

Limitations

1. Compatibility Update:Existing workflows may require updates to ensure compatibility with the latest version of Robility Flow.
2. Internal A2A Communication Only: External A2A integration is not supported in this release. A2A communication is limited to internal agents within the same Robility Flow project.
3. Non-Editable Memories: Created memories cannot be edited individually. To remove a Memory Base, the entire Memory Base must be deleted.
4. Session Limit: A maximum of 10 active sessions per user is supported in Robility Flow Assistant. An existing session must be deleted before creating a new one when the limit is reached.
5. Playground File Preview Formats: File preview in the Playground supports only HTML, PDF, DOCX, PPTX, XLSX, and CSV formats. Previewing other file formats is not supported.

What’s New

New Features

1. Memory: Enables AI Agents to retain and recall relevant information from previous interactions. It stores conversational information as embeddings in a configured vector database and retrieves relevant details through semantic search, reducing the need to repeat previously provided information.

2. Multi-Language Support: Robility Flow now supports seven languages English, French, Spanish, German, Portuguese, Japanese, and Chinese—allowing workflows to handle interactions across multiple languages.

3. DB Providers in Settings: Allows a vector database provider to be configured under Flow Settings. The selected provider is used by Memory and Knowledge Base to store embeddings and retrieve relevant information through semantic search.

4. A2A Server: Allows a Robility Flow workflow to be published as an A2A-compatible agent so other agents within the same project can discover and delegate tasks to it via the A2A protocol.
(Note: External A2A Server integration is currently under development; only internal A2A communication is supported in this release.)

5. Flow Assistant: An AI-powered assistant that helps create complete workflows, modify existing workflows, and answer workflow-related queries using natural language instructions, automatically using your currently open workflow as context.

6. Flow Radar: An interactive minimap on the workflow canvas provides a bird’s-eye view of every connected node, making large workflows easy to navigate. Pan across the canvas, jump straight to any section, zoom in or out, or fit the entire flow to the screen in one click.

New Components

1. A2A Agent: Connects a workflow to other agents supporting the Agent-to-Agent (A2A) protocol to delegate tasks and process responses. Provides Internal mode (connects to a published A2A Agent flow within the same project) and External mode (provides configuration for an external A2A-compatible agent).
(Note: External agent execution is currently under development; only internal A2A agent communication is supported in this release.)

2. Memory Base: Provides access to stored information from past interactions within the same flow. It stores user inputs and Agent responses as embeddings in a configured vector database and retrieves relevant memories through semantic search to generate context-aware responses.

3. Human Input: Enables Human-in-the-Loop (HITL) checkpoints by pausing workflow execution until a user reviews the data and submits a decision. Once an option is selected, the workflow automatically resumes and routes down the corresponding path.

4. Playground LLM: Allows the Language Model used by an Agent to be selected or switched directly from the Robility Flow Playground without modifying the existing workflow configuration.

5. Playground Instructions: Allows flow-specific instructions to be configured directly from the Playground to guide how the connected Agent processes queries and generates responses.

6. Azure Blob Storage: Bring Azure Blob Storage operations into workflows with a dedicated component for managing containers, blobs, and virtual folders. Automate file storage, retrieval, organization, and maintenance while securely interacting with Azure storage resources.

7. Google Cloud Storage (GCS) File System: Integrate Google Cloud Storage directly into workflows to manage buckets, objects, and folder structures. Automate file handling, storage organization, and secure content access while seamlessly incorporating cloud storage operations into end-to-end workflow automation

Enhancements

1. Enhanced Playground UI: Provides a centralized model selection experience displaying models available through configured Model Providers, allowing you to switch Language Models directly from the Playground without modifying the underlying workflow configuration.

2. Deep Agent: Deep Agent now offers enhanced human-in-the-loop controls, with approval required before built-in tools run. New capabilities include Context Summarization, Rubric Validation, and Token Optimizer, making agent runs more controlled, reliable, and efficient.
a. Human-in-the-Loop Approval: Built-in tool actions can be reviewed and approved before execution, keeping humans in control of what the agent does.
b. Context Summarization: Long conversations and intermediate results are condensed automatically, so the agent retains key context across extended runs.
c. Rubric Validation: Agent outputs are checked against defined rubrics to confirm they meet the expected quality criteria.
d. Token Optimizer: Token usage is reduced across agent runs, lowering cost and latency.

3. Agent Component Updates:
a. Default System Prompt – Provides a default system prompt to guide the Agent’s behavior during task execution.
b. Structured Response Output – Supports structured responses in Data and JSON formats through the structured_response option.
c. Default Calculator Tool – Includes the Calculator tool by default, allowing the Agent to perform calculations without additional tool configuration.

4. Tool Approval Required: Available across all tool-enabled components. When enabled, the Agent must obtain user approval before executing the selected tool, providing additional control over operations that create, modify, or interact with data and external systems. If approval is denied, the tool is not executed.

5. File Selection Pagination: The Select Files screen across all components now supports pagination, displaying files across multiple pages rather than loading the entire list at once to improve responsiveness when browsing large collections.

6. Search File: The Search File component now integrates with Directory Search, enabling AI Agents to dynamically discover, locate, and retrieve files during workflow execution.

7. Python Interpreter Security Enhancement: Run custom Python scripts with an added layer of protection. An optional isolated virtual sandbox safely executes code without risking access to underlying system files, networks, or resources, complete with customizable execution limits.

8. URL and API Request Components: Automatically blocks requests to internal networks and private addresses, helping protect the system from unauthorized access and redirect attacks.

Bug Fix

1. Knowledge Base Ingestion Fixes: Resolved Knowledge Base ingestion issues related to Azure Embeddings and document indexing. Fixes include addressing embedding dimension mismatch errors and eliminating incorrect duplicate-entry validation failures when adding or updating documents.

Known Issues

1. Workflow Execution Timeout: Long-running workflows may encounter execution timeouts, particularly when they contain multiple components, large inputs, or models with slow response times.
Workaround: Break workflows into smaller sequential workflows, use concise inputs, and select faster models to help reduce execution timeouts.

2. MCP Client – Project-Based Flow Filtering: The MCP Client may display flows that are not available to the selected project. However, access restrictions remain enforced, and only flows the user is authorized to access can be executed.

IMPORTANT: For details on existing Known Issues and Existing Limitations in Robility Flow, refer to the following sections:

Known Issues
Existing Limitations

Release Date: 19.11.2026

Share this Doc

v.26.11.19.1

Or copy link

CONTENTS
Robility Chatbot
Robility Assistant
Online