Orkes Conductor Changelog

LLM tasks now support inputGuardrail and outputGuardrail; pre- and post-call hooks that scrub or validate text going into or out of the task without adding a separate workflow step. Guardrails can run inline as a JavaScript expression, call an external HTTP service, or trigger a registered workflow, with a configurable failure mode (FAIL or WARN) so a broken guardrail never silently lets unredacted content through. Each guardrail invocation runs as its own auditable sub-workflow execution.

Orkes Conductor now provides audit logs that record who changed what and when across a cluster's resources. Each entry can be compared against the resource's previous state, making it easier to investigate unexpected changes and support compliance reviews. Viewing audit logs requires administrator access.

Conductor now supports out-of-the-box integrations with popular external services, including Slack, GitHub, Jira, Azure Storage, Azure Functions, Google Cloud Storage, Google Cloud Functions, AWS Lambda, MySQL, PostgreSQL, Redis, and Stripe. Use these integrations directly in your workflows to perform actions across these platforms without writing custom connectors.

Conductor now supports out-of-the-box integrations with popular external services, including Google Slides, Sheets, Calendar, Drive, and Docs, as well as Notion, Discourse, WordPress, AWS S3, CommonRoom, and HubSpot. Use these integrations directly in your workflows to perform actions across these platforms without writing custom connectors. 

AI Assistant, an AI-powered chat interface in Conductor UI which supports building workflows, debugging executions, and exploring your workflow catalog, can now be powered by Claude (Anthropic) or OpenAI models.

Introduced a new GET /api/metadata/tags/resources endpoint that retrieves all resources associated with a specific tag.

POST /workflow/test and POST /workflow/test/sync now support mocking Sub Workflow tasks. When a mock output is supplied for a Sub Workflow's task reference name, the task completes immediately with that output, and no child workflow is started.

The HTTP task and HTTP Poll task now support an acceptedStatusCodes field in inputParameters. Use this to define which HTTP status codes or families (e.g., "404", "4xx") should be treated as successful task outcomes. When not set, only 2xx responses are accepted. When explicitly set, only the listed codes are accepted; 2xx is not implicitly included.

The Wait task now supports a yield input parameter. When set to true, synchronous API calls return immediately upon reaching a Wait task with a duration, rather than holding the HTTP connection open.

A new endpoint POST /workflow/test/sync has been added for testing workflows synchronously. The key addition in this endpoint is support for mocking HTTP tasks. You can supply mock outputs for HTTP tasks the same way as any other task. If no mock is provided for an HTTP task, it falls through and executes normally.

The Conductor AI Assistant can now leverage configured integrations when helping build workflows, making it easier to create integration-connected workflows through conversational UI.

A new GET /api/metadata/tags/grouped endpoint returns all tags organized by resource type with a count per type. This supports tag management and auditing across large Conductor deployments.

Orkes Conductor’s Gateway lets you expose any workflow as an API endpoint or MCP tool. API Gateway routes map HTTP requests to workflows, while MCP Gateway surfaces workflows as tools for AI agents. Both support authentication, rate limiting, CORS configuration, and pre/post transformation scripts.

Orkes Conductor now supports integrations with AWS and Google Cloud Platform, enabling workflows to securely access private S3 and GCS storage buckets using storage-related tasks.

The new AI task, Chunk Text, splits large text content into smaller, structured s

The new AI task, List Files, retrieves file listings from S3, GCS, Azure Blob, or Git repositories. It is designed for workflows that need to dynamically index or process file-based data sources.

The new AI task, Parse Document, extracts and converts documents from cloud storage, Git repositories, and websites into Markdown format. It supports Office documents, PDFs, HTML, images (via OCR), ZIP archives, and text files.

A new GET /api/scheduler/schedules/search endpoint has been added to support paginated search over schedule definitions. This improves manageability for clusters with large numbers of scheduled workflows.

The Sub Workflow task now supports a Workflow definition parameter, allowing the complete workflow definition to be passed as a variable, object, or array at runtime. This enables more dynamic workflow composition patterns.

Event handlers can now be tagged, bringing them in line with other Conductor resources that support tagging. This improves discoverability, filtering, and management of event handler definitions at scale.

Orkes Conductor now supports integration with Git repositories, including GitHub, GitLab, Bitbucket, and any other repository that supports API tokens. This enables workflows to read from and interact with version-controlled file sources.

Orkes Conductor now supports integration with Ollama, enabling workflows to use locally hosted LLMs in AI tasks. Supports optional authentication headers for secured or proxied Ollama deployments.

The Business Rule task now supports five new string operators: startsWith, endsWith, contains, matches, and length. These expand the range of rule conditions that can be evaluated without custom code.

Business Rule tasks now support a cacheTimeoutMinutes parameter to control how long downloaded rule files are cached at the task level. This gives teams finer control over rule file freshness without relying on system-wide defaults.

A new POST /api/workflow/execute/{name}/{version} endpoint runs workflows synchronously with configurable consistency modes (SYNCHRONOUS, DURABLE, REGION_DURABLE). New signal endpoints let you resume paused Yield or Wait tasks synchronously (POST /api/tasks/{workflowId}/{status}/signal/sync) or asynchronously, with a return strategy to control what the API returns.

AI Prompts now support versioning, starting at version 1 and incrementing with each new save. Teams can maintain multiple prompt variants in parallel, making it easier to iterate and roll back prompt changes without affecting production workflows.

Workflow definitions now support a maskedFields parameter to designate sensitive fields that should be hidden during introspection. Masked values are displayed as ****** in workflow input and output views, preventing unintended data exposure.

Remote Services lets you register HTTP and gRPC service endpoints once and reuse them across HTTP, HTTP Poll, and gRPC tasks. Supports automatic endpoint discovery from Swagger specs or gRPC reflection, circuit breaker configuration, and request hedging for reduced tail latency.

The new Yield task pauses workflow execution until it is explicitly resumed via signal APIs. When used with synchronous workflow execution, Conductor returns immediately at the Yield point rather than holding the connection open, enabling tightly coordinated request-response patterns with external systems.

Workflows triggered by webhook events can now use dynamic idempotency keys defined as workflow inputs via expression, ${workflow.input.<variable>}. This prevents duplicate workflow executions from repeated webhook deliveries using data-driven key values.

The NATS message broker integration now includes default parameters and new pull consumer configuration options: Batch Count, Pull Period, and Pull Batch Wait Time. These additions give teams more precise control over message consumption behavior.

Orkes Conductor now supports integrations with Grok and Perplexity as AI/LLM providers. Teams can connect these models through the Integrations panel and use them in AI-powered workflows.

A new SendGrid integration allows Conductor workflows to send transactional and marketing emails directly. The accompanying SendGrid task enables email sending as a workflow step after configuring the integration.

Orkes Conductor now supports importing standard BPMN files (XML and .bpmn format) directly into workflow definitions. This simplifies migration from BPMN-based systems by eliminating the need to manually recreate workflows.

When a Sub Workflow task has an idempotency strategy configured but no explicit key provided, it now automatically inherits the idempotency key from its parent workflow. This reduces the need for manual key management in nested workflow patterns.

Workflows can now be configured with a rate limit to control how many executions run concurrently. Any workflows triggered beyond the limit are queued by trigger time, providing better control over system load.

A third idempotency strategy, Fail on Running, has been added alongside the existing ReturnExisting and Fail strategies. When selected, the workflow request fails if another workflow with the same idempotency key is currently in a Running or Paused state, preventing unintended concurrent executions.

Conductor now supports defining schemas that enforce the payload structure for workflow and task inputs/outputs. This helps teams catch structural mismatches early and maintain consistency across workflow and task definitions.

The Business Rule task can now download rule files from URLs that do not have CSV, XLS, or XLSX file extensions. As long as the URL's Content-Type header indicates a supported format, the file is processed correctly.

NATS Messaging integration now supports token-based authentication via a token-based endpoint. This gives teams a secure, flexible way to authenticate using a JWT token fetched from their NATS server, without relying on credential-based methods.

Conductor webhooks now support SendGrid and Twilio as sources, enabling workflows to be triggered from email and SMS delivery events.

Orkes Conductor now supports MongoDB as a vector database integration, enabling workflows to store and query embeddings using MongoDB Atlas Vector Search.

Orkes Conductor now supports Postgres (pgvector) as a vector database integration, enabling workflows to store and search embeddings using a Postgres-backed vector store.

Conductor webhooks now support Stripe as a source, enabling workflows to be triggered from Stripe payment events.

Orkes Conductor now supports IBM MQ as a message broker integration, enabling workflows to consume and produce messages on IBM MQ queues.

Orkes Conductor now supports integrations with AWS Bedrock, including Anthropic, Cohere, and Titan models hosted on Bedrock. Teams can use these models in AI tasks without managing model infrastructure directly.

Conductor webhooks now support Microsoft Teams as a source, enabling workflows to be triggered from Teams events. 

Environment variables let you define cluster-level values that can be referenced in any workflow. Supports plain text and JSON values, with JSON fields accessible via standard path notation.

The new LLM Chat Complete task sends messages to a configured LLM and returns the model's response as task output. Supports system prompts, message history, prompt templates, and prompt variables for dynamic inputs.