Changelog

Event Streams message size limits are raised from 512KB (recipe) / 1MB (API) to 10MB — large JSON payloads publish natively, no splitting or trimming required.

• 10MB Max Message Size — Publish large payloads directly across both connector actions/triggers and public APIs, with consistent behavior no matter how you publish.
• Dynamic Batching in recipes — Batches are capped at 50MB total payload; when messages approach the 10MB limit, Workato automatically returns remaining messages in subsequent batches with no data loss.

Learn more in Event Streams.

Native undo and redo support is available in the recipe editor. Experiment freely, backtrack instantly, and iterate without fear — no manual reversals, no restoring to a previous version.

• Full Action Coverage — Step additions, deletions, connection swaps, field-level changes, and AIRO-made changes all flow through a unified undo/redo stack.
• Familiar Shortcuts — Ctrl+Z to undo, Ctrl+Shift+Z to redo — just like any other editor you already use.

Learn more in Recipe Editor.

Enterprises running heavy XML — SOAP services, SAP IDocs, EDI batches, legacy middleware — now have first-class platform support. Three pieces, one story.

• Native SOAP Services — Host SOAP API endpoints natively on Workato by providing a WSDL, the same way REST APIs are hosted today. The WSDL is read directly from a governed Workato Resource.
• Workato Resources — WSDLs, XSDs, and XSLTs are now versioned, governed assets living in the project folder — referenced everywhere instead of pasted inline, and portable across environments with nothing to re-wire.
• XML Connector — Resource Support — Transform, Validate, and Generate from Sample XML actions can now reference resource assets instead of inline content, with full reuse and no UI lag.
• XML Streaming — Up to 200 MB — Large XML payloads stream natively, handling SAP IDocs, EDI batches, and other high-volume XML flows without size constraints.
• Generate XML from Sample — A new action generates a working XML shape from a representative sample, enabling mapping even when the schema is non-deterministic (anyType).

Learn more in Native SOAP and Workato Resources.

Share one MCP server across your entire org and control exactly which tools each user group can call — reads for everyone, writes and destructive actions only for the roles that need them. No duplicate servers, no code changes.

• Per-Tool, Per-Group Access — Grant a group access to a server, then scope it to specific tools, so each person's agent sees only what they're authorized to use.
• One Server, No Sprawl — Share a single mixed-sensitivity server across the whole org instead of maintaining watered-down copies per role.
• Contained Blast Radius — A hijacked or over-eager agent can only reach tools within its user's authorized scope — nothing outside it.

Learn more in MCP.

✨ August’s spotlight is an AI voice-and-browser stack. Vapi lets recipes place outbound calls and control live conversations (speak, transfer, end). AssemblyAI handles the other half — upload audio, transcribe, export SRT/VTT, and run LLM prompts over transcripts, with webhooks when jobs finish. Airtop adds cloud browser sessions so agents can scrape pages, ask natural-language questions, extract lists, and click or type without running their own browser fleet.

Also this month: Clay for company/people lookups and ICP signals; Tanium for endpoint management via the Core Platform API; Azure Event Hub for high-throughput event publish and batch ingest; plus two OPA extensions — Keeper Secrets Manager for in-network secret retrieve/rotate, and Email (Custom) OPA for SMTP/IMAP/POP3 over a secure tunnel.

🔧 Enhancements lean on auth, schemas, and connectivity: OPA secure tunnels for Google Pub/Sub; OAuth/PKCE for Looker and Employment Hero; new Telegram actions and granular event subscriptions; hierarchical field picks and richer time-off filters in HiBob; $select/$expand-aware pills in OData; multi-select job status in SmartRecruiters; Qualtrics mailing lists on XM Directory; Embedded export fixes; and Ironclad Custom action with leaner non-admin OAuth. Veeva Vault continues upload and picklist polish.

🛡 Stability fixes cover OpenAPI anyOf schema rendering, Azure DevOps Service Principal token refresh after idle/expiry, and SOAP DateTime and Boolean unmarshalling.

Genies run support tasks up to 30 minutes — covering deep research, large document generation, and multi-step tool chains — with no setup or configuration required.

• Built for Longer-Running Work — The platform supports Genies running up to 30 minutes across any mix of LLM generation, knowledge base search, and tool calls.
• Real-Time Visibility — End users see agent messages, tool calls, and search calls as they happen, in the correct order, no matter how long the run takes.
• Same Guarantees for Headless API — Teams building on the Headless Genie API get the same real-time, ordered event stream as native Slack and Teams users.

Learn more in Agent Studio.

Skills can be tested directly from the skill editor — no Genie required, no tab switching, no prompt-engineering your way into a skill invocation.

• Test in the Skill Editor — Hit Test and run the skill in one window, exercising its logic directly without going through a Genie.
• No Genie in the Loop — A failed run no longer leaves a Genie stuck waiting for output that never arrived; debug and re-run from the skill itself.

Learn more in Agent Studio.

A stop button is now available in the AIRO input bar whenever AIRO is actively thinking or executing — click it and AIRO halts within a few seconds, preserving everything up to your last completed turn as context.

Learn more about AIRO.

End users can rate Genie responses with a thumbs up or thumbs down directly in the conversation — giving builders systematic, real-time signal on what's working and what isn't. On by default, no setup required.

• Native, Instant Capture — Feedback is captured directly on each response in Slack, Teams, and custom interfaces via API, with an optional short comment from the user.
• Full Context on Feedback — Drill into any flagged conversation from the conversation history tab to see the exact message and full trace behind it.
• Home Page Visibility — A feedback summary widget shows your positive/negative split and trend over a selected timeframe.
• Flows Everywhere You Already Look — Feedback data is included in conversation exports, the Dev API, and log streaming.
• Foundation for What's Next — Native feedback data feeds directly into the batch-run and automated grading phases of the broader Agent Studio evaluation framework.

Learn more in Agent Studio.

A single Genie can connect to multiple chat interfaces and triggers simultaneously — no more creating and maintaining duplicate Genies per connection.

• Multi-Connection Support — Connect one Genie to multiple unique Slack workspaces, unique Teams tenants, Workato GO, and unique custom interfaces all at once.
• Centralized Trigger Management — Chat Interfaces have moved to a new Trigger module, where all connected chat interfaces and recipe triggers are listed and managed in one place.

Learn more in Agent Studio.