Changelog

Add notes directly on the workflow canvas to document what a flow does — what it's for, why a branch exists, what an input expects. Useful when workflows get long or when someone else has to read them.

A single workflow can now have more than one trigger. Add and remove triggers without rebuilding the flow around them.

• Canvas lock toggle prevents accidental node dragging
• Paste a data pill path into any property input and it resolves as a real data pill
• Workflow issues now point to the exact node, not the workflow level
• Warnings use a readable color and are listed separately from errors
• Output tab flags when it's showing placeholder sample data, not real output
• Execution filter shows which workflow version each run used and matches all versions
• Webhook testing fires the trigger you picked, not always the first one
• Fork/Join and Branch cases no longer trigger false task-order warnings

LLM components now expose Thinking and Reasoning effort settings. Configure how much the model reasons before it responds — relevant for complex tasks where output quality matters more than speed.

SSH: run commands on remote servers over SSH directly from a workflow step.

Neon: connect workflows to Neon serverless Postgres.

ByteChef runs in two modes.

Automation: connect your tools and automate work across 200+ integrations.

AI Agents: configure agents with a model, memory, a knowledge base, and live tool access. Deploy them into workflows, embed them in your product, or expose them via API.

ByteChef now runs on H2 out of the box. No database to provision, no credentials file, no configuration step before you reach the editor. Run one Docker command — on amd64 or arm64 — and you're in.

docker run -p 8080:8080 bytechef/bytechef

A lot of what previously required an Enterprise license is now available to all users:

• User management and invites
• AI Providers and AI Skills
• Streaming chat and the Form trigger
• AI Agent tools
• Project and workflow sharing

If you've looked at ByteChef before and hit a feature wall, it's worth another look.

Execution logs that tell you what actually happened

When a workflow fails, you need to know why — fast. Logs now show readable values per task, stored durably. Failed triggers get their own rows. Formula errors surface instead of going quiet. Every node shows which component version it ran.

New connectors

Beamer, Cal.com, Browser Use V4, and ScrapeUnblocker. Plus a Google Sheets modified-row trigger, and a corrected New Row trigger — it now detects rows by content rather than position, so reordering a sheet no longer fires false inserts.

What you can build

ByteChef runs in two modes.

Automation — connect your tools and automate work across 200+ integrations.

AI Agents — configure agents with a model, memory, a knowledge base, and live tool access. Deploy them into workflows, embed them in your product, or expose them via API.

AI workflows become more flexible with expanded provider support and smarter defaults.

AI Highlights

• Embeddings support across AI providers
• New LiteLLM provider
• Automatic model selection for every provider
• Improvements to the Copilot provider catalog

Whether you're switching providers or experimenting with new models, ByteChef makes it easier to build AI workflows without changing how you work.

Managing Model Context Protocol servers is now cleaner and easier to navigate.

What's new

• Redesigned card-based MCP components
• Organized tool rows for better readability
• Server-side keep-alive connections
• Visibility into every active MCP connection

Less clutter, better visibility, and a smoother experience when working with MCP servers.

We focused heavily on eliminating the kinds of issues that interrupt your flow.

This release fixes editor freezes, improves layout behavior, and makes the workflow builder more resilient when dealing with invalid expressions, duplicate node names, and stale components.

Combined with dozens of UI refinements, the result is an editor that feels faster, more stable, and more predictable.

• New integrations: Zenrows, Attio, and WordPress
• Faster Knowledge Base loading with lazy-loaded document chunks
• More reliable MCP connections and timeout handling
• Better approval workflows and workflow inputs
• SDK and embedded experience improvements
• Google Mail connection fixes
• Multi-tenant, performance, and reliability improvements across the platform

Undo & Redo

Mistakes happen. Now they're one click away from disappearing. Undo and redo support is now available throughout the workflow editor.

Faster editing

Building workflows is now smoother with:

• Cut nodes directly from the editor
• Improved node context menu
• Cleaner node hover states
• "See Executions" shortcut for Subflow executions
• Fixed ComboBox layering issues

• Knowledge Base improvements
• Cleaner AI Skill creation experience
• Updated AI Providers page

PagerDuty

Respond to incidents and automate operational workflows with the new PagerDuty integration.

Also included

• Read File as Bytes action
• Connector improvements for Zendesk, Ollama, Snowflake and MailerLite

This release also improves platform stability, including:

• Fixed execution details freezing on large outputs
• Better handling of connectionless nodes
• OAuth connection improvements
• Trigger form fixes
• Scheduler reliability improvements

The AI Agent is a new workflow node that lets you configure a fully custom AI agent from scratch - with both a Simple and Advanced mode.

Each agent is built from four building blocks you assemble directly in the workflow editor: a model, memory, a knowledge base, and tools. You choose the LLM, connect your integrations as callable tools, and write a system prompt that defines how the agent behaves.

The result is a deployable agent that reasons, retrieves, and acts - without any custom backend code.

CRM & sales

Zoho, Copper, Freshsales, Nutshell, Apollo, Hunter and more

Marketing & forms

Mailchimp, ActiveCampaign, Typeform, Jotform, SendFox, Amplitude

Productivity

Canva, Airtable, LinkedIn, Teamwork

AI & web

ElevenLabs, Firecrawl, Brave Search

Database

MongoDB + MongoDB Atlas Vector Search

Comms & infrastructure

SendGrid, Infobip, FTP

• AI skill management moved to the Community Edition - create, update, and delete skills, including file attachments, regardless of your plan
• All major LLM providers verified end-to-end: Anthropic, OpenAI, Azure OpenAI, Gemini, Amazon Bedrock, Mistral, Groq, NVIDIA, Stability, OpenRouter, Perplexity, and DeepSeek
• Microsoft integrations expanded: SharePoint recursive folder listing and file download, MS Teams reply trigger, Outlook batch email trigger
• Security hardening in the execution engine: SpEL injection defenses, path containment in file storage, resume-token verification
• Frontend performance improved with JS compression, immutable asset caching, and lazy bundle loading