Changelog

✨ A Fresh New Workflow Experience

The workflow listing on the homepage has been completely redesigned to make managing and discovering your workflows faster, easier, and more intuitive.

With the new experience, you can:

• Navigate workflows more easily with a cleaner and more modern layout
• Find what you need faster with improved organization and visibility
• Manage workflows with less friction through a more intuitive experience
• Take advantage of more powerful capabilities designed to simplify your workflow management

This update is more than just a visual refresh. We rebuilt the workflow listing experience to help you work faster and stay focused on what matters.

We think you’re going to love it.

Breaking change: PostgreSQL now uses a Docker named volume

Heym used to store its database in the ./data/postgres folder on your machine. It now stores it in a Docker volume named heym-postgres-data.

You need to migrate before your next deploy. Nothing happens automatically and nothing is deleted. Heym refuses to start rather than come up on an empty database.

Why we changed this

Docker on macOS and Windows does not pass PostgreSQL the write guarantees it depends on when the database sits in a host folder. On affected systems this shows up as:

• PostgreSQL crashing with PANIC: could not close file ... Input/output error, sometimes several times a day
• The database dropping into recovery after each crash, which stalls every workflow running at that moment
• In the worst case a table left with a missing data page, which breaks every write to it until it is repaired

Linux was never affected, but all platforms now use the same setup.

Migrating with ./deploy.sh

Run this once, then deploy as usual:

./deploy.sh --migrate-pgdata
./deploy.sh

• ./deploy.sh now stops before it builds anything if it finds the old folder, so you cannot accidentally start on an empty database.
• Your ./data/postgres folder stays where it is as a backup. Delete it once you have confirmed your workflows are intact.
• If any step fails the new volume is removed, so you can never end up deploying a half-migrated database.
• Expect a short outage while the copy and the index rebuild run.

Migrating a GHCR image setup

The release image does not include a database, so the PostgreSQL behind your DATABASE_URL is yours to move. Running & Deployment has the commands step by step.

Two things to watch:

• Give each Heym deployment on the same machine its own volume name. Two instances pointed at heym-postgres-data would end up sharing one database.
• Start the migrated database with the same PostgreSQL image you were using before. Extensions such as pgvector live inside the image rather than in your data folder, so a database that uses them will fail under a plain postgres:16 image.

Why the migration also rebuilds indexes

Copying a data folder copies it exactly, including any damage the old setup left behind. That damage usually takes the form of index entries pointing at rows that are no longer there, and pg_amcheck reports a clean result even when it is present, so a passing check is not proof your database is healthy.

--migrate-pgdata therefore rebuilds every index after the copy. That repairs those entries, though it cannot bring back data pages that were already lost.

If your database is large and you are confident it is healthy, skip the rebuild with ./deploy.sh --migrate-pgdata --skip-reindex.

Using ./run.sh for local development

Nothing to do. Your existing database container and its data are reused exactly as before.

Full details in Running & Deployment.

Security

Fixes GHSA-6x65-w7q7-wg93: webhook header auth secrets, MCP API keys, portal session tokens, and Discord interaction tokens were stored or returned in plaintext where a database read or a collaborator could recover and replay them. Reported by @SashaMIT.

Breaking: MCP no longer accepts credentials in the query string. Move ?key= and ?token= to the X-MCP-Key or Authorization header before upgrading. The ?session= parameter used by the SSE handshake still works.

Behavior: MCP keys are now hashed at rest and shown once, when you generate or regenerate them. Existing keys keep working, but the tab can no longer display them.

Upgrading: Run migrations as usual. They backfill existing keys and sessions in place, so nothing already issued breaks. They are one-way, so back up first if you might roll back.

What's New in Heym: Alerts

Heym now includes an Alerts tab to help you catch workflow issues early and monitor performance more proactively.

Alert Types

Alerts are evaluated over configurable time windows, helping reduce noise from short-lived spikes.

• Error Count: Track accumulated errors
• Run Duration: Monitor max, average, and p95 execution times
• Token / USD Spend: Set cost thresholds
• Execution Count: Monitor workflow volume

Secure by Design

Alerts can target a specific workflow or all workflows the owner can access. Metrics are never exposed for workflows a viewer does not have permission to see.

Reliable Alerting

The alerting system is designed to prevent duplicate or lost alerts, handle fire and recovery states, and keep failures isolated.

Metrics use the same pricing and analytics systems as Traces and Analytics, so data stays consistent across Heym.

Alert notifications also include contributing workflows and their share of the metric, making issues easier to investigate.

Get Started

Open the Alerts tab in Heym, choose a workflow and alert type, then set your time window and threshold.

Stay ahead of workflow issues before they become bigger problems.

Watch the v0.0.88 demo video

Release notes

Breaking change: file conversion moved from the Drive node to the Converter node

The Drive node's convertFile operation has been removed. Converting a file to another format is now a Converter node with conversion: "fileConvert".

Update any workflow that used it:

Behavior and outputs are unchanged. The same formats are supported, the source file is still left untouched, the result is still stored as a new Drive file, and $label.id, $label.filename, $label.mime_type, $label.size_bytes, and $label.download_url still read the same way. There is no data migration; only saved workflow definitions need editing, and a workflow left on the old operation fails with an unknown operation error.

Drive keeps everything else it did: storing, fetching, downloading, sharing, and expiring files.

New: OCR in the Converter node

The Converter node gained imageToText and pdfToText, backed by Tesseract. Point converterFileId at a Heym Drive file and it returns the recognized text.

Language defaults to auto, which detects the script on the page and picks the matching model, so a Turkish invoice and an English one both work without being told which is which. You can also name a code such as tur, or combine several with eng+tur. Output is UTF-8 with optional NFC normalization, and narrower charsets such as cp1254 are available when a downstream system needs one. pdfToText rasterizes and reads every selected page, so scanned and digital PDFs behave the same way.

Tesseract and poppler ship in every Heym image, and run.sh installs them for native development runs. There is nothing to configure.

Security

MCP stdio servers now run in a hardened, throwaway container instead of on the backend host, with no Docker socket, no backend environment secrets, and no caller-supplied mounts (GHSA-378x-q589-34mv). Isolation is controlled by HEYM_MCP_STDIO_SANDBOX. Thanks to @sajdakabir for the report and the follow-up review.

Security

• Fixed an SSRF in the HTTP node (blocks internal/metadata targets, pins IPs against DNS-rebinding/redirect/proxy bypasses; opt out with HEYM_HTTP_ALLOW_PRIVATE_URLS). Thanks @0neOfU4 (GHSA-8wj7-v2w6-wfcx)
• Removed the setup command field from the Codex and OpenCode Go nodes to close a backend shell-exec surface; existing values are ignored. Thanks @0neOfU4.

Security: fixed a Playwright node backend RCE where custom code could run arbitrary Python for any authenticated user, so custom code is now off by default and sandboxed when enabled (GHSA-mp23-7m6r-jfw4). Thanks to @EQSTLab and @min8282 for the responsible disclosure. We recommend upgrading to this release.

What's New in Heym 🔥

Agentic Kanban Board for Loop Engineering

We're excited to introduce the Agentic Kanban Board, a powerful new feature designed specifically for managing AI agent workflows. This visual workspace transforms how you orchestrate multi-agent systems, bringing the same clarity and control to AI workflows that Kanban has brought to software development for decades.

Key Improvements:

Benefits:

The Agentic Kanban Board integrates seamlessly with Heym's existing visual workflow builder, MCP connectors, and multi-agent orchestration capabilities. Start automating your loop engineering workflows today.

board.webm

Security

Security: Agent skill execution hardened (GHSA-hcv7-mg77-pg73). Untrusted Python skills now run in an isolated, non-root, fail-closed Docker sandbox with no Docker socket or backend secrets. Network access, generated files, and Heym Drive remain supported. Self-hosted and multi-tenant deployments should upgrade to 0.0.65. Thanks to @sajdakabir for the responsible disclosure.

What's New: Jira Node

Bring Jira into your workflows and automate issues end to end.

• New Jira node with 20+ operations: search issues by JQL, create, update, delete, changelog, notify, comments, transitions, attachments, and user management.
• Works with both Jira Cloud and self-hosted Jira Data Center / Server, selectable per credential.
• Text fields adapt automatically: rich ADF on Cloud v3, plain text on Data Center v2.
• Every field supports expressions and AI autofill, and can be exposed as an agent tool.
• Jira credential with a built-in Test Connection check, masked secrets, and clear labels for Cloud email/token or Data Center username/password.
• Full docs, unit tests, and end to end coverage included.

Thank you so much @eryue0220 once again 💯

What's New in Heym: OpenAI Codex Node

We're excited to announce the native OpenAI Codex node for Heym's workflow builder.

This addition brings fully automated, agentic coding capabilities directly into your visual workflows, enabling you to trigger autonomous coding tasks from cron jobs, webhooks, Slack messages, or GitHub events.

Authentication Made Simple

The Codex node introduces a groundbreaking authentication option.

Instead of managing OpenAI API keys and paying per token, you can now authenticate using your existing ChatGPT Plus or Pro subscription through a secure OAuth PKCE flow.

Heym securely stores and auto-refreshes your token bundle, eliminating surprise API bills while maintaining full functionality.

Traditional access-token mode remains available via an auth-mode toggle for teams that require it.

Seven Publish Modes

Choose exactly how your changes ship with flexible publish options:

• diff_only: Edit locally and return the patch with changed files, nothing pushed
• draft_pr: Commit, push, and open a draft pull request
• open_pr: Commit, push, and open a review-ready pull request
• commit_push: Commit and push a branch without creating a PR
• direct_commit: Commit straight to the base branch
• update_existing_pr: Add a commit to an existing branch or PR, opening one if none exists
• patch_artifact: Save the diff as a downloadable file and return a patchUrl

Agent Tool Integration

Attach the Codex node to any AI Agent so the agent can delegate coding tasks dynamically.

All node fields, including task prompt, repository, and branch, can be populated by the agent at runtime, enabling sophisticated multi-agent workflows.

Human-in-the-Loop Workflows

When Codex requires a product decision or missing information, it returns a needs_input status and pauses execution with a shareable follow-up link.

Route the question output to Slack, email, or Discord for team input.

Answering resumes the same Codex thread and execution, ensuring you maintain control without becoming a bottleneck.

Structured Results

Every Codex execution returns comprehensive structured output, including:

• A summary of changes
• Validation notes
• The complete git diff
• A list of changed files

This enables downstream nodes to process results programmatically and integrate with your existing CI/CD pipelines.

Isolated Execution

Codex clones repositories in an isolated runner environment, ensuring safe and predictable execution.

The node handles the complete lifecycle from repository cloning through code generation to result packaging.

Why It Matters

This release represents a significant step forward in workflow automation, enabling teams to save an average of 3.6 hours per developer per week on repetitive coding tasks while maintaining full control over how changes are integrated into their codebases.

Security

• Template get-by-id IDOR (GHSA-5748-x76g-v68m) and team member management privilege escalation (GHSA-vxpw-x7j7-8723), both reported by @sajdakabir. Thank you.

Heym adds Canvas Execution Highlights, a new debugging and observability layer for visual workflows.

After an execution completes, highlighted node outputs can be displayed inside a searchable and collapsible canvas panel. The feature supports live executions, restored execution history, and runs initiated through dashboard widgets.

For developers building complex agent and LLM workflows, this provides a faster way to understand what happened during execution without inspecting every node individually.

🔌 New in Heym: Plugins are here! You can now extend heym with plugins ! Add new tools and capabilities to your canvas / workflows in just a few clicks. No setup headaches, no code required. Build your own , zip it and deploy !

Live now at the Heym website

New Nodes: Linear and ClickHouse

This release adds two new integration nodes to Heym.

Linear

Automate Linear workflows directly from the canvas:

• Manage issues, comments, teams, projects, states, and members
• Create, update, list, and delete Linear records
• Use Linear data in downstream workflow steps

ClickHouse

Bring ClickHouse data into your automations:

• Connect to ClickHouse
• Run queries from workflows
• Use query results in AI and automation flows

Security Fixes : Security: fixed an authenticated sandbox escape in the workflow condition evaluator that allowed arbitrary Python and OS command execution on the backend host (GHSA-pm6h-x3h5-j38h). Condition expressions now run through a hardened, AST-based evaluator that blocks attribute traversal, and Slack/Telegram webhook verification now fails closed. Reported by @okcomputerfan (also known as pixileaf). Thank you.

What’s new in Heym 🤗 : full Notion integration with internal token and OAuth credentials, workflow support for pages, databases, data sources, and blocks, plus expression-powered Notion node inputs and editor discovery tools.

Thanks for the great contribution! @eryue0220

RAG just got simpler in Heym 0.0.50.

You can now run vector stores on Postgres with pgvector, not just Qdrant. Vectors live in Heym's own database, so there is no extra service to deploy.

Pick your backend per store, point a RAG node at it, done. Same insert, search, metadata filters, and reranking either way.

Existing Qdrant setups keep working, nothing to migrate.

Two new things in Heym this week.

1. The AI assistant now plans before it builds. Give it a vague request and instead of guessing, it asks you a few quick questions (tap a chip or type your own), then builds the workflow from your answers.

2. A workflow analysis tool that reviews your runs and tells you what to improve first.

Less guessing, better workflows.

Heym news 🚀 A GitHub node has been contributed to Heym. Thanks for the support! 🙌 @eryue0220

🚀 New in Heym: the Supabase node is here.

Query and mutate your Postgres tables right inside your workflows. Select, insert, update, upsert and delete with operator filters, logical AND/OR groups, auto table and column discovery, and a one-click connection test. No custom HTTP wiring needed.

#Heym #Supabase #PostgREST #AIworkflows #automation #nocode

This version adds the ChatGPT connector to Heym, allowing OpenAI users to invoke Heym workflows 🎉 DataTable improvements added and Heym introduce support for Mongo-like operators such for count / find operations.

New in Heym :

Build database tables in seconds. Just describe what you need or paste JSON, and Heym's AI generates the columns for you. Review, tweak types and constraints, then save.

No more manual schema setup. ✨

New in Heym 🚀 : Build a complete observability dashboard in Heym without leaving the canvas.

Describe any metric in your own words, in any language, and our AI will generate the right widget: bar, line, area, pie, gauge, scatter, proportion, or bar gauge. Then refine any widget with a single prompt.

Big thanks to @eryue0220 for their amazing aws/s3 node contribution!

New : Discord nodes are here. Heym now supports Discord incoming webhooks for sending channel messages, plus Discord interaction triggers for slash command style workflows with verified signatures and follow-up replies from workflow output. Thx @eryue0220

Security: fixed an authenticated sandbox escape in user defined Python tools that allowed OS command execution on the backend host (GHSA-wcgw-9hfw-f6f2). Python tools now run inside a hardened, isolated container by default. Reported by @jashidsany. Thank you.

Heym now speaks OpenTelemetry. Every workflow run becomes a trace: one span per run, one per node, with model and token usage on each. Send it to Jaeger, Tempo, Honeycomb, or any OTLP backend. No instrumentation code, just one env var.

Heym now has a runbook. Press play and watch the canvas build and run a workflow on its own, cursor and all. The fastest way to understand AI-native automation is to just watch it happen.

Breaking change: the Docker Logs tab is now opt-in

Starting this release, the Docker-backed Logs tab is disabled by default. Previously any signed-in user could read container logs. Access is now gated behind two settings.

To re-enable it, set:

DOCKER_LOGS_ENABLED=true DOCKER_LOGS_ALLOWED_EMAILS=admin@example.com (comma-separated list of trusted user emails) For safety, create the trusted admin account before enabling this, or keep ALLOW_REGISTER=false, so an unverified self-registration cannot claim an allow-listed email.

In this release we merged many commits to the repo by community(devs) .

That makes us incredibly happy. @ckakgun

Thank you for being part of this journey. We are lucky to have you. 🧡

⚠️ Breaking change · Security SECRET_KEY and ENCRYPTION_KEY are now required and validated at startup

The application no longer accepts publicly known default secrets. On startup, it refuses to boot if:

• SECRET_KEY is empty, too short, or still set to the old public default (your-super-secret-key-change-in-production-min-32-chars)
• ENCRYPTION_KEY is empty or still set to the old public placeholder (change_this_to_a_random_32_byte_hex_value)

This closes vulnerabilities where deployments left on public defaults could have authentication tokens forged, or stored credentials encrypted under a publicly known key (CWE-798).

Who is affected

Any deployment that never set its own SECRET_KEY or ENCRYPTION_KEY, or copied the old example values. These instances may fail to start after upgrading until valid secrets are configured.

Deployments already using strong, unique values for both keys are not affected.

Required action before upgrading

Generate a strong SECRET_KEY:

python -c "import secrets; print(secrets.token_urlsafe(32))"

Generate a strong ENCRYPTION_KEY:

python -c "import secrets; print(secrets.token_hex(32))"

Set them in your environment:

• Docker / Compose: set SECRET_KEY and ENCRYPTION_KEY in the host environment or the .env file next to docker-compose.yml.
• Manual / run.sh: set SECRET_KEY and ENCRYPTION_KEY in your .env.

For fresh manual installs, run.sh / deploy.sh can auto-generate missing empty keys. Existing deployments using the old ENCRYPTION_KEY placeholder will fail loudly instead of silently rotating the key.

Restart the application. With Compose restart: always, a missing or rejected key may cause a crash loop until valid values are provided.

Impact of changing SECRET_KEY

Changing SECRET_KEY invalidates all previously issued tokens, including:

• User access and refresh tokens
• Long-lived workflow HTTP bearer tokens used by curl or external integrations
• Temporary OAuth state, MCP session, and Playwright execution tokens

Users will need to log in again. Existing external integrations using workflow bearer tokens will receive 401 Unauthorized until their tokens are revoked, recreated, and updated in the calling systems.

Impact of changing ENCRYPTION_KEY

Changing ENCRYPTION_KEY can make previously encrypted stored credentials unreadable. No database rows are deleted, but affected credentials may need to be re-created or reconnected.

You can now talk to your AI workflows in Heym.

New in Chat: 🔊 Read any message out loud 🎙️ Hands-free voice mode (you talk, it talks back)

Integration with ElevenLabs. Automatic language detection, works in any language, dark mode and mobile ready.

--

Your favourite AI workflow platform just learned how to shapeshift files

generate a report as markdown, deliver it as PDF. capture a screenshot, compress it to JPEG. extract text from a PDF and turn it into a Word doc.

one node. any format. fully automated.

Heym Drive just got a whole lot more powerful.

New in Heym Traces: full LLM cost observability, built in.

Tokens by model, USD spend, latency, error rate, all filterable by time range (1h, 24h, 7d, 30d, all). Per-model pricing syncs from Helicone for 1,100+ models, with per-user overrides and custom model support.

Applies security deps update & license updates

Heym tutorials playlist on YouTube

Shipping a smoother Chat tab in Heym today.

Your Chat / AI assistant keeps working after you close the tab.

Kick off a long-running task, close the browser, come back whenever. The progress bar is still spinning, and the moment it finishes you get a notification dot and a soft ding.

Small details, big flow upgrade.

Just shipped a smoother Drive experience in Heym.

Workflows can now list Drive files with getAll, return metadata like name, size, type, source, and download URL, and let AI agents fill File IDs when using Drive as a tool.

Also made drag and drop uploads work across the Drive page.

Watch the v0.0.19 demo video

MCP Call (Node client): Direct MCP tool invocation is supported, so tools exposed by MCP servers can be called from the Node client without extra glue.

Just shipped multi-MCP server support for Heym.

Create multiple named MCP endpoints, assign specific workflows to each, and connect different AI clients to different toolsets from one platform.

1000 workflows? No problem. Cluster them however you want.

New in Heym: Build expressions with AI : Natural language → valid $node / $vars DSL, wired to your last run + variables. Wrong preview? Hit Regenerate; it sees the error and output and tries again under the hood 🤗

AI agents in Heym can now use workflow nodes as tools.

Connect a node to an agent, mark which fields the agent should fill at runtime, and keep fixed values where you need control. It works with tool parameters, Set mappings, output fields, and tidy canvas layout for agent tools and sub-agents.

Added MCP workflow trace linking so executions launched through MCP carry the correct user trace context. This restores trace links for LLM and agent spans when viewing historical MCP runs on the canvas. Improved the mobile chat drawer behavior. The chat sidebar now starts closed on mobile, responds to viewport changes, and avoids hover-open behavior on small screens. Improved chat input resizing. The message textarea now auto-resizes consistently after typing, voice transcription, transcription correction, and sending.

Watch the v0.0.7 demo video

Heym AI Builder just met chat.

Same workflow generation engine, now inside a conversational interface.

Describe what you want, iterate in chat, preview the canvas, open the workflow when ready.

Building automations is starting to feel less like configuration and more like conversation.

v0.0.1 — Initial Public Release

We're open-sourcing Heym today. This is the first public release of a self-hosted, source-available AI workflow automation platform we've been building in Berlin.

What ships in this release

Visual Workflow Canvas

Drag-and-drop node editor. Workflows are directed acyclic graphs — independent nodes run concurrently, dependent nodes resolve automatically. Streaming mode emits execution events in real time.

AI Assistant

Describe a workflow in plain English. The assistant generates the full canvas — nodes, connections, pre-filled properties. Supports voice input and streaming output. Works with any configured LLM credential.

38 Node Types

Triggers, AI, logic, integration, and control nodes. Full feature reference: [features.md]

Multi-Agent Orchestration

Multiple Agent nodes on the same canvas, each with its own model, tool set, and Python execution environment. Chain, branch, or run conditionally. Automatic context compression keeps long-running agents stable.

Built-in RAG

Vector store management powered by Qdrant. Upload documents, create stores, wire semantic search directly into workflows. Full pipeline visible in a single trace.

MCP Server

Every Heym instance exposes a built-in MCP Server. Any workflow can be exposed as an MCP tool — callable from Claude Desktop, Cursor, or any MCP client. Agent nodes can also connect to external MCP servers as tool sources.

Human-in-the-Loop

Pause execution at any checkpoint, generate a public one-time review URL, resume from exact stored snapshot. The same run can pause more than once.

Portal

Turn any workflow into a public-facing chat interface with a single toggle. Custom URL slug, streaming responses, file uploads, per-user authentication.

Traces

Every LLM call, agent step, token count, and latency logged automatically. Filter by source, model, or credential.

Skills System

Portable capability bundles: a SKILL.md instruction file + optional Python tools + assets. Drop onto any Agent node. AI Build assists with drafting and iterating skills.

Scheduled Tab

Full calendar view (day / week / month) of all active cron workflows.

Chat Tab

Direct LLM interface built into the dashboard. Run workflows by name, test prompts, query your data conversationally.

Drive

Built-in file storage for workflow outputs. Files generated by agents land here automatically.

Templates

Ready-made workflows: Gemini image creator, PDF→DOCX pipeline, Cursor/Claude blog crawler, Jina website reader.

Also ships

Auto Heal, Parallel Execution, Span Timeline, Error Handling & Retry, Guardrails, Variables, DataTable, Teams & Credentials.

Deployment

Single docker-compose up to get started. Kubernetes also supported.

License

MIT + Commons Clause. Source-available, free to use, not available for commercial resale without separate terms.

Links

• Site: the Heym website
• Discord: join from the site