Open-source AI workflow automation — your workflows are just markdown files.
Use your Copilot/Claude/Codex subscriptions · 39 integrations · Tool calling · Local LLMs · Self-hosted
Visual workflow designer — drag-and-drop nodes, AI-assisted editing, one-click execution
npm install -g marktoflow
marktoflow init
marktoflow run workflow.mdThat's it. Your workflow is a markdown file with YAML frontmatter:
---
workflow:
id: hello-world
name: Hello World
tools:
slack:
sdk: '@slack/web-api'
auth:
token: '${SLACK_BOT_TOKEN}'
steps:
- action: slack.chat.postMessage
inputs:
channel: '#general'
text: 'Hello from marktoflow!'
---Already have a GitHub Copilot, Claude, or Codex subscription? Use it directly — no API keys:
marktoflow run workflow.md --agent copilot # GitHub Copilot subscription
marktoflow run workflow.md --agent claude # Claude CLI login
marktoflow run workflow.md --agent codex # Codex CLI
marktoflow run workflow.md --agent ollama # Free, runs locally
marktoflow run workflow.md --agent gemini # Google Gemini subscription| What you get | |
|---|---|
| Markdown-native | Workflows are .md files — readable, auditable, diffable, version-controlled |
| Use your existing AI | Copilot, Claude, Codex subscriptions — no extra API keys needed |
| AI tool calling | Agentic loops — models call tools, get results, and reason autonomously |
| Local LLMs too | Optional llama.cpp, VLLM, Ollama support for air-gapped or offline use |
| Secure by design | Self-hosted, no telemetry, your data never leaves your infrastructure |
| Structured output | JSON mode and JSON Schema validation for reliable AI responses |
| MCP-first | Native Model Context Protocol support with zero config |
| Direct SDK calls | Official SDKs, full TypeScript types, no wrapper APIs |
| Parallel execution | Run multiple AI agents concurrently — 3x faster code reviews, batch processing |
| Visual editor | Optional drag-and-drop GUI with marktoflow gui |
| Cost tracking | Know exactly what each workflow run costs |
marktoflow is designed for teams and individuals who care about where their data goes.
| marktoflow | Cloud agent platforms | |
|---|---|---|
| Data residency | Your infrastructure — nothing leaves unless you send it | Data processed on vendor servers |
| Local LLM support | First-class — llama.cpp, VLLM, Ollama | Typically cloud-only |
| Air-gapped operation | ✅ Full offline support with local models | ❌ Requires internet |
| Audit trail | Git history + SQLite execution logs — fully inspectable | Vendor-controlled logs |
| Workflow transparency | Plain markdown — read every step before it runs | Opaque agent behavior |
| Telemetry | None — zero phone-home, zero tracking | Varies |
| Credential storage | Local encryption (AES-256-GCM) or your vault (HashiCorp, AWS, Azure) | Vendor-managed |
| Source available | AGPL-3.0 — inspect, modify, self-host | Often proprietary |
Use case: You want AI-powered automation but your company policy (or common sense) says sensitive data — code, credentials, customer info, internal docs — shouldn't flow through third-party agent platforms. marktoflow runs entirely on your machine or your servers with your choice of LLM.
| Feature | marktoflow | Zapier | n8n | GitHub Actions |
|---|---|---|---|---|
| Open source | Yes | No | Yes | No |
| Workflow format | Markdown | Proprietary | JSON | YAML |
| Version control | Git-native | No | Limited | Git-native |
| AI agent support | Built-in | Add-on | Plugin | Limited |
| AI tool calling | Built-in | No | Plugin | No |
| Local LLMs | Yes | No | Via plugin | No |
| Direct SDK access | Yes | No | No | Via actions |
| Visual editor | Yes | Yes | Yes | No |
| Self-hosted | Yes | No | Yes | Runners only |
| Per-task pricing | Free | Yes | Free (self-host) | Minutes-based |
| Feature | marktoflow | OpenClaw | LangChain | CrewAI |
|---|---|---|---|---|
| Self-hosted | Yes | Gateway only | Yes | Yes |
| Local LLM support | First-class | Limited | Yes | Yes |
| Air-gapped / offline | Yes | No | Partial | Partial |
| No telemetry | Yes | No | Opt-out | Opt-out |
| Workflow format | Markdown (auditable) | Opaque | Python code | Python code |
| Visual editor | Yes | No | No | No |
| 39 native integrations | Yes | No | Community | Community |
| Non-AI workflows | Yes | No | No | No |
| Data stays local | Always | Routed through gateway | Depends on LLM | Depends on LLM |
39 native SDK integrations — all with TypeScript types, retry logic, and input validation.
| Category | Services |
|---|---|
| Communication | Slack, Teams, Discord, Telegram, WhatsApp, Twilio |
| Gmail, Outlook, SendGrid, Mailchimp | |
| Google Workspace | Sheets, Calendar, Drive, Docs |
| Project Management | Jira, Linear, Asana, Trello |
| Knowledge | Notion, Confluence |
| Developer | GitHub, Airtable |
| Payments | Stripe, Shopify |
| Support | Zendesk |
| Storage | Dropbox, AWS S3 |
| Databases | Supabase, PostgreSQL, MySQL |
| Universal | HTTP client (any REST API), RSS/Atom feeds |
| AI Agents | GitHub Copilot, Claude, Codex, Gemini — use your existing subscriptions, no API keys — plus OpenAI, Ollama, llama.cpp, VLLM for local/custom setups — control via Slack/Telegram |
| Package | Description |
|---|---|
marktoflow |
All-in-one install: npm install -g marktoflow |
@marktoflow/marktoflow |
Scoped metapackage (CLI + GUI + integrations) |
@marktoflow/core |
Parser, engine, state management, plugin system |
@marktoflow/cli |
Command-line interface and workflow runner |
@marktoflow/gui |
Visual workflow designer (web UI) |
@marktoflow/integrations |
39 service integrations and AI adapters |
Production-ready workflow templates in examples/:
-
agent-task-executor — Control AI agents (Claude, Copilot, OpenCode, Ollama, Gemini) via Slack/Telegram. Send task instructions, get structured results with pass/fail status.
Example: Send "Create a React component called UserProfile with email validation" → Agent executes with safe permissions → Returns structured results (3/3 tasks passed ✅)
# Run the Slack version marktoflow run examples/agent-task-executor/workflow-slack.md # Or use Telegram marktoflow run examples/agent-task-executor/workflow-telegram.md
-
parallel-agents — ⚡ NEW: Run multiple AI agents concurrently
- multi-agent-code-review — Security + Performance + Quality reviews in parallel (3x faster)
- batch-pr-processing — Review 50 PRs in 5 minutes instead of 25
- consensus-decision — Gather diverse AI perspectives for better decisions
-
codebase-qa — AI-powered Q&A via Slack/Telegram
-
copilot-code-review — PR review with GitHub Copilot
-
daily-standup — Jira + Slack standup automation
-
incident-response — Multi-service incident coordination
-
approval-workflow — Human-in-the-loop with web forms
-
sprint-planning — AI-assisted sprint planning
📋 CHANGELOG — Version history, new features, security fixes, breaking changes
- Installation Guide
- Detailed Guide
- YAML API Reference — Complete API including ⚡ parallel execution
- REST API Guide
- Template Expressions
- Control Flow
- GUI User Guide
- GitHub Discussions — Questions, ideas, show & tell
- Contributing — PRs welcome
- Issues — Bug reports and feature requests
AGPL-3.0 — Free for personal and open source use. Commercial licensing available for organizations that cannot comply with AGPL terms. Contact scottgl@gmail.com for details.