Features

m9m ships with roughly 95% of n8n's backend feature surface, plus a first-class agent runtime, and a handful of things n8n doesn't have.

Workflow runtime

• 32 built-in node types — HTTP Request, Set, Switch, Merge, Filter, Item Lists, Sub-workflow, and 25 more.
• n8n-compatible workflow JSON — import and run your existing exports unchanged.
• Expression engine — full {{ $json.field }} / {{ $node["name"].data }} syntax with the standard string / math / date / array built-ins.
• Triggers — cron schedules, webhooks, manual CLI invocation.
• Custom logic — inline JavaScript and Python when a built-in node isn't enough.

Integrations

• Databases: PostgreSQL, MySQL, SQLite.
• Object storage: AWS S3, Google Cloud Storage, Azure Blob.
• AI: OpenAI (GPT-4 and beyond), Anthropic Claude, and sandboxed CLI agents (Claude Code, Codex, Aider).
• Communication: Slack, Discord, SMTP email.
• Code platforms: GitHub, GitLab.
• Productivity: Google Sheets.
• Universal: HTTP Request node + webhook receivers for any API.

Agent orchestration

• CLI nodes — spawn Claude Code, Codex, and Aider in sandboxed Linux namespaces with CPU, memory, and network limits.
• MCP integration — 37 tools exposed over Model Context Protocol for AI-driven workflow management.
• Human-in-the-loop — checkpoints that pause a workflow for review and resume on approval.
• Multi-model pipelines — chain OpenAI, Anthropic, local models, and human steps in a single workflow.

Operations

• Git-based workflow versioning — workflows are JSON; version them like code.
• Audit logs — every run, every node execution, every credential use.
• Multi-workspace — isolate workflows, credentials, and runs per team or tenant.
• Credential management — encrypted at rest, environment-variable substitution, OAuth2 flows.
• Queue backends — in-memory, Redis, or RabbitMQ.

Observability

• Prometheus metrics — per-node, per-workflow, per-queue.
• OpenTelemetry tracing — one span per node, propagated across sub-workflows.
• Structured logs — JSON-formatted, run-scoped.

Extensibility

• Custom nodes in Go — interface-driven, registered at compile or load time.
• SDKs — Go, Python, and Node.js for programmatic workflow execution.
• Plugin architecture — load nodes from external binaries or shared libraries.

Performance

• ~500 ms cold start, ~150 MB RAM, 300 MB container.
• 500 concurrent workflows per process.
• 5–10× faster execution than Node-based alternatives on typical pipelines.
• Run m9m benchmark to reproduce on your hardware.