Your team outgrows "just use Zapier" the week an invoice line item hits four figures and compliance asks where customer data touches the automation vendor. You still need non-technical teammates to wire up flows. You just want credentials and run history on a box you can name in an audit.
Activepieces (~25k GitHub stars) markets itself as an open-source Zapier-style platform with an AI angle: visual flows, 280+ integrations called pieces (mostly TypeScript npm packages), and those same pieces doubling as MCP servers for Claude Desktop, Cursor, or Windsurf. Community Edition is MIT; enterprise-only features ship under a separate commercial license. Self-hosting is first-class in their docs, not an afterthought.
What you are running
- Flow builder — loops, branches, retries, HTTP, code steps with npm, versioning, templates, translations; chat and form triggers for human-in-the-loop.
- Pieces ecosystem — Google Sheets, OpenAI, Slack, Discord, and hundreds more; roughly 60% community-contributed per upstream claims.
- AI inside the builder — native AI pieces and an SDK for agents that help author flows; "ask AI" in code steps for non-dev data cleanup.
- MCP bridge — each piece can expose tools to LLM clients, which is the headline for agent-first shops in 2026.
- Approvals and delays — human-in-the-loop built as pieces, not a separate product SKU.
Compare mentally to tools we have already written up: n8n is the incumbent node graph; Windmill leans code-first and scriptable. Activepieces optimizes for approachable UI plus MCP and piece hot-reload for TypeScript authors. Pick on who builds the flows and whether MCP matters to your stack.
Why self-host?
Data stays inside your boundary. OAuth tokens, payload bodies, and logs live on your Postgres/Redis (or your single-container volume), not a multi-tenant US region by default.
Network-gapped option. Upstream explicitly pitches self-hosted, air-gapped style deployments for orgs that cannot route CRM webhooks through SaaS.
Customize pieces. Fork or publish private npm pieces when a connector does not exist; your flows still version in the UI.
Canadian VPS angle. Beauharnois or Toronto hosting plus AP_FRONTEND_URL on your domain gives webhooks a stable Canadian entry point for teams documenting data location.
License clarity. MIT Community Edition is friendly for internal automation. Check the EE/commercial boundary on their pricing page before you need SAML, advanced roles, or other paid-tier features.
Installation paths
Hobby / trial (one container):
docker run -d --name activepieces -p 8080:80 \
-v ~/.activepieces:/root/.activepieces \
-e AP_DB_TYPE=PGLITE \
-e AP_REDIS_TYPE=MEMORY \
-e AP_FRONTEND_URL="https://flows.yourdomain.ca" \
activepieces/activepieces:latestEmbedded PGLite and in-memory queue: fine for a laptop, not for production scale or Enterprise upgrade paths (their docs warn you cannot migrate hobby DB to Enterprise later without starting over).
Production self-host: use their Docker Compose stack with PostgreSQL and Redis (curl -fsSL https://get.activepieces.com | sh or the docs install assistant). Set AP_FRONTEND_URL to your real HTTPS URL so webhook signatures line up. Expose 8080 (or your chosen port) behind nginx or Caddy with TLS.
Who it's for (and who should skip it)
Good fit: ops and marketing teams wanting Zapier ergonomics without per-task SaaS tax; dev shops that will write TypeScript pieces and wire MCP to Cursor; Canadian SMBs centralizing integrations on one VPS.
Skip it: if you only need cron plus bash (use systemd); if you need hard air-gap but cannot run Postgres/Redis; if you want pure-code workflows only (Windmill may fit better).
Hosting in Canada
Production Activepieces wants Postgres, Redis, and enough RAM for workers when flows spike. Backup the database and volume holding execution artifacts; rotate encryption keys if you store OAuth secrets.
Talk with Swift Host for Docker Compose sizing, TLS in front of webhooks, and backup scope on a Canadian CloudBox. We are in St. Albert and actually answer the phone.
