Activepieces vs n8n (2026)

Several hundred documented app nodes, HTTP, GraphQL, SSH and code nodes, community nodes and an MCP Client node give agents tools across every major system class.

If, Switch, merge, loop, wait and error handling nodes control flow, and sub-workflows can be attached to agents as tools.

Breadth spans app events, schedules, webhooks and human initiated interfaces across the full piece library, which is why this moves from P.

Sixteen named trigger nodes cover schedules, webhooks, chat, email, RSS, forms, files and MCP, and a durable scheduler keeps scheduled runs through restarts.

Corrected N>F on 29 September 2026: the August None found no knowledge feature; the agent docs now document a vector-backed knowledge base over uploaded files and tables.

Vector stores, embeddings, retrievers and document loaders attach as sub-nodes, so the retrieval index can live in a customer-run database.

Corrected F>P on 29 September 2026: the August Full rested on Tables holding state between runs, which is a database the builder uses rather than an agent memory layer meeting the Q78 clauses.

Memory attaches as a sub-node with session IDs, but the vendor states it does not persist between sessions unless an external chat-memory backend is wired in.

Tools can require human approval before an agent runs them, a Guardrails node enforces policy, and workflow reviews gate publishing.

VERIFICATION FLAG: a SOC 2 Type II for the managed cloud is reported by several third party reviews but was not retrieved on any first party page; the site publishes no trust centre or compliance page. Graded F on the documented control surface, which is deep and named, with the attestation searched for and unconfirmed. Governance is paywalled to Team and Ultimate, which is a commercial gate rather than an absence of capability.

Annual independent SOC 2 audits, a public SOC 3 report and yearly penetration tests sit beside SSO, SAML and LDAP on Enterprise, role-based access on paid plans and a deep set of documented security controls.

Every run is an execution the customer can open and filter, traced node by node to an OpenTelemetry collector, with Prometheus metrics and redaction for compliance.

Self-hosting is documented across Docker, npm and seven named cloud providers, with an isolate configuration that stops the instance contacting n8n's servers, beside managed n8n Cloud.

Corrected N>P on 29 September 2026: the August None said no template library was offered as a product surface; the docs now document one, with categories but no named templates on the pages read.

A template library, a custom templates library for self-hosted instances, the AI Starter Kit and portable .n8np packages give ready-made starting points.

Customer chooses the model and the provider, and can route through their own gateway; this is the buyer facing choice surface the F bar requires, not vendor internal routing.

Agents require a Chat Model sub-node chosen from OpenAI, Anthropic, Google Gemini, Mistral, Perplexity, Qwen Cloud, MiniMax, Moonshot Kimi or local models, and each step can use a different model.

The MCP Server Trigger makes n8n an MCP server for outside clients, beside a public REST API, custom and community nodes, portable .n8np packages and a documented OEM embedding path.

Real customer facing testing of flows, but scoped to workflow debugging rather than agent evaluation; no scoring, regression suite or eval product.

Dedicated Evaluation and Evaluation Trigger nodes run quick tests against known cases in development and metric-based scoring in production.

Downgraded for consistency with the axis rule: Comp is non zero only when the agent operates software exposing no programmatic interface. Everything here runs through APIs, and the prior P carried no basis at all.

The Airtop app node wires in a third-party cloud browser that clicks, fills forms, types and scrolls; n8n ships no browser of its own.