Research report · verified 22 Sep 2026
Four layers for open-source marketing work: a starter shortlist.
The useful question is not “which agent is best?” It is “which layer owns the job I need—and what must I operate?” This report compares four candidates by execution boundary, license evidence, setup and operations.
Decision rule
Choose the lowest layer that can safely execute the job. Use Mautic for an owned marketing system; Activepieces CE for cross-system triggers and transformations; Marketing Agent OS for planning/research in an existing harness; Loopmark only for a constrained individual drafting experiment. Combine layers only after a named owner, review step, cleanup route and operating budget exist.
At-a-glance comparison
| Candidate | Use it for | License status | Start with | Decision |
|---|---|---|---|---|
| Mautic Platform | Owned contacts, forms, campaigns, landing pages and lifecycle email. | GPL-3.0-or-later; OSI-style for the named component. | Database, PHP/web environment, HTTPS, mail transport and cron/queue operations. | Conditional fit for teams able to operate a self-hosted marketing system. |
| Activepieces CE Orchestrator | Auditable flows across existing CRM/forms/email/analytics systems. | MIT for CE/outside named EE paths; check the path and edition. | Docker Compose, app/worker/Postgres/Redis, public URL for external webhooks and backed-up encryption key. | Conditional fit for technical RevOps with a flow owner and failure test. |
| Marketing Agent OS Harness library | Reusable marketing skills in a supported agent harness. | Apache-2.0 declared wrapper/normalized layer; inspect retained upstream licenses. | An approved harness, limited project-scope skills and no secrets for the first test. | Conditional fit for developers and consultants; strongest for planning, not execution autonomy. |
| Loopmark Agent Exploratory | Local content, audience and funnel drafting in a Python CLI. | MIT. | Python 3.10+, a low-spend test OpenAI key and disposable local JSON. | Exploratory solo-founder pilot only; leave social credentials unset. |
Where the boundaries matter
Mautic is a system, not a shortcut
Documentation covers forms, tracking, campaigns, queues, webhooks and scheduled operations. It does not make sending, consent, deliverability or every desired connector safe by default.
Activepieces is not a CRM
Its CE deployment has operational pieces and stores connections. Target-system connector scope, permissions, retries and data paths remain workflow-specific.
A skill library inherits its harness
Marketing Agent OS is instruction-first with optional helpers. Any account access, network action or data handling comes from the host harness and separately configured tools.
Loopmark has explicit operating gaps
Its issue tracker names approval/RBAC, managed scheduler retries/alerts and ad/follower analytics as open work. Treat it as an exemplar, not a production team service.
Maintenance signals are not guarantees
Mautic and Activepieces CE show explicit documentation, release and security-process signals in the reviewed sources. Marketing Agent OS has a short visible history and a provenance caveat. Loopmark’s visible project evidence is thinner and no dedicated security disclosure process was found in the checked sources. All four still need version, backup/restore, upgrade, secret, monitoring and incident checks before production.
Uncertainty & exclusions
- Release and commit details are dated snapshots, not quality scores.
- Exact integrations were not tested; absence from the checked evidence is not proof of absence.
- Model, mail, tracing and social providers bring separate pricing, terms and data-handling decisions.
- n8n, other alternatives, source-available counterexamples, hosted/proprietary tools and paid-ad execution are outside this first four-item scope.
Read the primary evidence
See sources and method; the private research record preserves the detailed, dated evidence notes.