Research decision · 23 Sep 2026
Loopmark Agent: retain the slot with a stricter label.
Decision: retain Loopmark only as an exploratory local drafting CLI. It is not a team-approval system, managed scheduler, CRM, or production/unattended publishing service.
Decision card
Technical solo operator; local and draft-only
Use only for a fictional-data drafting experiment with social credentials absent. The checked material supports a local CLI and local JSON records, not managed marketing operations.
- Hands-on status
- No hands-on install or test occurred.
- License
- MIT for the named repository.
- Replacement decision
- MarketingOS AI was examined but is not recommended.
Evidence gaps that control the decision
| Gap checked | What the checked material shows | Decision consequence |
|---|---|---|
| Release/package provenance | The README offers a package install line and source fallback; August packaging commits are visible in history. The checked repository surface listed no Releases or Packages, and a PyPI artifact was not independently retrieved. | Do not claim a published package or versioned release provenance. |
| Security reporting | The README gives secret-handling advice. No dedicated SECURITY.md, private reporting route, advisory process, or supported-version policy was verified in checked material. | Secret advice is not a vulnerability-disclosure process. |
| State and recovery | The README describes local JSON under data/ or DATA_DIR. No backup, restore, retention, migration, or locking procedure was found. | Local persistence is not documented operational recovery. |
| Approval/RBAC | Orgs, RBAC, and post-approval workflows are listed as planned hosted/paid work in the open issues. | Not a team-approval system. |
| Retries, alerts, run visibility | Scheduler source shows a local process, status output, and exception printing. Managed retries/alerts remain planned. | Not a managed scheduler or production operations service. |
“Not verified” and “not documented” refer only to the checked primary sources; they do not prove a capability cannot exist elsewhere.
Compact comparison: examined but not recommended
| Area | Loopmark | MarketingOS AI |
|---|---|---|
| Project boundary | MIT local Python/LangGraph CLI for drafting, with optional social paths. | MIT private 0.1.0 Next.js marketing workspace with HubSpot, Supabase, Clerk, and OpenAI dependencies. |
| Documented improvement | No verified approval/RBAC path. | README claims review states, reviewer decision, immutable approved content, run history and traces; publishing is intentionally disabled. |
| Limitation that prevents replacement | See evidence gaps above. | README says live persistence and production HubSpot sync are incomplete; demo run storage is in-memory. No verified backup/restore, retry/alert policy, private security-reporting route, or release/package was found. Four visible commits are dated July 2026. |
| Sources | Repository · License | Repository · License · Package manifest · History |
MarketingOS AI is not a recommendation and is not catalogued as a product. Its claimed approval boundary made it a useful comparison, but it does not clear the replacement threshold.
Draft-only first test
Do not import customer or CRM data, schedule, send, post, resolve a real complaint, mutate a CRM, use production credentials, or run any social integration. Record unresolved setup, error, data-path, and cleanup questions before any later evaluation.
Replacement threshold
Replace this slot only with a narrow OSS marketing-agent application that has primary evidence for an exact OSI-style license and scope, an installable tagged release or verifiable package tied to source, documented credential/data path and no-action mode, approval or enforced draft-only external-action boundary, durable state with backup/restore guidance, retry/error/run visibility or an explicit limit, and a usable security disclosure route.
Uncertainty and exclusions
- No hands-on test occurred. Installation behavior, output quality, provider handling, reliability, and error behavior were not independently measured.
- External providers are separate decisions: Loopmark can send data to OpenAI and optional providers; MarketingOS AI’s live path can involve Supabase, Clerk, OpenAI, and HubSpot.
- Hosted/proprietary products, documentation-only connectors, broad runtimes/frameworks, and broad marketing teams were excluded from this narrow-agent slot. MarketingOS AI alone was compared because its documented approval-first boundary was relevant.
- No autonomous action is recommended.