governed autonomy for marketing
The whole funnel, on autopilot
For B2B SaaS founders without a marketing team: marketinque plans and drafts every stage of the funnel, and nothing publishes, sends, or spends without your say-so, item by item or by a rule you set.
One email at launch. Unconsented blasts are BLOCK-classed around here.
- AUTO
content.draft_articleRan on autopilot. Logged.
- APPROVAL
linkedin.publish_postHeld in your approval queue.
- BLOCKED
cold_email.unconsented_blastNo execution path exists.
Three risk classes from the live policy matrix. One gate before anything reaches the outside world.
Every stage of the funnel. One agent.
8 stages, all 48 channels, one approval queue. The shape is the strategy: demand generation narrows attention into customers, lifecycle marketing widens customers into advocates.
Then it loops. Case studies feed attraction, reviews feed activation, referrals skip the top entirely. Run end to end, a funnel behaves like a flywheel.
Measure wraps all of it
Measurement is not a ninth stage; it wraps all eight. The strategist reads your goal and live numbers each cycle and works the next unlocked steps of one plan across the funnel, instead of the channel that shouts loudest, and what each cycle measured is what the next one learns from.
One agent, on a loop
Plan, draft, approve, audit — that is the view from your seat. Underneath runs a longer pipeline: a self-written brand brief, live web research, an adversarial critic, and a memory that learns what works for you.
Plan
Reads your goal and live numbers each cycle, works the next unlocked steps of your launch plan, and drafts for the channels that need it. Fresh work is skipped; its own load is capped.
Draft
Articles, posts, ad creative, outreach, briefs. Drafting is risk-classed AUTO and runs immediately, because a draft cannot hurt you.
Approve
Publishes, sends, and spend stop at the gateway. By default each waits for your yes; per action, you can let it go after a veto window, or on its own once it has earned it. Spend always asks.
Audit
Every action is appended to a hash-chained, append-only log. The chain is re-verified in the database, so edits cannot hide.
What you actually look at
The console running marketinque’s own marketing, captured on 25 September 2026. Real ideas, real plan, real log.
Designed around one question: what could go wrong?
Autonomy is a governance problem, not a model problem. So the guardrails are structural, not promises in a system prompt.
Every action passes one gate
Before anything runs, the gateway looks up its risk class. AUTO runs and is logged. REQUIRE_APPROVAL waits for you. BLOCK never runs: fake reviews, unconsented blasts, mass DMs, and paid link schemes have no execution path at all. An action with no policy row fails closed.
A log that cannot be quietly edited
Every action lands in an append-only log, each row hash-chained to the one before it. The database re-verifies the chain live; tampering breaks it visibly.
Kill switches
Stop one channel or the whole agent instantly. Irreversible actions, like money leaving, can never autopilot regardless of policy.
It markets itself, in the open
marketinque is customer zero: the agent runs the early stages of its own funnel in production. The guide and the blog attract, the free tools and the waitlist capture, and this page is the positioning. Every article was drafted by the agent and approved by a human before it went live.
Questions operators ask
- What can it do without asking me?
- Draft. Articles, social posts, ad creative, outreach drafts, audits, and briefs are risk-classed AUTO and run on autopilot, because a draft cannot hurt you. Anything that leaves the system (a publish, a send, a dollar of spend) waits for your approval by default. Per action, you can give it a veto window, where it goes out unless you cancel, or let it run on its own once it has earned that with five clean approvals and marketinque sends it itself. Email to your own list earns it too, and a bounce or complaint spike sends it back to asking. Spend and cold outreach always ask.
- Does it really cover the whole funnel?
- Every stage from positioning to advocacy has channels the agent drafts for, and all 48 channels in the catalog map to a stage. The human moments stay human: sales calls, demos, and kickoffs are prepped by the agent with battlecards, objection pages, and scripts rather than performed by it.
- What stops it from publishing something embarrassing?
- By default nothing publishes on its own: every publish waits in your queue. If you let a kind of action go on a veto window, a daily email lists what goes out next with a one-tap Cancel, and cancelling sets that action back to asking. Kill switches, per channel and global, stop the agent instantly.
- Will it sound like AI?
- It is told not to. The drafting prompt carries a list of the words and moves that mark text as machine-written, and the review panel names any that slip through, before you approve rather than after you post. Drafts start from your brand dossier, and the edits you make to them become voice rules for the next ones. You still read every draft: nothing goes out that you did not approve or set a rule for.
- Will it post in communities as me?
- No. Answers for Reddit, Hacker News, Slack groups and forums are drafted for you to post yourself, from your own account, if you want to. The only places it sends to itself are yours: your Bluesky, your dev.to, your Discord, your own email list and the project’s own blog, and only after your approval or a rule you set.
- Can it spam people?
- No. Unconsented blasts, mass DMs, and fake reviews are BLOCK-classed in the policy matrix, which means no execution path exists for them. Legitimate outbound (cold email, ABM, social DMs) is drafted by the agent but consent-gated and human-approved before any send.
- How do I know the audit log is honest?
- Each action is appended to a write-once log where every row carries a hash of the previous one. The database re-verifies the whole chain on demand, so a deleted or edited row breaks the chain visibly on the governance dashboard.
- When can I use it, and what will it cost?
- marketinque currently runs one production deployment, its own funnel, while the governance model proves itself in the open. It is free for now, with no card asked for, and projects open to the waitlist in order. The platform is built lean on purpose: it skips fresh work, caps its drafting per cycle, and runs on a single small server.