The content pipeline
Ideas become masters, masters become platform-native variants, variants wait for a human thumbs-up. Never blind cross-posting.
Strategy, creation, publishing, community, attribution. One operating system, seven platforms, AI you can actually trust.
Autonomous pipeline demo: seven stages complete in sequence — research, script, voice, visuals, assembly, a human-approved QA gate, and publish — with a settled cost of 38 cents.
7
Platforms, native adapters
5-tier
AI quality routing
100%
Audited AI actions
7platforms, natively6official APIs0scrapers
Every stage shares one data model, one design system, one honest AI layer.
Ideas become masters, masters become platform-native variants, variants wait for a human thumbs-up. Never blind cross-posting.
Every job, its state, and why — failures explain themselves instead of hiding.
ECONOMY for inbox triage, PREMIUM for flagship concepts. Cost shown before it runs.
Seven platform adapters over official APIs — with capability labels so you always know what is automated.
Content → click → lead → revenue, with honest scope about what platforms allow.
Roles from OWNER to CLIENT, RLS-enforced isolation, and an approval inbox every risky action must pass.
Normalized cross-platform metrics, preserved raw values, and AI that answers why — not just what.
Seven stages, each with a price. The one that matters is sixth: nothing paid runs until you have seen the bill and said yes.
Pulls what is already working in your niche and what is missing, into a brief the writer can use.
Writes the episode in your brand voice, then cuts it into the narration parts the voice step needs.
Narrates it, aligns the audio to the script, and produces timed captions from the alignment rather than guessing.
Plans the shot list from the script and renders it. The most expensive stage, which is why it sits behind the gate.
Cuts picture to narration, lays in captions and cards, and renders the master.
Checks the cut against the script and shows you the bill before anything publishes. Quote $0.38 — you approve.
Uploads through the official API with thumbnail, captions and description, and records what it did.
Autonomy is easy to sell and hard to trust. Four things make it checkable rather than promised: nothing risky ships without a person, everything that happened is written down, one workspace can never read another, and every capability says how automated it actually is.
Anything that spends money or speaks publicly waits for a person. Nothing ships because a model felt confident.
approvals table · approvalRequired on every audit entry
Who did it, or which agent did, what it cost and what it changed. Written as it happens, not reconstructed after.
audit_log · actorKind: user | agent | system
One workspace cannot read another, enforced in the database rather than remembered in the code.
Postgres RLS · is_workspace_member() on every table
Each platform capability says how automated it really is, including the ones that are not. A matrix with no "no" in it is marketing.
AutomationLevel · per capability, per platform
Every figure here was counted out of the product, and each card prints the thing that decides it underneath — so none of them has to be taken on trust.
144
tables behind row-level security
Workspace isolation is a rule the database enforces on every tenant-owned table, not a WHERE clause somebody has to remember. No migration has ever dropped one, so the count only goes up.
supabase/migrations · ENABLE ROW LEVEL SECURITY
58 × 14
triggers and actions, both closed sets
An automation starts from an event this product really emits and ends in an action it can really run. Anything outside either list is refused on the way in rather than accepted and failed later.
TRIGGERS · ACTION_CATALOG · automation.module.ts
$0.38
one finished episode, itemised
Seven stages, each with its own price and its own quality tier, adding up to the figure the demo settles on. The bill is shown at the QA gate before anything publishes.
Research 0.06 → Publish 0.01 · the ledger above
5 of 5
automation levels published, UNSUPPORTED included
Every capability on every platform wears one of five labels, and the unflattering one is on the list. A matrix with no "no" in it is a marketing claim.
AutomationLevel · packages/contracts
8
roles, OWNER through VIEWER
Client and approver are roles with their own reach, not a shared password and a promise. The API checks them and row-level security checks them again.
Role · packages/contracts · is_workspace_member()
1,559
API tests, green
The five cards before this one are not documentation. They are assertions, and this is how many of them stop passing the moment one of them stops being true.
76 spec files · pnpm --filter @mad/api test
Value dimensions, not arbitrary credits. AI usage is billed transparently against real provider cost.
about $3.04 of AI per month — quoted per run, settled per stage. No black-box credits.
One brand, consistent everywhere.
$ 29 / mo
For teams shipping campaigns weekly.
$ 89 / mo
Multi-brand governance and client delivery.
$ 249 / mo
YouTube, Instagram, Facebook, TikTok, LinkedIn, X and Pinterest — all over official APIs, each with honest capability labels for publishing, analytics, inbox and ads.
Every AI task declares the quality it needs. Cheap deterministic providers handle classification; premium models handle flagship creative. Cost is shown before anything runs.
Not by default. Publishing, ad spend and sensitive replies always route through the Approval Inbox, and every autonomous action is audited with cost and model.
Yes — workspace isolation is enforced twice: application-level role checks and database row-level security on every tenant-owned table.
We tell you. Features are labeled AUTOMATED, GUIDED, MANUAL or UNSUPPORTED — we never fake automation that the platform does not permit.
Plans price the value dimensions (brands, accounts, seats); AI volume draws from transparent credits settled against real provider cost.
Your whole marketing lifecycle, one calm workspace.
No credit card · Official APIs · Your data stays yours