GrowthMachina conducts SignalMachina (the scout), ContentMachina (the creator) and DistroMachina (the distributor) around one metric. Strategy is written as executable Markdown, every action sits behind an autonomy gate you set per channel, every paid call sits under a cap — and what a post actually did feeds the next idea.
growthmachina.app · the panel runs on your own machine · public data only
01 The shift
Each agent owns one thing and hands over one object: the scout produces an Idea you rate, the creator turns an Act idea into a ContentPackage you approve, the distributor turns an approved piece into a Post with an Outcome that flows back to the scout. Each has a "never" list — never post, never spend, never contact, never invent a fact. You set the north star and the autonomy level; the loop does the rest, at the pace you allow.
02 What it does
Every claim here is a page, a job or a table in the product today.
Every brand names its north star (for a WhatsApp-native app: weekly active users) and a target such as 100 % month over month. The Growth page turns that into a required weekly factor, compares it with the trailing four weeks — behind, on pace, ahead — and projects 30 days out. The pace line is injected into ideation, so ideas lean toward acquisition when you are behind.
A strategy is a versioned file: YAML frontmatter (lever, priority, status, channels, what it measures, what unblocks it) plus four fixed sections — why, how the system runs it, steps, measurement. The panel checks unblockers live (a missing secret, a TBD parameter, a source switched off), tracks one experiment per active strategy and shows its KPI tiles.
Draft: it proposes, you approve everything (the default for every brand). Copilot: it drafts and plans, you edit and press publish. Autopilot: it publishes inside your guardrails through official APIs, within ±10 minutes of plan. Set per brand and per channel; levels above draft require a connected account.
Outcomes at 24 h, 72 h and 7 days are compared with the idea's ICE estimate and the owner's rating. Winners (top-quartile outcomes) go back into ideation as "what worked", losers as "avoid", observed best hours into the planner — a calibration table every week, not a dashboard you babysit.
Each paid source and model has its own monthly cap; a spend guard checks the worst case before the call and records the actual cost after. The Command Center shows API spend today, this month and projected; the studio prints "render approved (≈ US$X)" before any video money moves. Local models do the default work.
Trends tagged with deaths, disasters, violence or hard politics are blocked before any brand sees them (100 % blocked on a labeled set of real headlines). Unverified facts stay marked [verify], never invented. A brand-guardian critic scores every piece; clones need written consent and carry AI labels; publishing uses official APIs only.
03 The agents
One Postgres, one model gateway, one config tree per brand (brief, voice, watchlists, channels, strategies). Each agent reads the same brand memory and writes to the same loop.
The scout
Watches what is rising in your markets, what competitors ship and what their users complain about, and turns every opening into a growth hack that names its loop — hook, format, timing, ICE score. Sources: Google Trends and News, X, YouTube, TikTok, charts, app stores, press.
Live · suggests, never postsThe creator
Turns an approved idea into pieces per channel in a voice learned from the creator's own posts, with tested hook variants and a critic. The Creator Studio takes a script to a finished reel: looks, storyboard, keyframes with the consenting creator's likeness, image-to-video, cloned voice, lipsync, captions, music bed, provenance.
Live · creates, never postsThe distributor
Plans every approved piece into a slot — boost windows, best hours, spacing, blackouts, weekly cadence — and publishes through official connectors for Instagram, TikTok, YouTube, Telegram and WhatsApp behind the autonomy gate. Then it measures and learns.
Live · 3 connectors in platform reviewThe conductor
One local command center: north star and pacing, Strategy Playbooks, API spend, the agent swarm's live state, and one section per agent. Runs on your own machine, reachable only from your network or tailnet.
You are here04 From signal to post
The same loop in the strategy files: sense, decide, create, distribute, learn — with your two short reviews where they matter.
Ideas land. The probe writes the day's ideas per brand: trend, bridge, idea, growth loop, ICE score, lifecycle stage. Competitor moves (a new version, a listener jump, a review theme) enter the same check. When the pace is known, one line steers the prompt: behind vs 100 % MoM — prioritise acquisition.
Rate. In Ideas, each card gets Act / Maybe / No and a short why. The ratings calibrate ICE and are the only way an idea moves on.
Draft. Every Act idea becomes a package: one piece per enabled channel (TikTok video, Instagram reel or carousel, WhatsApp forward, ad creative, X post), three hook variants each, the loop made explicit in the CTA. A separate critic scores voice fit, platform fit, specificity and loop clarity; below 6 it is redrafted once.
Review. Pieces side by side in the Review Queue: pick a hook, edit, approve. Every edit and rejection is summarised into the brand's learned voice memory and injected into the next draft.
Produce. Covers and narration render automatically under their caps. For video, the Creator Studio proposes a 4–6 scene storyboard with thumbnails; you approve scene by scene, then it renders, assembles with captions and a music bed, masters to −14 LUFS and exports with a provenance sidecar. A 27-second reel cost US$0.42 on its first real run.
Plan and publish. A slot inside the boost window (rising and peak trends go first), honouring best hours per geo, spacing, blackouts and the channel's weekly cadence. Draft: a copy-ready checklist and you post. Copilot: "Publish now". Autopilot: published within ±10 minutes of plan, with the platform's AI label when the video uses a clone.
Measure. Outcomes per post — views, likes, comments, shares, saves, follows, clicks — pulled from the platform API for connected channels, entered by hand otherwise.
Close the loop. Learn compares ICE with realised results normalised by the channel's 90-day median, lists winners and losers with their loops and the observed best hours, and appends guidance to the next ideation prompt. The operational gate comes first: ≥ 80 % published within ±1 day of plan and ≥ 70 % with a 7-day outcome, or no strategy result is trusted. Then the 20-minute experiment review in Strategy Playbooks.
05 Control
Three levels are shipped; the fourth is the stated direction. Click a level to see who does what.
The default for every brand. Signal suggests, Content drafts into a review queue, Distro plans a slot and prepares a copy-ready checklist. Nothing leaves the panel until you act.
06 Sources & connectors
What the scout listens to, what it watches for your brand and competitors, where the distributor can publish, and what renders the work. Logged-out public data, polite rates, no scraping behind logins, no cookies, no browser automation.
Publishing: TikTok public posting and YouTube public uploads are pending platform review, Instagram needs the account added as a tester until app review — details on the DistroMachina page. X and ad accounts are never published to.
We are onboarding a small group of creators and consumer apps in Latin America and Spain. Bring your brand brief, your channels and your metric; set the dial where you are comfortable.
Built by the team behind SignalMachina, ContentMachina & DistroMachina.