Growth OS
Forty-eight agents find a first audience and pick the next experiment, without publishing for you. Owns the first audience and the next experiment.
$19.99once
This OS, lifetime. Studio stays on monthly plans.
- Agents
- 48
- Skills
- 11
- Teams
- 10
- Objects
- 8
- Programs
- 1
- Workflows
- 3
- Tools
- 2
- Evaluators
- 3
Why this OS
What it is for
A founder who needs a first audience and one next experiment, not a GTM essay or a launch.
A model asked to do marketing writes a plan with no audience object, no frozen brief, no independent critic, and no experiment you can refuse.
The week has a bottleneck and one next experiment. Demand is not invented. Nothing ships until you accept it.
A founder asks a model to do their marketing. They get a GTM essay. There is no audience object, no frozen brief, no independent critic, and no next experiment they can refuse. The essay sounds like a department. It is a paragraph with confidence.
Growth work fails in the gap between a clever paragraph and an inspectable run: who we reach first, what we sound like, what we try next. Each of those should be a file, with evidence and gaps left explicit. A prompt pack of channel personas cannot remember the foundation when it writes the campaign. It will happily invent demand, a conversion rate, or a customer who was never spoken to.
Growth OS is that department as an organization. The compiled package defines forty-eight specialist profiles across ten disciplines. Your host runs the Director. A mission activates a subset. Defined is not mobilized. Mobilized is not concurrent. The Director chooses one bottleneck, allocates the week, delegates, and stops before outreach. The Director does not launch.
The life this changes is the week that used to disappear into channel spray. You bring product facts you already have. Unknown customer and market facts stay unknown. Intake freezes the brief. Intelligence selects evidence worth acting on. Audience and strategy name who first, and the promise. Editorial, studio, and distribution work that frozen context. Learning names one next experiment. Trust critiques the dossier independently. You accept the experiment, or you do not. Nothing ships.
Three workflows are compiled. Marketing foundation prepares market, ICP, buying dynamics, and positioning. Brand system prepares identity, voice, tone, messages, and examples. Campaign strategy prepares objective, offer, channels, experiment, measurement, and costs. Six further missions stay documented (competitive review, growth diagnosis, daily build, community review, launch, avatar content) until they are compiled as live capabilities. Documented is not compiled.
What people want from growth software is usually a feeling of motion: more posts, more channels, a dashboard that goes up. That is vanity if the first audience was never named and the next experiment was never refused. This OS owns the first audience and the next experiment. It will not publish, spend, or treat a reviewed draft as proof of demand.
Trust reviews a frozen dossier. The reviewer does not grant publication rights and does not validate the market. Authorized publish is a declared binding slot. It does not auto-post. Paid spend is not in this package. If a catalog listed LinkedIn or Instagram, that is a specialist who prepares utility, not a bot that posts.
A day with the OS is a frozen brief instead of a new GTM essay. A week is one experiment with a critic, not seven channels that cannot remember the foundation. A month is a short log of experiments you accepted or deferred, not a follower count you cannot explain. A year is a memory of what was already tried. It is not a claim that revenue multiplied.
The objects are product brief, audience, offer, campaign, experiment, review. Objects you can reopen, not a chat that forgets. Forty-eight agents are defined on the organization graph. A mission activates a subset. We do not claim 246/246 PASS. The compiler refuses RELEASE. Host LLM jobs are NOT_RUN until your host executes them.
The upstream Python and Railway service, Discord bot, and Station showcase are a separate runtime. This page sells the Agentik host-first package. If you wanted a launch button, this is the wrong OS. If you wanted someone to invent a customer so the week feels full, this is the wrong OS.
If you have a product and you do not yet know who first, or you keep restarting the GTM from a blank chat, this is the machine. Bring facts and sources. Leave the fiction out. The next experiment is the deliverable, not the launch.
MCP is a control layer
This OS is an organization you install on a host. It can sit on top of other connections that host already has: an HTTP MCP server, a CLI, or a signed API. App connections use a platform key plus a user id (Composio model B). They are not a second raw MCP URL. The OS owns the job, the objects, and the gates. The connection is I/O the owner authorized. Agentik-OS on this site is the live control connector. Ads, CRM, CMS, or repo apps are not pre-wired here. You attach them on the host or through Connections when a run needs them.
A growth run can drive ads, CRM, analytics, social, and mail once those exist on the host. This package still does not publish or spend.
Tools you can connect
Curated for this OS. Not a Composio catalog dump. Nothing in this list is a live vendor on mcp.agentik-os.com except Agentik-OS itself.
Ads
Connectable / planned
Google Ads or Meta Ads as a host connection. Paid spend is not in this package.
CRM
Connectable / planned
HubSpot or Salesforce if the host already has them. Not installed on this site.
Analytics
Connectable / planned
GA4 or Mixpanel for facts you already have. No invented conversion rate.
Social
Connectable / planned
LinkedIn, Instagram, or X as prepared utility. Operators in the roster do not auto-post.
Mail
Connectable / planned
Gmail or Resend if the owner authorizes send. Outreach stays off until consent.
Orchestration
- Director
- Growth OS Director: choose one bottleneck, allocate the week, delegate, and stop before outreach. The Director does not launch.
- NanoTeams
- Ten NanoTeams, mapped from the upstream micro-OS: Direction (founder liaison, chief of staff, portfolio economics), Intelligence, Strategy, Editorial, Studio, Distribution, Community, Revenue, Learning, Trust.
- Workflows
- Three compiled workflows: marketing foundation, brand system, campaign strategy. Six further missions stay documented (competitive review, growth diagnosis, daily build, community review, launch, avatar content) until they are compiled as live capabilities.
- Gates
- Trust reviews a frozen dossier independently. The compiled owner-gate waits for you. Authorized publish is a declared binding slot. It does not auto-post. Paid spend is not in this package.
- Objects
- Product brief, audience, offer, campaign, experiment, review, plus mission and evidence records. Objects you can reopen, not a chat that forgets.
Missions
Nine missions in the upstream Growth OS. Three are compiled capabilities in this package. The rest stay documented until they are.
- marketing_foundationMarket, ICP, buying dynamics, positioningcompiled
- brand_systemIdentity, voice, tone, messages, examplescompiled
- campaign_strategyObjective, offer, channels, experiment, measurement, costscompiled
- competitive_reviewObservations, objections, alternative explanations, a proposed actiondocumented
- growth_diagnosisA diagnosis and an experiment to investigate itdocumented
- daily_buildA story and distribution plan for today’s progressdocumented
- community_reviewCommunity findings and a response plandocumented
- launchPositioning, editorial work, and a launch plandocumented
- avatar_contentA persona, script, and storyboard for connected production toolsdocumented
Methods
- Frame the week, delegate, demand a receipt
- Collect the smallest useful founder input
- Select evidence worth acting on
- Choose segment, promise, and channel focus
- Run an evidence-based content calendar
- Prepare production briefs; never invent a published asset
- Plan channel workload; never auto-publish
- Prepare community plans without inventing members
- Activate monetization only when requested
- Turn observed outcomes into the next experiment
- Independently check claims; keep effects inside authority
You get
- A framed product brief with forbidden moves
- Audience and positioning that can be inspected
- An independent critique before synthesis
- One next experiment, not a launch
It will not
- It will not auto-post because a catalog listed a channel
- It will not invent demand or a conversion rate
- It will not claim 248/248 PASS or RELEASE
- It will not run 48 agents at once because 48 exist
Honesty
- This is a preview OS. The package compiles. The compiler refuses RELEASE. Your host runs the model when it calls MCP.
- Forty-eight agents are defined on the organization graph. A mission activates a subset. Defined is not mobilized. Mobilized is not concurrent.
- The upstream Python/Railway service and Station showcase are a separate runtime. This page sells the Agentik Host-First package.
- A reviewed draft is not market proof. We do not claim 248/248 PASS.
How a run looks
The user path for this OS. As many steps as the run needs.
- You bring product facts you already have
- Director opens one mission, not all 48
- Intake freezes the brief
- Intelligence selects evidence worth acting on
- Audience and strategy name who first, and the promise
- Editorial shapes the message from that frozen brief
- Distribution plans channel workload without publishing
- Learning names one next experiment
- Trust critiques the dossier independently
- Synthesis returns the reviewed draft
- You accept the experiment, or you do not. Nothing ships.
Workflows
Nodes and edges from the compiled package. Nothing decorative.
Intake freezes the brief. Independent contexts are required for the analyses. Trust reviews a draft. It does not validate the market and does not authorize publication. You accept the experiment, or you do not.
Result synthesis · Completes at owner-gate
Agent
Intake
Agent
Intelligence
Agent · isolated
Strategy
Agent · isolated
Editorial
Agent · isolated
Distribution
Agent · isolated
Revenue
Agent · isolated
Learning
Agent · isolated
Trust
Agent
Synthesis
Program
Record
Gate · approval
Owner gate
Agents
48 agents. A run mobilizes a subset.
| Agent | Does |
|---|---|
| Growth OS Director | Choose one bottleneck. Do not launch. |
| Founder Liaison | Collect the smallest useful founder input. |
| Chief of Staff | Turn direction into accountable work. |
| Portfolio Economist | Keep effort and spend tied to unit economics. |
| Intelligence Lead | Select evidence worth acting on. |
| Journal Editor | Extract inspectable progress from the founder journal. |
| GitHub Scout | Turn selected repositories into verified demo notes. |
| Audience Researcher | Discover audience jobs and objections. |
| Market Radar | Find differentiated opportunities and external changes. |
| Strategy Lead | Choose segment, promise, and channel focus. |
| Positioning Architect | Make product value understandable. |
| Journey Architect | Map first contact to repeated value. |
| Channel Planner | Select realistic channel workload. |
| Partnership Planner | Find complementary ecosystem partnerships. |
| Editorial Lead | Write messages from the frozen brief. Do not schedule posts. |
| Hook Writer | Develop honest attention openings. |
| Script Writer | Write useful scripts that fulfil their hooks. |
| Repurposing Editor | Adapt one insight to channel-native drafts. |
| Localization Editor | Adapt meaning to language and context. |
| Studio Lead | Prepare production briefs. Never invent a published asset. |
| Persona Designer | Create distinct disclosed synthetic presenters. |
| Storyboard Artist | Make each visual carry useful information. |
| Asset Custodian | Track rights, identity, and production provenance. |
| Finishing Reviewer | Review the production brief independently. |
| Distribution Lead | Plan channel workload. Never auto-publish. |
| LinkedIn Operator | Prepare LinkedIn utility. Never auto-post. |
| Instagram Operator | Prepare Instagram productions. Never auto-post. |
| YouTube Operator | Prepare searchable YouTube packages. Never auto-post. |
| X Operator | Prepare concise X drafts. Never auto-post. |
| Community Lead | Prepare community plans without inventing members. |
| Onboarding Concierge | Design a useful first action for a new member. |
| Conversation Host | Prepare discussion prompts. Do not send Discord messages. |
| Member Success | Draft re-engagement around expressed goals. |
| Moderation Triage | Write proportionate community rules. Do not apply them live. |
| Event Producer | Prepare community ritual drafts. Do not run them. |
| Revenue Lead | Activate monetization only when requested. |
| Offer Architect | Make requested offers clear and credible. |
| Conversion Editor | Write requested conversion paths. Do not ship them. |
| Lifecycle Operator | Plan opted-in follow-ups. Do not send them. |
| Learning Lead | Name one next experiment from observed outcomes. |
| Measurement Engineer | Make events and denominators trustworthy. |
| Experiment Designer | Design bounded falsifiable growth tests. |
| Retention Analyst | Find what predicts repeated useful participation. |
| Trust Lead | Independently check claims. Do not grant publication rights. |
| Fact Reviewer | Independently check consequential claims. |
| Privacy Steward | Keep secrets and private material out of public artifacts. |
| Reliability Operator | Recover jobs without duplicate external effects. |
| Budget Controller | Enforce spend and invocation envelopes. |
Before / after
A day, a week, a month, a year. Honest deltas. No invented multiple.
| Horizon | Without | With | Delta |
|---|---|---|---|
| Day | A GTM essay. No audience object. No frozen brief. | Intake freezes the brief. Unknown facts stay unknown. | One inspectable brief, not a new essay each morning. |
| Week | Channel spray. Posts without a first audience. | One bottleneck, a subset of the forty-eight, one next experiment. | Experiments started, not channels opened. Nothing ships. |
| Month | A follower count you cannot explain. Deferred work forgotten. | A short log of experiments accepted or refused, plus gaps Trust named. | Experiment count and refusals. Not vanity reach. |
| Year | We posted a lot. Demand was never evidenced. | A memory of first audiences tried and experiments deferred. | What was already tried stays on file. No invented revenue multiple. |
How to use it
Unlock ($19.99 once, lifetime, or a plan ($19 / $99 / $199) that includes official OS.) then install on the host you already pay. Talk to the Director. Call a team. Approve named artifacts. The host pays tokens.
How to talk
Copy a starter and paste it after install. list_agents shows real ids. ask_director starts a run. ask_team calls one named specialist. Replace the brackets with your facts.
list_agents({ osId: "growth-os" })ask_director({ osId: "growth-os", prompt: "Open one mission from product facts I already have: [facts]. Do not mobilize all 48. Stop before outreach." })ask_team({ osId: "growth-os", teamId: "audience_researcher", prompt: "From these product facts, who first? Do not invent a customer." })approve({ jobId: "<job-id>", approved: true })- 01
Unlock before you install
Buy Growth OS for $19.99 once, lifetime, or use a plan that includes official OS ($19 for three official OS, $99 for ten, $199 unlimited). You must buy it to use it. Payment is confirmed before access opens.
- 02
Install on the host you already pay
MCP is the HTTP connector. CLI is the repo commands. Installing does not start posting, spending, or scheduling outreach. Automations stay off until consent.
- 03
Talk to the Director
Bring product facts you already have. Ask the Director to open one mission, not all forty-eight roles. The Director chooses a bottleneck and stops before outreach.
- 04
Pick a compiled workflow
Three are compiled in the package: marketing foundation, brand system, campaign strategy. The other six missions stay documented until they are compiled. Do not treat a documented mission as a run.
- 05
See who can be called
list_agents returns the forty-eight compiled rows. A run mobilizes a subset. ask_team calls a specialist (for example audience_researcher) with a job, not a free-form mention.
- 06
Accept the experiment, or refuse it
Trust critiques the dossier. You accept the next experiment, or you do not. approve is for an identified artifact. Nothing publishes from this package.
- 07
What the host pays
Tokens stay on Claude, ChatGPT, Codex, or Hermes. Agentik unlocks the organization. A reviewed draft is not market proof. We do not invent a conversion rate.
Access
Unlock to use
You must buy Growth OS to use it, or choose it as one of 3 on the $19 plan.
$19.99 once, lifetime
You unlock the organization, not tokens. Tokens stay on the host you already pay.
Payment is confirmed before access opens. A success page never grants access.
Install
MCP is the HTTP connector. CLI is the repo commands.
Choose a host
MCP · Cursor
Checking…
One-click
Adds Agentik-OS as an OAuth MCP connector. Cursor reads PKCE from the 401 and well-known metadata.
Add in CursorOr paste MCP JSON
~/.cursor/mcp.json or project .cursor/mcp.json
{
"mcpServers": {
"Agentik-OS": {
"url": "https://mcp.agentik-os.com/api/mcp"
}
}
}OAuth for Cursor. API key for CLI. GET is public.
Using
Who uses it
- Checking public handles…
Using counts appear only from a live Convex query of public @handles with a verified grant. The house operator is @agentik only. Never email.
Other OS
An editorial system for ideas, briefs, assets, and approvals, then an authorized package, with a human gate before anything leaves the building.
A knowledge intelligence layer: discover sources, record access, extract claims with provenance, keep contradictions, teach to a learner, and feed other OS. Never pretend a book was read.

