unifounder
Skip the intro
unifounder
BlueprintsBlog
Log inJoin the waitlist
MarketsProblemsOffersFeaturesProofContentChannelsContactsMARKETINGPRODUCTREVENUETEAMWORKFLOW 2WORKFLOW 3WORKFLOW 1S1 · ICP & Problem Discovery· Needs youWorkflow · Validation EngineS1S1 · ICP & PROBLEM DISCOVERYIndependent Italian consultants who sellknowledge-based services are a promisingICP for an AI content creation system.1RESEARCHSegment & anti-ICP definitionRunning2ACTION AGENTProspect sourcing & reachabilityUnlocked · starting soon3ACTION AGENTProblem discovery interviewsLocked · needs Sourcing4RESEARCHDemand & timing desk researchLocked · needs SourcingOpen gate — close & decide earlyValidateEvidence holds → S2 opensPivotNew bet · stage re-runsInvalidateBack to an earlier stageKillLine closes · learning keptCOMPANY MEMORYDecision & reasoning written backS1ValidatedS1 archived · its learning stays in memoryS2S2 · SOLUTION WEDGEWorkflow advanced — the next stage opensunits queuedS3S4S5
The premise
You are the center.

Your agent orbits your business. It does the work — it doesn’t make the calls.

Company Memory
Everything your company knows.

Who buys, what hurts, what you sell, where they came from. Recorded as data, not as notes.

Not a list
It's a graph.

A problem is felt by a market and solved by an offer. A proof backs a claim and came from a customer. Your agent can follow these links.

The real state
Your company, company area by company area.

It's not how many records you have. It's how many hold up, connected to something.

workflows
Work runs stage by stage.

Several lines can run at once. Each advances one stage at a time.

The agent at work
Agent by agent.

The agent works through the stage one agent at a time — and where two don’t depend on each other, it runs both in parallel.

The gate & the return
Your decision returns to the map.

The stage completes and the gate opens: validate, pivot, invalidate, or kill — each choice decides how the workflow advances. Your call and reasoning are written back into the company memory.

Hover a record or a company area to see its connections
S1 · ICP & Problem Discovery
Workflow · Validation Engine
Runs inside the AI agent you already use

An operating method for
your AI agent.

Structured memory, an order of work, and mandatory stops where you decide. It proposes. You approve. Nothing moves without you.

Join the waitlist
✓ Works with any MCP client✓ Every decision stays yours✓ It tells you what it doesn’t know
The vocabulary

Eight words you just saw on the map.

The deck deals itself as you arrive. Click a card to flip it back.

Company Memory
Company Memory
What the system knows about you

Structured memory, not notes: markets, problems, offers, proofs — records joined by relationships, read before anything is done. It's why session two doesn't start from nothing.

Company Memory
Record
Record
One fact, with a status

A single structured fact in memory — a market, a problem, an offer, a proof. Each one carries a status: proven, assumed, or unknown. No label upgrades itself.

Record
Company Area
Company Area
Your company, area by area

Memory is organised by area of the company — audience, problems, offer, channels. What counts isn't how many records you have, but how many hold up in each area, connected to something.

Company Area
Workflow
Workflow
One bet, run to a conclusion

A reusable route to an outcome, run as an ordered set of stages taken from a blueprint. Several can run at once — they don't merge, and each keeps its own history.

Workflow
Stage
Stage
One question, with a hypothesis

A phase of the workflow. It states the question it exists to answer and the hypothesis it tests, in writing, before work starts. It ends at a gate.

Stage
Agent
Agent
A goal-directed unit of work

One piece of work that plans, acts and reports. Research Agents discover and update memory; Action Agents change something real, behind your approval; Recurring Agents run on a cadence.

Agent
Decision Gate
Decision Gate
Where you decide, not the AI

Every stage ends at one. The system reports what it found, recommends an outcome, and stops. Your choice is recorded with the date and your reasoning.

Decision Gate
Skill
Skill
Know-how, installed verbatim

A plain-text procedure your agent learns once — including where it must stop and ask you, and what it must never do.

Skill
How it works

It reads your business before it plans anything.

No setup wizard and no blank chat. You connect it to the agent you already use, and it starts from what's recorded.

01
Connect it

One endpoint over MCP in the client you already work in — Claude Code, Claude Desktop, or any compatible one. No second window, no second model subscription: you work where you're already working.

02
Your agent installs the skills

Procedures for research, positioning, offers, copy, experiments. They're plain text files — read one before you install it, and you'll know exactly where it stops and asks.

03
Start a workflow

Work is ordered into stages and units. Your agent always knows what's workable right now, what's blocked, and what's waiting on you — across sessions, weeks apart.

04
You decide, always

Nothing ships, sends, or signs without you. Each stage ends with a decision that's yours, recorded and dated. The agent proposes; the founder disposes.

Works insideClaudeClaude CodeChatGPTCodexGeminiPerplexityCursorOpenClawHermes+ any MCP client
Skills & Recurring Agents

Skills teach your agent how to work. Recurring Agents put it to work.

The method runs on two kinds of documents: skills your agent learns once and uses when needed, and recurring agents you run on demand — each one ending with a report in your hands.

segment-selection.skill.mdSKILL · KNOW-HOW
---
name:segment-selection
description:"Use this when choosing or scoring a target segment."
---
01Purpose
02Inputs it reads
03Evidence standard
⋮
08Where it must stop and ask youthe signature
12What it must never dothe signature
install_skill → written into your agent's files, verbatim

Lives in your agent, not in UniFounder. Activates on its own — the description is the only trigger. No report, no state, nothing to complete.

Personal brand — weekly runRECURRING AGENT · WORK
get_recurring_agent_instance· rules loaded · nothing starts yet
RECOMMENDS 5 SKILLS TO INSTALL FIRST
voicepositioningcontent-reviewclaimsdistribution
✓ REPORT · DELIVEREDrun #4

The report is the only record — there is no start call, and abandoning a run leaves nothing.

standalone → your run history  ·  stage step → advances the stage

Lives in UniFounder, never installs. Chosen deliberately — you ask for it, or it's a stage step. Repeatable as often as it's useful.

A recurring agent without its skills is a to-do list done badly. A skill without a recurring agent is competence nobody has used yet.

Questions, answered.

An AI agent that speaks MCP — Claude Code, Claude Desktop, or any compatible client. You connect one endpoint and keep working where you already work — UniFounder isn't another place to type, it's the layer your agent reads and updates. There's a web app too, but it's for you to read: the map, the reports, the decisions you made and when.

Your agent is already smart. Give it a memory and a method.

Joining the waitlist starts your founder profile. We onboard every founder personally — the earlier you join, the sooner you build.

Join the waitlist
unifounderRuns inside your agent · you decide at every gateBack to top ↑