Hypotize wordmark
Get started
The first agentic Growth OS

Your AI can generate 100 experiments. It cannot run one properly.

Hypotize gives your AI agent the stack of a disciplined growth hacker: a method it has to follow, with stages in order and outcomes measured, and a shared memory of every test and learning that your team and your agents can query.

Start for freeSee it in action
Works with
Kimi logoOpenClaw logoHermes logoand any MCP-compatible agent

What goes wrong without it

AI made testing faster. It also made bad testing faster.

An agent can draft ten experiments before lunch. What it cannot do on its own is hold the line between a measured result and a confident guess.

The agent declares victory

Confident answers before anyone measured.

Tests close without a number

“Looked good” cannot be reused.

Old tests come back as new ideas

Lost results get paid for twice.

The tracker becomes another job

Manual updates fall behind, then stop.

Live example

This is what running it properly looks like.

Four agents, one record. Each one checks what already exists before adding to it, and none of them can close a test without the measured result and the learning. What they write here is what your team reviews.

Reply

read

Writes to

Written to the record
IdeaPrioritizedRunningOutcome
Result
+6.0% completed checkouts
Evidence
ga4-checkout-march.png

Key learning

Removing the coupon field beat every discount idea scored above it. On this page, friction costs more than price.

The workspace

Agents write into it. Your team works in it.

Bring your team into one shared workspace. Members can create projects, score ideas, track progress, and review results together. Keep sensitive projects private and invite clients to individual boards as free contributors or viewers.

The backlog of candidate experiments

The pool of ideas before anything runs. Each one sits here to be scored and voted until the team approves it or drops it.

Before you connect

What your agent can touch. And what it never can.

The connection screen spells out the boundary before anything is shared. A connected client only reaches the projects you already have access to.

Mock of the OAuth authorization screen: Claude requests to connect to your Hypotize workspace. Granted: see the record, and keep the record up to date. Off unless you tick it: delete projects and experiments. No client can ask for billing, members or admin settings, projects you are not part of, or sending your record to another AI.
  • Data stays in the EU: Hetzner, Finland
  • OAuth: your password never leaves Hypotize
  • Disconnect in one click from Settings

From people running the tests

What the first teams using it say.

We test AI ideas daily, and the backlog was always the bottleneck. Now I brainstorm with Claude and the experiments land in Hypotize already scored and prioritized. The tool disappeared into the workflow, which is the best thing I can say about it.
Chris RubinFounder & CEO, brandmultiplier.ai
Hypotize gave my agency one uniform way to run ideation and experiments across clients. Our AI agents log ideas and results into the same record the team reviews, and the reporting is ready to present internally and to the client without extra work.
Emanuele Di BiaseHead of Growth
I manage experiments for multiple clients, and now my AI agent keeps the record for me: it finds what each client already tested and logs every result. I looked for this tool for a long time.
Gerardo ForlianoDigital Productivity Coach

Pricing

Start free. Pay when the team joins.

You pay for the people who edit the record. People who vote or watch are free on every plan.

People

1

Projects

1

Count the people who edit experiments, yourself included. Anyone who only votes or watches is free.

Your plan

Free

$0

forever

Free covers one project, with no time limit. Add a second project and Solo is $19.90/month.

Start for free

No card needed to sign up.

  • Free$0Just you, one project
  • Solo$19.90/moJust you, unlimited projects
  • Team$39/moUp to 3 members, then $10 for each extra member. Unlimited projects, private projects.

Before you start

Including the things it does not do.

The full stack a growth hacker works with, built for your AI agent. A framework with the method, an MCP server with 21 tools to read and update the record, three skills that teach the agent the method, a plugin that installs it all in Claude Code and Codex, and the app your team works in. Trackers are built for people clicking through a web app. Hypotize is built for the agent doing the work, under rules it cannot break.

The server. An experiment moves one stage at a time, and it cannot close without a measured result and a written learning. If the agent tries to skip that, the request is refused and the experiment stays open, so the agent has to go and get the result first.

Claude, ChatGPT, Codex, Cursor, and any client that connects to a remote MCP server with OAuth. In Claude Code and Codex you can also install the plugin, which adds the three skills to the connection.

No. Your existing tools run and measure the test. Hypotize is the system around it: what to test next, the stage each experiment is in, and every outcome with its evidence and learning. If your agent is also connected to your analytics, it reads the results there and records them in Hypotize.

A document accepts whatever anyone writes into it, agent included. Hypotize gives every experiment the same shape: hypothesis, priority score, owner, stage, outcome, evidence and learning. The server enforces it, so an agent cannot skip a stage or close a test without a result. A spreadsheet has no way to say no.

No, but that is where it pays off. The web app works on its own, with the same board, backlog and learnings. With an agent connected, the routine work moves to the agent: it finds what the team already tested, adds and scores ideas, moves tests through their stages and records outcomes. The app is where your team reviews that work.

It votes in whatever framework the experiment uses: ICE, PIE, RICE, or a custom one your team defined. Scores go through the same validation as the app, and votes aggregate into the priority with the framework's own formula. The agent also reads scores, adds ideas, moves statuses, logs interim readings as comments, and records outcomes.

Hypotize never crawls or syncs those tools. Your agent does the reading: if it can reach the source, the backlog import skill turns those notes into experiment records here. It skips duplicates and never invents a result the source does not contain.

No. The model runs inside the agent you connected, and Hypotize never sends your record anywhere for AI processing. The connection only lets that agent read and update experiment records, plus delete them if you grant that scope at connection, and nothing else.

Only what your own account and role allow. It cannot reach a project you cannot reach yourself, and every action goes through the same permission checks as the web app.

Yes, in one click. Access stops the moment you revoke the connection; there is no key that keeps working somewhere until it expires.

On Hypotize's own server at Hetzner in Finland, inside the EU. Nightly backups, and a restore is tested every week.

Workspace owners, admins and members, plus guests who are editors or owners on any project, count once per person. Pending invitations reserve members. Contributors and viewers are free on every plan. Free includes 1 member and 1 project; Solo includes 1 member and unlimited projects. Team is $39/month for up to 3 members, then $10 per extra member. AI agents and MCP connections never count.

Now your AI can run the experiment properly.

Start free with one project. Connect the agent you already use, or none at all.

Start for freeSee it in action
Works with
Kimi logoOpenClaw logoHermes logoand any MCP-compatible agent