Skip to content
Steptree

For enterprise knowledge programmes

Do the job once. Get the SOP your team follows and your AI can read.

Your AI rollout can only use what your organisation has written down, and most of how the work gets done never was.

Steptree records an expert doing the task on their own screen, instead of a week of interviews. Steptree then writes the SOP, with the goal, the steps and screenshots, and the judgement calls they flagged.

Their manager approves it before it is published. Staff follow it, and Claude or Steptree AI can search the same library.

Start freeSee an SOP

Free plan: 300 AI credits a month and two recordings, no card. Claude can connect on any plan.

How a process gets written

One expert does the job once, and the SOP is written from it

Nobody writes a specification first, and nobody sits through a week of interviews.

  1. An expert does the task, recording

    Every click is captured with a screenshot, in whichever app the work happens, on Windows or Mac. They state the goal first and flag each judgement call as they make it.

  2. Steptree writes the SOP as a draft

    The goal, the numbered steps with screenshots under them, the values that change each run, the rules that always apply and the decision points.

  3. Their manager approves it, then everyone reads it

    Staff open the approved SOP by link. Claude, Codex or GitHub Copilot can search the same library over MCP, and so can Steptree AI in the Steptree App.

Try it on one process

The document

The document a new starter works from

What one recording hands back: the goal, the numbered steps, the judgement calls and the things a newcomer would miss.

Your reviewers get something to mark up

A screen recorder captures what was clicked. A Steptree SOP also carries why: the goal the work was for, and the calls a click alone leaves out. Edit it like a doc, comment on it, then approve it. Any SOP can also become a SKILL.md skill file your AI loads.

The controls

What stays out of a recording, and who reads the result

Each one is enforced where the action happens, not described in a policy. See the FAQ below for what is not built yet.
Nothing is published until a person approves it
Every SOP is a draft until its owner, a team manager or an admin publishes it. An AI reading a draft over MCP is told it is a draft.
Password fields are never recorded
A password field is skipped entirely, screenshot included. Emails and card numbers are stripped from typed text.
Any recorded step can be deleted first
Screenshots are kept as taken, usually of just the window being worked in. A step that shows too much can be deleted before the SOP is written from it.
Your documentation stays in your workspace
Row-level security in Postgres separates every workspace, with a test suite over it, rather than a filter in application code.
AI access runs on a personal key that expires
Each person mints their own key for their AI. It sees only what they can see, and it is read-only unless they allow edits.
The audit log is append-only, and it exports
Administrative actions are written to a log your members cannot edit or delete, and it exports as JSON for your own security tooling.
Start on the free plan

Questions

What a review board asks, answered plainly

We already have a wiki. What does this add?
A wiki holds what somebody found time to type. Steptree writes the SOP from the task being done, with the screenshots and the reasons, and a manager approves it before anyone relies on it.
How does the AI we are rolling out read it?
Each person connects Claude, Codex, Cursor or GitHub Copilot in VS Code over MCP, the standard AI tools use to connect. It runs on their own key, which expires, and it comes with every plan.
Does the expert have to write anything?
No. They do the task at their normal pace, state the goal first, and press a button when they make a judgement call. Steptree writes the rest, and their manager corrects the draft.
Which computers does it record on?
Windows and Mac, in any application: the ERP, a spreadsheet, a vendor portal, the old system with no API. There is no Linux recorder.
What can our security team actually review?
Workspaces separated inside Postgres rather than by application code, with a test suite over it. Password fields skipped as a recording happens. An append-only audit log with an export route, multi-factor authentication, and dependency scanning that blocks a build. Send the questionnaire and a person answers it.
What will you not claim?
A SOC 2 report, because none exists yet. Proven single sign-on, because SAML and OIDC are built but have never met a live identity provider. A data region, because there is no selector for one. Customer-managed keys, because the platform underneath offers no honest way to hold them. These are four real gaps to weigh before you commit.
Can we start inside one business unit?
Yes, and it is the way we would suggest starting. Pick a process the unit already measures, have one expert record it, and judge the SOP against how the work is done today.

Next

Where to go from here

Write down one process your programme depends on

Have an expert record one real task, read the SOP it becomes, and connect Claude to the library. Start on the free plan today, no card.

Start freeCompare plans
Someone at a wall of sticky notes talking a seated team through a process.