NewRe-run diffs: see exactly which steps changed after a release

Help center automation

The help center that keeps up with your product.

Trailframe’s agent works through your app in a real browser, drafts step‑by‑step articles with evidence for every step, and flags anything it couldn’t verify.

Free to start · No interns were harmed in the writing of these docs

A Trailframe run: 12 articles written, 46 screens explored, with a list of drafted help articles and why the run stopped.

Why guides matter

Nobody wants to open a ticket. Not even your support team.

A good help center answers questions before they become tickets, gets new users to value faster and gives your support team its time back.

Fewer tickets

Most “how do I…” questions have one right answer. Put it where customers look first.

Faster onboarding

New users find the next step on their own instead of stalling at setup.

Found by search and AI

Public guides show up in search, and they’re what AI assistants read to answer questions about your product.

Trust before they buy

Buyers check your docs before they sign. Current guides show a product that’s cared for.

Everyone knows guides matter. The hard part is keeping them accurate. That’s what Trailframe does.

Why Trailframe

Docs fall behind the moment your UI changes.

Writing help articles by hand is slow, and generic AI writes steps that don’t exist. Trailframe documents what it actually did in your app.

Screenshots go stale

A renamed button quietly breaks half your articles.

With Trailframe

Re-run after each release. Only the steps that changed are rewritten.

Nobody owns the docs

Support writes them late at night and the backlog keeps growing.

With Trailframe

One run covers a whole app. Your team reviews instead of writing.

AI that guesses

Chatbots invent menu names for flows they never clicked.

With Trailframe

Every step cites the screen it happened on. Gaps are flagged.

How it works

From a URL to a reviewed help center, in one run.

No scripts to write and no recorder to babysit. Give Trailframe a login and it does the clicking.

A run in progress: queued, signed in and exploring 38 screens, with time, action and spend limits.

Evidence, not guesswork

Pics or it didn’t happen.

Trailframe only writes what it saw. Reviewers check each step against a recorded action and a masked screenshot, in seconds.

  • Masked screenshots with the target control highlighted
  • If it didn’t click it, it won’t pretend it did
  • Versioned: compare runs and see what changed
The article “Connect a Slack channel” with six steps. Each step shows where it was seen and an evidence ID.

Safe by default

Comes with seatbelts.

It clicks around your app like a curious new hire, minus the part where it deletes the workspace.

policy.log · run_5

  1. allowclick"Integrations"
  2. allowtypesearch "slack"
  3. blockclick"Delete workspace"destructive control
  4. blockgotoaccounts.google.comoutside allowed sites
  5. blockclick"Sign out"would end the session
  6. allowclick"Connect"

Never submits, deletes or signs out

Destructive controls, outside sites and uploads are blocked. Allow submissions on a test workspace when you need them.

Credentials stay yours

Logins resolve in the worker from a reference. The model never sees them.

Masked screenshots

Page text is minimized and sensitive fields are masked before anything is stored.

Versioned articles

Re-run after a release and compare versions step by step.

Gemini via Vertex AI or an API key. Switch per run.

Model optional

Gemini on Vertex AI or an API key, or a deterministic rule set with no model.

Publish

A help center your customers will actually use.

Approved articles publish to a fast, searchable help center on your own domain. It updates itself after every run.

  • Hosted on your domain, like help.yourproduct.com
  • Search, categories and answers with citations
  • Or export Markdown and JSON to any docs tool
Browse our help center, written by Trailframe →

One example run, on a demo app

screens explored
46
articles drafted
12
start to finish
14 min
model spend
$0.62

Pricing

Start free. Scale by guides, not seats.

Every plan includes safe mode, evidence for every step and Markdown and JSON exports.

Free

For trying Trailframe on one app

Free

Start free
  • 1 workspace
  • Safe mode and policy log
  • Export Markdown and JSON
  • Community support

Team

For larger products and many apps

10× the guides

Talk to sales
  • Multiple workspaces and roles
  • Self-host on Google Cloud
  • Custom models and limits
  • Priority support

FAQ

Questions you were too polite to ask.

Can’t find what you need? Our help center has the details, or talk to us.

Contact us →
Will the agent change anything in my app?

No. Safe mode never submits forms, deletes anything, signs out or uploads files, and every blocked action is logged. You can allow submissions on a test workspace.

Does the model see our passwords?

No. You give Trailframe a credential reference, and the browser worker resolves it at sign-in. Page text is minimized before anything is sent to a model, and screenshots are masked.

What happens when a step can’t be verified?

The article says so. Steps the agent didn’t perform are flagged with a review note instead of being written as if they happened, so a reviewer can confirm them or re-run on a test workspace.

Can we host it ourselves?

Yes. Run the CLI and worker locally, use the Docker image, or deploy the hosted service to your own Google Cloud project with the included Pulumi program.

Which models does it use?

Gemini, through Vertex AI or an API key. You can also run with a deterministic rule set and no model at all.

How do articles stay up to date?

Re-run after a release. Each run creates a new version of an article, so you can compare versions and see which steps changed before you publish.

Your help center called. It wants to be useful again.

Point Trailframe at your app and review the first drafts this afternoon.

Free to start · Safe mode on by default