For agencies
Ship for a client. Sabí writes it up.
In the client's own Notion, ready for you to check. Confluence is coming soon.
Your week, before and after
Before
The same change, written up two or three times.
Your Confluence says one thing. The client's Notion says another.
AI speeds up the code. Nobody has time for the docs.
One developer leaves, and a client's know-how leaves too.
After
Each change written up once, from the code.
One for their customers, one for their team, one for their developers.
Nothing reaches a client until you've checked it.
See what changed, and what Sabí left alone.
One workspace per client.
Each client's code, docs and reviewers stay apart.
Their template, not yours.
Coming in January: one template per client, so pages look like the rest of their docs.
Every client, their own template.
Your code visits. Your docs stay.
Each run analyses the client's code in a secured sandbox that's deleted when the run ends.
You stay in control.
You check everything first. Or let some pages publish on their own.
Every page shows the code change it came from.
Change a line yourself, and it goes through the same check.
Who approves what, per client
A typical week
- Thursday
You ship a new checkout for a client.
- Friday, 9:00
The write-up is waiting: one for their customers, one for their team.
- Friday, 11:30
Your project manager approves it.
Nobody wrote it twice.
Make the case to your co-founders
Every client keeps docs somewhere different, and your team keeps them in sync by hand.
A lead developer to connect each client's GitHub, and a reviewer for about an hour a week.
Hours a month spent on docs, per client.
Questions
Who buys Sabí, us or our clients?
You do. Your agency gets one workspace per client. Your clients just get docs that are current.
Our clients use Notion, Confluence, or both. Does that work?
Notion works today. Confluence is coming soon, so one change can land in both.
Will our clients let Sabí read their code?
Sabí reads it and never changes it. Each run uses a secured sandbox, deleted when the run ends, and Sabí doesn't train AI models on their code. The details are on the Security page.
Can Sabí handle a complex setup across many repos?
We're proving it on simpler setups first. Try it on one client's GitHub in the demo and judge for yourself.
